[
  {
    "id": "2026-11-optimal-scenario-design-climate-emulation",
    "title": "Unrealistic emissions scenarios help AI climate models prepare for unfamiliar futures",
    "summary": "MIT researchers with Imperial College London designed a small synthetic set of deliberately unrealistic emissions scenarios to train climate emulators, which normally need six to eight full Earth system model runs that each take months. The method needs only one model run, and the emulator predicted unseen scenarios and local pollutant-control effects where traditional emulators failed. It was validated with a simple climate model rather than a full Earth system model, and the article reports no quantitative accuracy metrics.",
    "field": "climate",
    "ai_technology": "Climate emulator trained on optimized synthetic scenarios",
    "organization": "MIT; Imperial College London",
    "source_url": "https://news.mit.edu/2026/new-research-could-lead-faster-trustworthy-ai-models-climate-prediction-1112",
    "source_type": "press release",
    "publication_date": "2026-11-12",
    "added_date": "2026-11-13",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An emulator trained on a small set of synthetic scenarios predicted unseen scenarios and a local pollutant-control effect where traditional emulators failed. The source says it was validated with a simple climate model.",
      "unknown": "The source says the method still needs testing on a full Earth system model scenario and on other climate model types; it does not say whether this has been done.",
      "ai_role": "An optimization algorithm generated the synthetic training scenarios; an AI emulator was trained on them to predict climate outcomes.",
      "evidence_urls": [
        "https://news.mit.edu/2026/new-research-could-lead-faster-trustworthy-ai-models-climate-prediction-1112"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-ipertensione-smartwatch",
    "title": "Signs of hypertension detected by a smartwatch",
    "summary": "Prospective study of 196 people at five US sites, compared with 24-hour ambulatory blood pressure measurement: 65.8% sensitivity and 90.0% specificity for screening, and 78.4% and 90.9% in the confirmatory test (48 participants). A feasibility study; it does not replace the cuff, and many cases are missed.",
    "field": "medicine",
    "ai_technology": "AI-PPG-ACC-HTN (photoplethysmography and accelerometer)",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/detecting-signs-of-hypertension-using-wearable-sensors/",
    "source_type": "blog",
    "publication_date": "2026-11-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-11336-3",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A prospective study of 196 participants at five US sites compared smartwatch screening with 24-hour ambulatory blood pressure monitoring: 65.8% sensitivity and 90.0% specificity, with 78.4% and 90.9% in a confirmatory subgroup of 48.",
      "unknown": "The authors call it early feasibility research; the smartwatch does not replace a cuff, many true cases are missed, the confirmatory subgroup is small, and generalizability is unclear.",
      "ai_role": "A 1-D convolutional network pretrained with self-supervised contrastive learning on PPG and accelerometer waveforms produced embeddings; a logistic regression output a hypertension probability.",
      "evidence_urls": [
        "https://research.google/blog/detecting-signs-of-hypertension-using-wearable-sensors/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-11-mitochondrial-interactome-alphafold",
    "title": "AlphaFold-Multimer predicts 2,895 interactions among human mitochondrial proteins",
    "summary": "Texas A&M AgriLife Research screened 630,003 protein pairs from about 1,100 human mitochondrial proteins with AlphaFold-Multimer and identified 2,895 likely interactions, more than half not previously reported, giving 85 uncharacterized proteins their first probable partners. The paper appeared in Nature Communications on 25 August 2026 and the institutional release on 5 November. Lab experiments in yeast and human cells confirmed several of the partnerships; most predictions remain untested, and the disease links are potential targets, not established ones. Predictions are shared through the MitoMatch web tool.",
    "field": "biology",
    "ai_technology": "AlphaFold-Multimer",
    "organization": "Texas A&M AgriLife Research; and Washington University School of Medicine in St. Louis",
    "source_url": "https://www.eurekalert.org/news-releases/1150930",
    "source_type": "press release",
    "publication_date": "2026-11-05",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-77112-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "AlphaFold-Multimer screened about 630,000 protein pairs; lab experiments in yeast and human cells confirmed several of the predicted partnerships, while the release does not say how many of the 2,895 were tested.",
      "unknown": "Only several predictions were confirmed, most remain unverified, the interactions are described as likely rather than established, and disease links are potential targets.",
      "ai_role": "AlphaFold-Multimer predicted pairwise interactions among mitochondrial proteins; the predictions were released as the MitoMatch web resource.",
      "evidence_urls": [
        "https://www.eurekalert.org/news-releases/1150930"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-vault-claude",
    "title": "First 3D structure of a prokaryotic vault particle, with Claude",
    "summary": "Claude searched protein databases, designed DNA constructs, reduced purification from about three days to about 90 minutes, and processed cryo-EM data into an initial model in 34 hours. The result is a particle made of 96 copies of a bacteriophage protein, with about 25% more internal volume than the mammalian version. Its biological function is unknown; this is a preprint, and Claude made errors that humans corrected.",
    "field": "biology",
    "ai_technology": "Claude",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/research/investigating-vault-particles-with-claude",
    "source_type": "blog",
    "publication_date": "2026-11-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.11.04.764334",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The 3D structure was determined by cryo-EM; cryo-EM confirmed the vault shape for four of seven particle-forming proteins, and fifteen altered protein versions were tested, with a reviewer checking the structure.",
      "unknown": "The biological function is unknown, vaults were made only in E. coli, and an unidentified pocket molecule resembling coenzyme A is unconfirmed.",
      "ai_role": "Claude searched protein databases, designed DNA constructs, optimized purification, and ran RELION cryo-EM processing to an initial model in 34 hours; it also made errors that humans corrected.",
      "evidence_urls": [
        "https://www.anthropic.com/research/investigating-vault-particles-with-claude"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-agentic-cheminformatics",
    "title": "Agentic cheminformatics: new macrocycles from literature analysis",
    "summary": "An agent using Cheminformatics Skills for Antigravity collected thousands of compounds analogous to crizotinib and lorlatinib, analyzed 568, and proposed new rigidified macrocycles with synthesis routes in about an hour. The proposals are computational hypotheses that still need to be synthesized and validated.",
    "field": "chemistry",
    "ai_technology": "Google Antigravity, Cheminformatics Skills",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/agentic-cheminformatics-with-google-antigravity/",
    "source_type": "blog",
    "publication_date": "2026-11-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The source describes computational evaluation only: benchmarks of several machine learning architectures on four readouts with scaffold splits, and predicted potency; no synthesis or experimental validation is reported.",
      "unknown": "Every proposed design must still be synthesized and validated; the data are sparse and assay noise is high; the designs are testable hypotheses, not finished drugs.",
      "ai_role": "An agent searched literature for crizotinib and lorlatinib analogues, curated 568 compounds, ran matched molecular pair analysis, benchmarked models, and proposed macrocycles and synthesis routes.",
      "evidence_urls": [
        "https://deepmind.google/blog/agentic-cheminformatics-with-google-antigravity/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-herculaneum-pherc-306b-first-letters",
    "title": "Writing seen in PHerc. 306B for the first time in 2000 years",
    "summary": "Vesuvius Challenge awarded a $50K First Letters Prize for scroll PHerc. 306B, where the winning submission reached the 10-character target and the papyrology team agreed on a reading, tentatively a Greek phrase with parallels in Aristotle and Philodemus. Ink was detected by multiplying two models (TRACE v3B and a fine-tuned A-hn). Letters remain faint, the readable area is small, and the work relied on hundreds of hours of manual labeling.",
    "field": "humanities",
    "ai_technology": "ML ink-detection models (TRACE v3B, A-hn) with VC3D GrowPatch",
    "organization": "Vesuvius Challenge",
    "source_url": "https://scrollprize.substack.com/p/writing-seen-in-pherc-306b-for-the",
    "source_type": "blog",
    "publication_date": "2026-11-02",
    "added_date": "2026-11-13",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 4,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The reading was checked by the papyrology team, which agreed on the letters with the winning submission, and was compared against 25 community submissions and a 10-character target; the page describes no independent replication.",
      "unknown": "The reading is partial and uncertain, letters remain faint, training on machine-generated labels could compound errors (not assessed), and the Grand Prize is not yet awarded.",
      "ai_role": "Machine-learning ink-detection models (TRACE v3B and a fine-tuned A-hn) produced the ink maps; the readings and prize review were done by humans.",
      "evidence_urls": [
        "https://scrollprize.substack.com/p/writing-seen-in-pherc-306b-for-the"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-10-evo2-in-context-enhancers",
    "title": "Evo 2 uses in-context learning to design de novo regulatory sequences with only tens of examples",
    "summary": "Arc Institute reports that Evo 2 designs functional regulatory sequences from 4 to 64 in-context examples without extra training, with 30-50% yield. Of nine AI-designed enhancers delivered by AAV to mice, seven drove detectable expression and two labeled layer 5 extratelencephalic neurons with 73% and 95% specificity. Results are bioRxiv preprints; in vivo work covers only layer 5 enhancers, and other designs (proteins, CRISPR loci, antibodies, phage genomes) are computational only.",
    "field": "biology",
    "ai_technology": "Evo 2 (genomic language model); in-context learning; reinforcement learning post-training",
    "organization": "Arc Institute; Stanford University",
    "source_url": "https://arcinstitute.org/news/evo-2-in-context-learning",
    "source_type": "press release",
    "publication_date": "2026-10-29",
    "added_date": "2026-11-13",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Nine AI-designed enhancers were delivered by AAV to mice: seven drove detectable expression and two labeled layer 5 extratelencephalic neurons with 73% and 95% specificity.",
      "unknown": "Only layer 5 enhancers were tested in vivo; the authors plan other cell types, and designs beyond enhancers (proteins, CRISPR loci, antibodies, phage genomes) are computational only.",
      "ai_role": "Evo 2 generated regulatory sequences from 4 to 64 in-context examples; reinforcement learning with predictive reward models sharpened cell-type specificity.",
      "evidence_urls": [
        "https://arcinstitute.org/news/evo-2-in-context-learning"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-sviluppo-farmaci",
    "title": "Claude's early progress in drug development",
    "summary": "Anthropic reports early results with six partners: a CAR T-cell receptor design about 70% shorter that enabled immune cells to produce about 10 times more receptors; about 35,000 antibody designs in one month, yielding 20 candidates; and a document review that found all 11 human-identified issues plus seven more in 89% less time. The work is mostly preclinical and reported by Anthropic.",
    "field": "medicine",
    "ai_technology": "Claude",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/research/early-progress-drug-development",
    "source_type": "blog",
    "publication_date": "2026-10-27",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Partner lab tests and a document review reported by Anthropic: a CAR T receptor design was tested in immune cells, antibody designs were screened for IgE binding, and a review of 1,061 pages was compared with 11 human-identified issues.",
      "unknown": "The work is mostly preclinical with no human tests; partner reports give limited methodological detail, and the page does not say whether the seven extra flags were confirmed.",
      "ai_role": "Claude proposed molecular and receptor designs, generated about 35,000 antibody designs, reviewed manufacturing documents, and drafted regulatory documents.",
      "evidence_urls": [
        "https://www.anthropic.com/research/early-progress-drug-development"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-openai-tre-risultati",
    "title": "OpenAI presents three results in harmonic analysis, complexity, and quantum theory",
    "summary": "OpenAI announces results on local smoothing, a disproof of the exponential time hypothesis, and the construction of quantum field theories. The source presents these as research results to be understood with academic groups: physical artifacts are not yet public, and independent verification is not established here.",
    "field": "multidisciplinary",
    "ai_technology": "OpenAI internal models",
    "organization": "OpenAI",
    "source_url": "https://openai.com/index/building-scientific-understanding-with-academic-communities/",
    "source_type": "press release",
    "publication_date": "2026-10-26",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 4,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The local-smoothing proof was formalized in Lean the day after the AI output; the refutation of the Exponential Time Hypothesis was verified in Lean and is being checked by an external group led by Impagliazzo, Paturi and Williams.",
      "unknown": "The physics result was not released; the external groups had not reported conclusions on the page.",
      "ai_role": "OpenAI internal models produced the proofs; external experts were asked to shepherd and explain the results.",
      "evidence_urls": [
        "https://openai.com/index/building-scientific-understanding-with-academic-communities/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-10-tb-screening",
    "title": "Chest X-rays and AI for tuberculosis screening",
    "summary": "Nexus AI says it has deployed Google Research's chest X-ray AI as a certified diagnostic tool at about 40 facilities in six countries: more than 25,000 X-rays assessed by over 90 clinicians, with response times under 45 seconds. These are author-reported deployment figures, with no sensitivity or patient outcome data; WHO and FDA approvals are still in progress.",
    "field": "medicine",
    "ai_technology": "Foundation models for chest X-rays",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/bringing-foundational-models-to-the-front-lines-of-tb-screening/",
    "source_type": "blog",
    "publication_date": "2026-10-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 1,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The post reports deployment figures (about 40 facilities, over 25,000 X-rays, over 90 clinicians) without sensitivity, specificity or patient outcome data from this deployment; it cites earlier peer-reviewed studies for performance.",
      "unknown": "Deployment-level accuracy and outcomes are not given, WHO prequalification and FDA clearance are not granted, and the facility count across countries is unclear.",
      "ai_role": "The AI analyzed chest X-rays and flagged TB-suggestive findings for triage in under 45 seconds; clinicians made final decisions.",
      "evidence_urls": [
        "https://research.google/blog/bringing-foundational-models-to-the-front-lines-of-tb-screening/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2026-10-forty-forest-map",
    "title": "ForTy 2020: global map of forest types at 10 meters",
    "summary": "A 10-meter map classifies each land area into six forest or land-use types (including primary forest, plantations, and tree crops), with an initial time series from 2017–2025. Validation across more than 9,000 sites identifies the correct type in 90% of cases and distinguishes forest from non-forest in 95%.",
    "field": "climate",
    "ai_technology": "Satellite-image classification models",
    "organization": "Google DeepMind; Google Research, World Resources Institute, IIASA",
    "source_url": "https://deepmind.google/blog/beyond-tree-cover-mapping-the-worlds-forest-types/",
    "source_type": "blog",
    "publication_date": "2026-10-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41597-026-08299-5",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The map was validated at more than 9,000 random sites, where two trained experts checked high-resolution imagery to determine land type: correct forest type in 90% of cases and forest vs non-forest in 95%.",
      "unknown": "Per-class accuracy and regional breakdowns are not reported, the time series is early-release without separate validation, and results depend on the probability threshold chosen.",
      "ai_role": "A custom remote-sensing model trained on about 1.7 million labeled locations, using AlphaEarth embeddings, outputs probabilities for six forest and land-use classes.",
      "evidence_urls": [
        "https://deepmind.google/blog/beyond-tree-cover-mapping-the-worlds-forest-types/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-ecografia-materna",
    "title": "AI-assisted ultrasound for dating pregnancy",
    "summary": "Prospective study of 385 pregnant women in Chicago and Nairobi: the model estimates gestational age from scans taken by minimally trained operators with a mean error of 4.18 days, non-inferior to the 4.48 days of expert sonographers. This is a non-inferiority result from only two sites, both relatively urban.",
    "field": "medicine",
    "ai_technology": "Ultrasound model for estimating gestational age",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/empowering-frontline-maternal-care-with-ai-assisted-ultrasound/",
    "source_type": "blog",
    "publication_date": "2026-10-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1001/jamanetworkopen.2026.22484",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A prospective multi-center study of 385 pregnant women at two sites (Chicago and Nairobi) compared AI gestational-age estimates from blind sweep scans by minimally trained operators with expert sonographer biometry: mean error 4.18 vs 4.48 days.",
      "unknown": "The source does not state the non-inferiority margin or confidence intervals; only two relatively urban sites and one commercial probe were used, and clinical trust and confidence safeguards are not validated.",
      "ai_role": "The model estimated gestational age from blind sweep ultrasound scans and ran live on mobile devices in clinics.",
      "evidence_urls": [
        "https://research.google/blog/empowering-frontline-maternal-care-with-ai-assisted-ultrasound/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-10-rosalind-atlante-crispr",
    "title": "GPT-Rosalind agents connect genetic perturbations and neuronal signals",
    "summary": "Analysis of a CRISPR atlas of mouse brain cells identified differences between Tsc1 and Tsc2 and a possible relationship between DCTN1 and cholesterol regulation. Experts assessed the interpretations, but the proposed mechanisms remain hypotheses to be tested in new experiments.",
    "field": "biology",
    "ai_technology": "GPT-Rosalind and distributed agents",
    "organization": "OpenAI; PerturbAI and Scripps Research",
    "source_url": "https://openai.com/index/scaling-ai-research-for-complex-biological-data/",
    "source_type": "press release",
    "publication_date": "2026-10-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.10.15.759436",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Thousands of GPT-Rosalind agents reanalyzed an existing whole-brain CRISPR atlas; they reproduced findings from human analyses, and their novelty and value assessments aligned with blinded human experts.",
      "unknown": "The proposed biological mechanisms still need experiments to test them.",
      "ai_role": "Each agent investigated one gene disruption, searched the literature and proposed explanations; reports were combined into a reasoning atlas.",
      "evidence_urls": [
        "https://openai.com/index/scaling-ai-research-for-complex-biological-data/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-leganti-scorpione",
    "title": "Claude designs binders against five scorpion neurotoxins",
    "summary": "In a mostly autonomous 64-hour campaign, Claude designed minibinders against five neurotoxins. External labs tested 211 designs, 17 of which met the criteria; the best (Cn2-B1 and Cn2-B2) had affinities in the low nanomolar range. Only recombinant toxins were tested, with no neutralization tests in cells or animals; Anthropic does not claim to have produced an antivenom.",
    "field": "medicine",
    "ai_technology": "Claude",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/research/claude-scorpion-toxin-binder-design",
    "source_type": "blog",
    "publication_date": "2026-10-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "External labs tested 211 designs by surface plasmon resonance: Adaptyv Bio in cell-free expression and Twist Bioscience in human kidney cells; a binder needed KD of 1 µM or tighter in every replicate at both labs and no CLEC12A signal, giving 17 binders. The two collaborating external laboratories tested candidates from the discovery project; this establishes direct testing rather than a separately established independent follow-up replication.",
      "unknown": "Only recombinant toxins were tested, with no neutralization in cells or animals; the authors do not know whether the binders neutralize venom and do not claim an antivenom; multitoxin designs failed.",
      "ai_role": "Claude selected toxin structures and hotspots, generated 318,484 designs, filtered and co-folded them, ran about 24 hours autonomously, then produced 180 selected designs after human follow-ups.",
      "evidence_urls": [
        "https://www.anthropic.com/research/claude-scorpion-toxin-binder-design"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-autonomous-wet-labs",
    "title": "AI agents in the lab: aptamers and proteins designed in a closed loop",
    "summary": "AI agents designed DNA aptamers and protein binders, which were then tested in the lab over multiple rounds. Target success rates reached 33.5% for myoglobin and 42.1% for α-thrombin (98.5% specificity). The post summarizes three arXiv papers whose peer review is unconfirmed; it also reports initial failures.",
    "field": "biology",
    "ai_technology": "AI agents with laboratory robots",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/",
    "source_type": "blog",
    "publication_date": "2026-10-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Agents ran three design-test-iterate rounds of DNA aptamers against myoglobin and alpha-thrombin in the lab, reaching 33.5% and 42.1% hit rates with 98.5% specificity for thrombin; protein binders against FAP were reported at single-digit nanomolar affinity.",
      "unknown": "Peer review of the cited arXiv papers is unconfirmed; there are two protein targets and three rounds; initial rounds failed, and protein expression adds steps to each loop.",
      "ai_role": "Agents designed and revised sequences across rounds and adjusted design rules after reading wet-lab data, under human oversight of protocols and robot runs.",
      "evidence_urls": [
        "https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-circuiti-quantistici",
    "title": "Claude designs more efficient quantum circuits",
    "summary": "Claude designed circuits for eight basic operations (chemistry, optimization, physical simulation, and number theory), assessed by an automated verifier. In several cases they outperform the best published circuits; for example, a point addition used 17 times fewer gates at the cost of 12% more qubits. Key identities were checked with simulation and Lean; a memory shortcut in the decoder fails about once in 10,000 times.",
    "field": "physics",
    "ai_technology": "Claude",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/research/designing-efficient-quantum-circuits",
    "source_type": "blog",
    "publication_date": "2026-10-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2610.22367",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "An automated evaluator checked each circuit for correctness and scored qubits times gates; results were compared with published designs, divsteps was checked by simulation and Lean, and decoder memory was measured over 403 million simulated decodings.",
      "unknown": "The decoder memory shortcut fails about once in 10,000 runs; gains apply to individual primitives, not a full fault-tolerant machine; Claude was conservative and needed prompts to broaden its search.",
      "ai_role": "Claude wrote circuits as Python code and revised them in a closed loop with evaluator feedback, with no human input between rounds, and reconstructed branch information to reduce memory.",
      "evidence_urls": [
        "https://www.anthropic.com/research/designing-efficient-quantum-circuits"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-antimicrobial-peptides-antigravity",
    "title": "Searching for antimicrobial peptides with Google Antigravity",
    "summary": "An agentic workflow searched UniProt for antimicrobial candidates in Streptococcus pseudopneumoniae and found a 97-amino-acid bacteriocin with no homologues among manually reviewed Swiss-Prot proteins. This is a computational candidate now undergoing laboratory validation; no potency or toxicity data are available.",
    "field": "biology",
    "ai_technology": "Google Antigravity (agentic platform)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/accelerating-antimicrobial-discovery/",
    "source_type": "blog",
    "publication_date": "2026-10-15",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 1,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Nothing was tested in the lab: the candidate's novelty was checked by BLAST against Swiss-Prot, its predicted structure was compared with a known bacteriocin fold, and its genomic context was examined; the wet-lab plan is underway with no results.",
      "unknown": "No potency or toxicity results exist; the wet-lab assays outcomes are unknown; the session covered a narrow sequence window in one species.",
      "ai_role": "An agent queried databases, filtered candidates, ran BLAST and structure comparisons, analyzed genomic neighborhoods, synthesized literature and drafted a wet-lab plan under researcher direction.",
      "evidence_urls": [
        "https://deepmind.google/blog/accelerating-antimicrobial-discovery/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2026-10-earth-ai-sanita",
    "title": "Earth AI for public health: geospatial models across five use cases",
    "summary": "The PDFM model increased the explained variance in MMR vaccination coverage from 16% to 22% across 146 counties on the US–Canada border, and improved weekly correct cholera-risk-area rankings in the DRC from 1.78 to 2.10. Gains for cardiovascular mortality and postpartum depression were small or not significant.",
    "field": "medicine",
    "ai_technology": "Population Dynamics Foundation Model (PDFM), TimesFM",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/earth-ais-planetary-geospatial-foundation-models-for-global-public-health/",
    "source_type": "blog",
    "publication_date": "2026-10-06",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2610.05699",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Five partner-led evaluations compared PDFM embeddings with baselines: MMR explained variance rose from 16% to 22% across 146 border counties and cholera Precision@5 improved from 1.78 to 2.10; cardiovascular gains were not significant.",
      "unknown": "Several results were non-significant or small, cholera gains were not significant at one or two weeks, and the authors cite static embeddings as a limitation.",
      "ai_role": "PDFM combines search trends, mobility, built-environment and environmental data into location embeddings used without task fine-tuning; for dengue it was paired with TimesFM 2.0.",
      "evidence_urls": [
        "https://research.google/blog/earth-ais-planetary-geospatial-foundation-models-for-global-public-health/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-synthid-bio",
    "title": "SynthID Bio: watermarking for proteins and synthetic structures",
    "summary": "Methods insert a detectable signal into protein sequences and predicted structures. For binders designed against three targets (VEGF-A, SARS-CoV-2 RBD, and PD-L1), watermarked designs matched the others in success rate and affinity. This is a proof of concept that needs to be made more robust against tampering.",
    "field": "biology",
    "ai_technology": "SynthID Bio, AlphaFold 3 (optimized component)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/introducing-synthid-bio/",
    "source_type": "blog",
    "publication_date": "2026-09-30",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-10965-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Wet-lab tests on three targets (VEGF-A, SARS-CoV-2 RBD, PD-L1), with Adaptyv Bio helping, showed watermarked designs matched unwatermarked ones in hit rate, affinity and diversity; the phage was tested only in bacterial cultures.",
      "unknown": "The watermark must become more robust against deliberate tampering, the bacteriophage work is early-stage, and the post says no single biosecurity measure is a silver bullet.",
      "ai_role": "SynthID Bio guided amino acid choice in a ProteinMPNN variant, adjusted predicted atomic coordinates, fine-tuned part of AlphaFold 3, and was integrated into Evo 2 to watermark a designed phage genome.",
      "evidence_urls": [
        "https://deepmind.google/blog/introducing-synthid-bio/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-lila-catalizzatori-palladio",
    "title": "Catalysts without iridium or ruthenium for acidic oxygen evolution",
    "summary": "An AI-guided, human-supervised platform assessed 2,942 catalysts and identified complex palladium oxides. InMnPdOx maintained an overpotential below 0.5 V for more than 1,000 hours in sulfuric acid at 10 mA/cm². This is a preprint and laboratory result, without validation in a PEM electrolyzer.",
    "field": "chemistry",
    "ai_technology": "Sequential learning, composition–property models, and LLMs",
    "organization": "Lila Sciences",
    "source_url": "https://arxiv.org/abs/2609.30133",
    "source_type": "paper",
    "publication_date": "2026-09-24",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Lead catalysts were tested in 1 M H2SO4 at 10 mA/cm2: InMnPdOx kept overpotential below 0.5 V for over 1,000 hours; a retrospective benchmark compared the agent with Bayesian optimization and in-context language-model selection.",
      "unknown": "Testing in a full PEM electrolyzer is not reported, and 'stable' means overpotential under 0.5 V rather than a broader durability assessment.",
      "ai_role": "A sequential-learning agent with ML composition models, adaptive optimization and LLM reasoning ran a human-supervised closed loop with over 90% automation.",
      "evidence_urls": [
        "https://arxiv.org/abs/2609.30133"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-claude-enzima-art",
    "title": "Claude identifies a new enzyme system: array-associated reverse transcriptases",
    "summary": "About 950 Claude agents searched a DNA-sequence database for 21 hours (210 million tokens) and flagged an uncharacterized enzyme system, found mostly in bacteriophages, with a DNA repeat array similar to CRISPR. Scientists performed the laboratory work; the function remains unknown, and the work is a preprint, not a peer-reviewed publication.",
    "field": "biology",
    "ai_technology": "Claude (agents)",
    "organization": "Anthropic",
    "source_url": "https://anthropic.com/news/claude-discovers-novel-enzyme-system",
    "source_type": "blog",
    "publication_date": "2026-09-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.10.05.755809",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Human scientists expressed the protein in laboratory strains and characterized it biochemically and structurally; early experiments showed the array is expressed as distinct short RNAs.",
      "unknown": "The primary function of the enzyme system is unknown; findings are preliminary, and most Claude-proposed candidates are eliminated in follow-up review.",
      "ai_role": "About 950 agents used 210 million tokens over 21 hours to search DNA sequences, narrowing about 3,500 candidate systems to 20; one agent noticed the repeat array and filed a report.",
      "evidence_urls": [
        "https://anthropic.com/news/claude-discovers-novel-enzyme-system"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-retrochimera",
    "title": "RetroChimera: predicting small-molecule synthesis",
    "summary": "Published in Nature, the method combines a Transformer model (R-SMILES 2) and a template-based graph model (NeuralLoc) through a learned vote. In blind tests with expert chemists, 90% of complete routes were accepted, compared with 20–50% for the component models or a baseline, and nine of ten difficult targets were solved. Evaluation covered only ten targets; one wrong prediction invalidates an entire route.",
    "field": "chemistry",
    "ai_technology": "RetroChimera (Transformer and template-based model)",
    "organization": "Microsoft Research",
    "source_url": "https://www.microsoft.com/en-us/research/blog/improving-synthesis-prediction-of-small-molecules-at-scale-with-retrochimera/",
    "source_type": "blog",
    "publication_date": "2026-09-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.nature.com/articles/s41586-026-11160-9",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Blind tests by PhD-level chemists rated individual steps and complete routes for ten difficult targets: 90% of RetroChimera routes were accepted, and nine of ten targets were solved, versus 20-50% for the component models or baseline.",
      "unknown": "The route-level evaluation covers only ten targets; one incorrect prediction invalidates a route; NeuralLoc is constrained on reactions outside its template library, and R-SMILES 2 may hallucinate.",
      "ai_role": "An ensemble of a Transformer (R-SMILES 2) predicting precursors and a graph neural network (NeuralLoc) selecting templates, combined by a learned rank-dependent vote.",
      "evidence_urls": [
        "https://www.microsoft.com/en-us/research/blog/improving-synthesis-prediction-of-small-molecules-at-scale-with-retrochimera/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-claude-ottimizza-modelli",
    "title": "Claude optimizes more than 30 open models for biology",
    "summary": "In just under four weeks, Claude optimized over 30 models for structure, protein design, and genomics. They were on average four times faster with a small accuracy loss, and nearly twice as fast with identical outputs. It also created a low-memory mode for systems over 10,000 tokens on a single GPU node. Design results are in silico only, and comparisons are internal to Anthropic.",
    "field": "computer science",
    "ai_technology": "Claude (in Claude Science)",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/research/claude-uplifts-biomolecular-modeling",
    "source_type": "blog",
    "publication_date": "2026-09-17",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Accelerated models were checked against their downstream tasks (for example structure prediction) and FlashPairformer kernels were compared with cuEquivariance (2.7-2.9x faster on triangle attention); design results are in silico with internal comparisons.",
      "unknown": "Design results have no wet-lab validation, large viral capsid predictions were incorrect, ColabFold kernels were not benchmarked, and the model is not reproducible.",
      "ai_role": "Claude wrote custom FlashPairformer kernels, optimized more than 30 models, built a low-memory Big mode, and ran autonomous protein binder design.",
      "evidence_urls": [
        "https://www.anthropic.com/research/claude-uplifts-biomolecular-modeling"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-alphagenome-atlas",
    "title": "AlphaGenome Atlas: predictive map of human DNA variation",
    "summary": "A precomputed database of predicted molecular effects for about 9 billion possible single-letter changes in the human genome, generated by AlphaGenome. Researchers can access it free through a portal and API. The predictions are a research tool, do not replace experiments, and are not validated for clinical use.",
    "field": "biology",
    "ai_technology": "AlphaGenome, AlphaGenome Variant Impact (AVI)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome/",
    "source_type": "blog",
    "publication_date": "2026-09-08",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Validation described is the company's own testing of the AlphaGenome Variant Impact score on pathogenicity benchmarks, plus case studies: a DNM1 variant prioritized with GREGoR and 22% more non-coding associations in UK Biobank.",
      "unknown": "The page does not detail how individual predictions were checked; AlphaGenome is not validated or approved for clinical use.",
      "ai_role": "AlphaGenome predicted variant effects across cell types and tissues; these were precomputed for about 9 billion variants and combined with AlphaMissense into a single score.",
      "evidence_urls": [
        "https://deepmind.google/blog/alphagenome-atlas-a-predictive-map-of-every-possible-dna-letter-change-in-the-human-genome/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-openai-navier-stokes",
    "title": "OpenAI proposes a solution to the Navier–Stokes problem",
    "summary": "OpenAI announces a proof that three-dimensional solutions to the Navier–Stokes equations can develop a singularity in finite time and publishes a text formalized in Lean. This is the source's claim: this record does not establish community consensus, correctness of the artifacts, or recognition as a Millennium Prize problem.",
    "field": "mathematics",
    "ai_technology": "OpenAI internal model",
    "organization": "OpenAI",
    "source_url": "https://openai.com/index/navier-stokes-solution/",
    "source_type": "press release",
    "publication_date": "2026-09-08",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 4,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "OpenAI shared an analytical proof and a Lean formalization that an initially smooth fluid at rest can develop a finite-time singularity under a smooth force with finite energy.",
      "unknown": "Not stated in the reviewed material",
      "ai_role": "An internal OpenAI model, described as more capable than GPT-6 Astra, produced the proof and the Lean formalization.",
      "evidence_urls": [
        "https://openai.com/index/navier-stokes-solution/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "source_type": "press release",
    "id": "2026-09-codex-calibrazione-qubit",
    "title": "Codex agents perform measurements and calibrations on a six-qubit chip",
    "summary": "Connected to MIT EQuS lab software, GPT-5.6 Sol selected parameters, ran measurements, and calibrated a six-qubit superconducting chip, measuring its frequencies and coherence. The demonstration covers known procedures on real hardware; weak or noisy signals sometimes required expert assistance, and no new physical property was demonstrated.",
    "field": "physics",
    "ai_technology": "GPT-5.6 Sol and Codex",
    "organization": "OpenAI; and MIT EQuS",
    "source_url": "https://openai.com/index/codex-quantum-computing-experiments/",
    "publication_date": "2026-09-08",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Connected to MIT EQuS lab software, GPT-5.6 Sol ran measurements on an uncalibrated six-qubit chip of a standard type, chose parameters, operated the hardware and analyzed results.",
      "unknown": "Not assessed: the technical case study with the detailed results was not read.",
      "ai_role": "GPT-5.6 Sol in Codex ran routine measurement workflows with measurement-specific skills provided by the researcher.",
      "evidence_urls": [
        "https://openai.com/index/codex-quantum-computing-experiments/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-fermat-lean",
    "title": "Claude formalizes Fermat's theorem in Lean",
    "summary": "A proof of Fermat's Last Theorem fully machine-verified in Lean: Claude produced most of it independently over 11 days, with about 13 million lines and 29,500 intermediate theorems, and a comparator confirms that its statement matches Mathlib's. It follows a simplified version of Wiles's proof, uses Mathlib and parts of an Imperial College project, and is probably longer than necessary.",
    "field": "mathematics",
    "ai_technology": "Claude, Lean",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/research/formalizing-fermats-last-theorem",
    "source_type": "blog",
    "publication_date": "2026-09-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The proof was checked by Lean using only standard axioms; a comparator tool confirmed the theorem statement matches Mathlib's, and mathematician Kevin Buzzard reviewed it.",
      "unknown": "The proof is about 13 million lines and likely much longer than needed; it follows a simplified version of Wiles's proof and is presented as verification, not new mathematics.",
      "ai_role": "Dozens of agents worked for 11 days on the Prove2Me platform, producing about 30,300 computer-verified theorems; humans gave occasional high-level instructions.",
      "evidence_urls": [
        "https://www.anthropic.com/research/formalizing-fermats-last-theorem"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-09-weathernext-3",
    "title": "WeatherNext 3: a global AI model for weather forecasting",
    "summary": "A model that learns from real-time geostationary satellite data and produces hourly forecasts, at resolutions down to 5 km for temperature and humidity (about five times finer than WeatherNext 2). For short-term precipitation, Google reports improvements of up to 60% over NASA's IMERG. These are ‘up to’ figures from Google's evaluations, with an independent ranking by Brightband.",
    "field": "climate",
    "ai_technology": "WeatherNext 3",
    "organization": "Google DeepMind; and Google Research",
    "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/",
    "source_type": "blog",
    "publication_date": "2026-09-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2609.03582",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Precipitation forecasts were scored with CRPS against baselines: gains of up to 60% over IMERG, 30% over MRMS and 10% over rain gauges for early lead times; independent live evaluations by Brightband ranked the model.",
      "unknown": "The 'up to' figures lack full test setup, longer-range gains are largest where forecasts are weak, and the atmosphere retains unpredictability.",
      "ai_role": "The model ingests live geostationary satellite data and historical analyses, trains on station observations and produces gridded fields, cyclone tracks and station-level forecasts.",
      "evidence_urls": [
        "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/introducing-weathernext-3/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-claude-leganti",
    "title": "Claude designs protein binders for 15 targets",
    "summary": "After an initial prompt, Claude selected binding sites, used design and folding models, and chose candidates, with success on 14 of 15 targets. Of 1,320 designs, 354 bound, for rates of 22–35% versus a typical 10–15%. None of the 90 designs for the maltose-binding protein worked. External labs ran the tests.",
    "field": "biology",
    "ai_technology": "Claude (Mythos Preview and Opus 4.8)",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/research/Claude-accelerates-protein-design",
    "source_type": "blog",
    "publication_date": "2026-08-18",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "External labs (Adaptyv Bio and Twist Bioscience) produced and tested the designs: 354 of 1,320 designs bound, with hit rates of 22-35% versus a typical 10-15%; none of 90 designs for maltose-binding protein bound. The collaborating external laboratories performed candidate production and binding assays within the discovery project; separate independent follow-up replication is not established by this account.",
      "unknown": "Anthropic plans further characterization to confirm hit rates and affinities; BBF-14 affinities are modest, GDF-8 data were inconclusive and excluded, and the TNFa result is unexplained.",
      "ai_role": "Claude chose binding sites, generated candidate structures and sequences by orchestrating structure, sequence and co-folding models, ran in silico optimization and chose candidates.",
      "evidence_urls": [
        "https://www.anthropic.com/research/Claude-accelerates-protein-design"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-radical-rai939",
    "title": "RAI-939: an autonomous-lab alloy tested with a torch",
    "summary": "Radical AI reports that AI-generated hypotheses and an autonomous lab produced new alloys, which were then tested by Purdue Applied Research Institute. During one minute under a torch, RAI-939 gained 0.25% mass while C103 lost 32%. This is a company announcement without peer review or replication details; scaling production remains to be developed.",
    "field": "materials",
    "ai_technology": "AI-generated hypotheses and autonomous laboratory",
    "organization": "Radical AI; and Purdue Applied Research Institute",
    "source_url": "https://www.radical-ai.com/news/purdue-applied-research-institute-confirms-superior-performance-of-radical-ai-materials",
    "source_type": "press release",
    "publication_date": "2026-08-07",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "A head-to-head torch test run by Purdue Applied Research Institute: RAI-939 gained 0.25% mass in one minute while C103 lost 32%; the page describes no repeated trials, sample sizes or replication.",
      "unknown": "Replication details are missing, the test articles were small, and production scaling is still to be developed.",
      "ai_role": "AI-generated hypotheses and an autonomous lab produced over 300 novel materials over 16 weeks.",
      "evidence_urls": [
        "https://www.radical-ai.com/news/purdue-applied-research-institute-confirms-superior-performance-of-radical-ai-materials"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-weathernext-cicloni",
    "title": "WeatherNext Cyclones: forecasting cyclones one day earlier",
    "summary": "Three-day forecasts match what earlier models achieved at two days: about 100 km trajectory error and 11 knots intensity error, compared with ECMWF-ENS and HWRF. The work is with the National Hurricane Center, CIRA, and the Met Office. The comparison uses historical 2023–2024 cyclones; why results are good at 28 km remains an open question.",
    "field": "climate",
    "ai_technology": "WeatherNext Cyclones",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/weathernext-ai-model-achieves-breakthrough-in-forecasting-cyclones/",
    "source_type": "blog",
    "publication_date": "2026-08-06",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-10953-2",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Historical cyclones from 2023-2024 were benchmarked deterministically and probabilistically against ECMWF-ENS (track) and HWRF (intensity): three-day errors matched earlier two-day forecasts; a 2025 storm is cited only as an anecdote.",
      "unknown": "Why the model is accurate at 28 km resolution is an open question; the 'extra day' is an average dependent on the chosen benchmarks, and the evaluation window is slightly unclear.",
      "ai_role": "A single model using Functional Generative Networks produced track, intensity and wind-structure ensembles of up to 1,000 members.",
      "evidence_urls": [
        "https://deepmind.google/blog/weathernext-ai-model-achieves-breakthrough-in-forecasting-cyclones/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-openai-dieci-risultati",
    "title": "OpenAI presents ten results in mathematics and theoretical computer science",
    "summary": "OpenAI announces ten results on codes, non-sofic groups, arithmetic circuits, quantum games, lattices, and combinatorics, accompanied, according to the source, by Lean certificates. This record groups one announcement; it does not mean this collection independently verified ten results.",
    "field": "mathematics",
    "ai_technology": "OpenAI internal model",
    "organization": "OpenAI",
    "source_url": "https://openai.com/index/ten-advances-in-mathematics/",
    "source_type": "press release",
    "publication_date": "2026-08-01",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 4,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Each of the ten arguments was prepared into a manuscript by humans with the model and then formalized by the model in a Lean certificate.",
      "unknown": "Not stated in the reviewed material",
      "ai_role": "An internal version of Astra found the results; humans prepared the manuscripts with the same model.",
      "evidence_urls": [
        "https://openai.com/index/ten-advances-in-mathematics/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-07-quantum-rl-willow",
    "title": "A quantum computer that learns from its own errors",
    "summary": "A reinforcement-learning agent tuned over 1,000 control parameters on the Willow chip: logical stability was 3.5 times higher under injected drift, with a further 20% reduction in logical error after expert calibration, and error below one per thousand cycles in the surface code. Drift was injected artificially, and scalability was shown only in simulation; published in Nature.",
    "field": "physics",
    "ai_technology": "Reinforcement learning for quantum control",
    "organization": "Google Quantum AI",
    "source_url": "https://research.google/blog/towards-a-quantum-computer-that-learns-from-its-errors/",
    "source_type": "blog",
    "publication_date": "2026-07-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.nature.com/articles/s41586-026-10759-2",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "On the Willow chip, drift was injected artificially into control parameters; RL steering raised logical stability 3.5-fold, and a further 20% error reduction followed expert calibration; scalability was tested only in simulation.",
      "unknown": "The drift was artificial rather than natural; scalability evidence is simulation-only; environmental decoherence cannot be fully prevented; fuller results need tighter integration.",
      "ai_role": "An RL agent used error-detection events as a learning signal and adjusted frequencies, amplitudes and phases during the computation without stopping it for recalibration.",
      "evidence_urls": [
        "https://research.google/blog/towards-a-quantum-computer-that-learns-from-its-errors/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-fastnet",
    "title": "FastNet: physically realistic AI forecasts from the Met Office",
    "summary": "Trained with a loss function that preserves energy across scales, it reaches a mean squared error equal to the Met Office global model and exceeds it on some metrics, with more realistic storm cores in Hurricane Ian and Storm Ciarán. The announcement gives no figures, covers only two extreme cases, and the system is not operational.",
    "field": "climate",
    "ai_technology": "FastNet (spherical-harmonic loss)",
    "organization": "Met Office; and Alan Turing Institute",
    "source_url": "https://www.metoffice.gov.uk/about-us/news-and-media/media-centre/weather-and-climate-news/2026/new-research-shows-ai-can-produce-trustworthy-physically-realistic-weather-forecasts-",
    "source_type": "press release",
    "publication_date": "2026-07-08",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1175/AIES-D-25-0090.1",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "FastNet was compared with the Met Office global model on verification measures (RMSE and some other metrics, shown only in a figure) and qualitatively on Hurricane Ian and Storm Ciaran storm cores.",
      "unknown": "The model is not operational, the text gives no numerical results, only two extreme cases are described, and comparable skill does not establish the same validation as the Global Model.",
      "ai_role": "A machine-learning weather model co-developed with the Alan Turing Institute, trained with a modified spherical-harmonic loss to keep energy across scales physically realistic.",
      "evidence_urls": [
        "https://www.metoffice.gov.uk/about-us/news-and-media/media-centre/weather-and-climate-news/2026/new-research-shows-ai-can-produce-trustworthy-physically-realistic-weather-forecasts-"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-06-gonorrea-a1",
    "title": "A1: deep learning identifies an antibiotic candidate against gonorrhea",
    "summary": "A screen of about six million compounds found candidates active against resistant Neisseria gonorrhoeae. A1 reduced bacteria in animal models, and proteomics, genetics, and biochemistry confirmed its target, alanine racemase. These are preclinical results, published in Science Translational Medicine.",
    "field": "medicine",
    "ai_technology": "Deep learning for molecule screening",
    "organization": "MIT; Karolinska Institutet, and collaborators",
    "source_url": "https://news.ki.se/ai-helps-identify-new-antibiotic-candidates-against-drug-resistant-gonorrhoea",
    "source_type": "press release",
    "publication_date": "2026-06-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/scitranslmed.ads4699",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Phenotypic screen of 38,650 molecules trained a graph neural network, which screened about 6 million compounds; 213 were validated experimentally, 83 inhibited growth. Two hits were tested in vitro against multidrug-resistant strains, by proteomics (alanine racemase target), in a vagina-on-a-chip model and in a mouse vaginal infection model.",
      "unknown": "Not reported (the abstract describes the in vivo and organ-chip results as showing only early promise).",
      "ai_role": "A graph neural network trained on phenotypic screening data, used to virtually screen the compound library before experimental validation.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/scitranslmed.ads4699&resultType=core&format=json",
        "https://news.ki.se/ai-helps-identify-new-antibiotic-candidates-against-drug-resistant-gonorrhoea"
      ],
      "test_type": "direct test"
    }
  },
  {
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "source_type": "press release",
    "id": "2026-06-o3-malattie-genetiche",
    "title": "Genetic reanalysis assisted by o3 contributes to 18 diagnoses",
    "summary": "Among 376 unresolved cases, o3 Deep Research proposed explanations reviewed by specialists: 18 diagnoses were clinically confirmed, seven of which had already been established elsewhere but were missing from local records. This was a retrospective study across heterogeneous cohorts, without a measure of time saved or effect on care; clinicians made the decisions.",
    "field": "medicine",
    "ai_technology": "o3 Deep Research",
    "organization": "OpenAI; Boston Children’s Hospital, and Harvard University",
    "source_url": "https://openai.com/index/diagnose-rare-childhood-diseases/",
    "publication_date": "2026-06-18",
    "paper_url": "https://doi.org/10.1056/aics2501343",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Retrospective multicohort reanalysis of 376 unresolved cases; a diagnosis required a pathogenic or likely pathogenic variant confirmed in a CLIA-certified lab and returned to families, and seven rediscoveries were also reported.",
      "unknown": "The abstract states that the retrospective design motivates prospective multicenter evaluation with predefined endpoints, calibration reporting and comparator baselines.",
      "ai_role": "An LLM ingested clinician notes, HPO terms and a filtered variant table and proposed candidate hypotheses for expert adjudication under ACMG/AMP criteria; experts made the classifications.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1056/AIcs2501343&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-06-gpt54-chan-lam",
    "title": "An AI system improves a Chan–Lam coupling reaction",
    "summary": "The system identified TEMPO as a useful additive for coupling primary sulfonamides; chemists reproduced yield improvements in 11 of 14 substrate pairs at bench scale. The work retains human supervision and specialized infrastructure; independent reproducibility and generalization remain to be checked.",
    "field": "chemistry",
    "ai_technology": "GPT-5.4 and Maria",
    "organization": "OpenAI; and Molecule.one",
    "source_url": "https://openai.com/index/ai-chemist-improves-reaction/",
    "source_type": "press release",
    "publication_date": "2026-06-17",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Over two cycles in an automated lab, yields improved for 88% of boronic acids and 83% of sulfonamides tested (mean 16.6% to 25.2%); human chemists repeated representative reactions at bench scale, with higher yields for 11 of 14 substrate pairs.",
      "unknown": "Not stated in the reviewed material",
      "ai_role": "GPT-5.4 connected to Molecule.one Maria proposed, designed and analyzed experiments; humans steered prompts, selected proposals, made limited corrections and validated the final result.",
      "evidence_urls": [
        "https://openai.com/index/ai-chemist-improves-reaction/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-electrolytegpt-battery-formulations",
    "title": "ElectrolyteGPT can generate new formulations for battery development",
    "summary": "University of Chicago researchers trained a GPT model with a new notation (fLine) to generate complete battery electrolyte formulations, including solvent ratios, salt concentrations and temperature. Several AI-proposed compositions were synthesized and, per the release, performed on par with the state of the art in lithium metal batteries. No cycling or capacity figures are given, the work is early, and generated candidates still need experimental verification.",
    "field": "materials",
    "ai_technology": "ElectrolyteGPT (GPT with fLine notation)",
    "organization": "University of Chicago Pritzker School of Molecular Engineering",
    "source_url": "https://news.uchicago.edu/story/electrolytegpt-can-generate-new-formulations-battery-development",
    "source_type": "press release",
    "publication_date": "2026-06-03",
    "added_date": "2026-11-13",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/jacsau.5c01628",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Generated solvents and formulations were synthesized and fabricated, found to meet the target property requirements, and enabled long-term cycling in anode-free lithium metal batteries.",
      "unknown": "The release states the limited data and parameter set, and that the results matched but did not exceed top existing electrolytes; the abstract states no limitations.",
      "ai_role": "A transformer-based generative model (ElectrolyteGPT) generated candidate solvents and electrolyte formulations conditioned on property targets.",
      "evidence_urls": [
        "https://news.uchicago.edu/story/electrolytegpt-can-generate-new-formulations-battery-development",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1021/jacsau.5c01628&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-openai-distanza-unitaria",
    "title": "OpenAI presents a disproof of the unit-distance conjecture",
    "summary": "OpenAI presents a construction of planar point configurations with more unit-distance pairs than predicted by the Erdős conjecture. The announcement says mathematicians outside the team checked it and provides a complementary explanation; this record checks the source, not an independent reconstruction of the proof.",
    "field": "mathematics",
    "ai_technology": "OpenAI internal model",
    "organization": "OpenAI",
    "source_url": "https://openai.com/index/model-disproves-discrete-geometry-conjecture/",
    "source_type": "press release",
    "publication_date": "2026-05-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 4,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The proof was checked by a group of external mathematicians, who also wrote a companion paper explaining the argument.",
      "unknown": "Not stated in the reviewed material",
      "ai_role": "A general-purpose internal OpenAI reasoning model produced the proof while being evaluated on Erdős problems.",
      "evidence_urls": [
        "https://openai.com/index/model-disproves-discrete-geometry-conjecture/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-05-co-scientist-fibrosi-epatica",
    "title": "Co-Scientist: repurposed drugs against liver fibrosis",
    "summary": "In Gary Peltz's Stanford lab, two of the three existing drugs proposed by Co-Scientist blocked fibrosis and promoted regeneration in human liver cells, while Peltz's own two choices showed no benefit. The anticancer drug vorinostat blocked 91% of a damage response that can lead to scarring. Cell experiments only; published in Advanced Science.",
    "field": "medicine",
    "ai_technology": "Co-Scientist",
    "organization": "Google DeepMind; and Stanford",
    "source_url": "https://deepmind.google/blog/uncovering-repurposed-medicines-to-fight-liver-fibrosis/",
    "source_type": "blog",
    "publication_date": "2026-05-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202508751",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The anti-fibrotic efficacy and toxicity of 25 drugs were assessed serially in human liver organoids grown in microwells; Co-Scientist's candidates were tested in cells alongside the authors' own picks.",
      "unknown": "Not reported.",
      "ai_role": "The AI co-scientist multi-agent system proposed three candidate drugs with reasoning, and helped direct their experimental characterization.",
      "evidence_urls": [
        "https://deepmind.google/blog/uncovering-repurposed-medicines-to-fight-liver-fibrosis/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1002/advs.202508751&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-co-scientist-invecchiamento-calico",
    "title": "Co-Scientist and aging: metabolism and the integrated stress response",
    "summary": "Calico Life Sciences used Co-Scientist to formulate the hypothesis that metabolism regulates the cellular integrated stress response, refined experiments with the tool, and reports new findings. The post gives no numerical data, and the results have not yet been published or peer reviewed.",
    "field": "biology",
    "ai_technology": "Co-Scientist",
    "organization": "Google DeepMind; and Calico Life Sciences",
    "source_url": "https://deepmind.google/blog/opening-new-paths-in-aging-research/",
    "source_type": "blog",
    "publication_date": "2026-05-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Calico reports that experiments were run with the Co-Scientist hypothesis and produced new findings; the post does not describe the experiments, assays or results.",
      "unknown": "Not reported (the results are not yet published; the team plans to publish them).",
      "ai_role": "Co-Scientist generated the hypothesis that metabolism regulates the integrated stress response, and was used to refine experimental design.",
      "evidence_urls": [
        "https://deepmind.google/blog/opening-new-paths-in-aging-research/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2026-05-co-scientist-fegato-mash",
    "title": "Co-Scientist and mechanisms of metabolic steatohepatitis (MASH)",
    "summary": "Filippo Menolascina (Edinburgh) used Co-Scientist to search the literature for combinations of MASH therapies. The tool proposed the NLRP3 inflammasome as a link between inflammation and metabolism, and the post says the hypothesis was later tested experimentally. No numbers are given; the drug combinations were suggested for testing, not demonstrated.",
    "field": "medicine",
    "ai_technology": "Co-Scientist",
    "organization": "Google DeepMind; and University of Edinburgh",
    "source_url": "https://deepmind.google/blog/accelerating-discovery-of-liver-disease-mechanisms/",
    "source_type": "blog",
    "publication_date": "2026-05-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The article says the NLRP3 hypothesis was later experimentally verified but does not describe the experiments, who carried them out, or their results.",
      "unknown": "Not reported (the article does not state limitations or whether the work is published).",
      "ai_role": "Co-Scientist searched the literature, proposed drug combinations for MASH therapy and proposed the NLRP3 inflammasome as the link between inflammation and metabolism.",
      "evidence_urls": [
        "https://deepmind.google/blog/accelerating-discovery-of-liver-disease-mechanisms/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2026-05-co-scientist-invecchiamento-cellulare",
    "title": "Co-Scientist: genetic factors to rejuvenate cells",
    "summary": "Omar Abudayyeh and Jonathan Gootenberg used Co-Scientist to search tens of thousands of papers, obtaining over 20 proposed genetic factors for reversing cellular aging, a couple of which were validated in the lab, and reducing analysis of genetic-screening data from six months to a few days. The post gives no sample size or peer review status.",
    "field": "biology",
    "ai_technology": "Co-Scientist",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/fast-tracking-genetic-leads-to-reverse-cellular-aging/",
    "source_type": "blog",
    "publication_date": "2026-05-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Lab tests validated a couple of the more than 20 proposed genetic factors, which pushed cells toward a younger state; the experiments and exact number are not described.",
      "unknown": "Not reported (the post gives no sample size or peer-review status and contains only the investigators' quotes).",
      "ai_role": "Co-Scientist scanned tens of thousands of papers, proposed genetic factors for reversing cellular aging and analysed screening data alongside the literature.",
      "evidence_urls": [
        "https://deepmind.google/blog/fast-tracking-genetic-leads-to-reverse-cellular-aging/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-weathernext-melissa",
    "title": "WeatherNext and Hurricane Melissa: earlier warning in Jamaica",
    "summary": "WeatherNext forecast a Category 5 landfall in Jamaica five days ahead with 80% probability, rising to nearly 100% at three days. According to the post, it was the best single model for track and intensity in the National Hurricane Center's 2025 verification report, which used it alongside its HAFS model. The post gives no numerical comparisons; rapid intensification remains very hard to predict.",
    "field": "climate",
    "ai_technology": "WeatherNext",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/how-weathernext-helped-the-national-hurricane-center-better-predict-hurricane-melissas-historic-landfall-in-jamaica/",
    "source_type": "blog",
    "publication_date": "2026-05-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The National Hurricane Center's 2025 annual verification report found WeatherNext the top-performing individual model for track and intensity; the Melissa forecast is presented as a single case.",
      "unknown": "The page notes that predicting rapid intensification is very difficult and that Weather Lab is experimental; no dedicated limitations section.",
      "ai_role": "WeatherNext, an AI weather model with 50-member ensembles, forecast the Hurricane Melissa landfall alongside physics-based HAFS and was used by NHC forecasters.",
      "evidence_urls": [
        "https://deepmind.google/blog/how-weathernext-helped-the-national-hurricane-center-better-predict-hurricane-melissas-historic-landfall-in-jamaica/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-05-mattersim-tap",
    "title": "MatterSim and tantalum phosphide as a good heat conductor",
    "summary": "A candidate identified with MatterSim-v1, tetragonal tantalum phosphide, was synthesized and showed thermal conductivity of 152 W/m/K in the best samples, close to silicon. MatterSim-v1 inference is 3–5 times faster, and the new multi-task model MatterSim-MT predicts energies, forces, and properties. Battery and other property results are predictions, not experimental confirmations.",
    "field": "materials",
    "ai_technology": "MatterSim-v1 and MatterSim-MT",
    "organization": "Microsoft Research",
    "source_url": "https://www.microsoft.com/en-us/research/blog/advancing-ai-for-materials-with-mattersim-experimental-synthesis-faster-simulation-and-multi-task-models/",
    "source_type": "blog",
    "publication_date": "2026-05-12",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "A MatterSim-v1 screen of over 240,000 candidates picked tetragonal tantalum phosphide, which was synthesized at UT Dallas and whose thermal conductivity was measured at UIUC, reaching 152 W/m/K in the best samples; the page does not describe the measurement method or number of samples.",
      "unknown": "The page states that practical constraints such as mass density, elemental abundance and environmental stability are needed, but does not say whether TaP meets them.",
      "ai_role": "MatterSim-v1 machine-learning interatomic potentials estimated phonon thermal conductivity to screen candidates before experimental validation; MatterSim-MT is a multi-task model.",
      "evidence_urls": [
        "https://www.microsoft.com/en-us/research/blog/advancing-ai-for-materials-with-mattersim-experimental-synthesis-faster-simulation-and-multi-task-models/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-redmod-pancreas",
    "title": "REDMOD: pancreatic cancer signals in CT scans taken before diagnosis",
    "summary": "Among nearly 2,000 abdominal CT scans originally considered normal, an AI radiomics model identified 73% of prediagnostic pancreatic cancers, with a median lead time of about 16 months. Validation included images from multiple institutions and protocols. This analysis used scans taken before diagnosis; clinical benefit and false positives in practice still need prospective evaluation.",
    "field": "medicine",
    "ai_technology": "REDMOD (Radiomics-based Early Detection Model)",
    "organization": "Mayo Clinic",
    "source_url": "https://newsnetwork.mayoclinic.org/discussion/mayo-clinic-ai-detects-pancreatic-cancer-up-to-3-years-before-diagnosis-in-landmark-validation-study/",
    "source_type": "press release",
    "publication_date": "2026-04-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://gut.bmj.com/content/early/2026/04/22/gutjnl-2025-337266",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "REDMOD was trained on 969 patients from a multi-institutional cohort and tested on an independent set of 493 patients (63 pre-diagnostic), with external specificity checks on further datasets and a comparison with radiologists; AUC 0.82 and 73.0% sensitivity, versus 38.9% for radiologists.",
      "unknown": "The article states that prospective validation in high-risk cohorts is the necessary next step; no explicit limitations section.",
      "ai_role": "A fully automated radiomics framework with AI segmentation and an ensemble classifier on a 40-feature signature from standard-of-care CT scans.",
      "evidence_urls": [
        "https://newsnetwork.mayoclinic.org/discussion/mayo-clinic-ai-detects-pancreatic-cancer-up-to-3-years-before-diagnosis-in-landmark-validation-study/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1136/gutjnl-2025-337266&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-04-gpt55-ramsey",
    "title": "GPT-5.5 and a proof about Ramsey numbers",
    "summary": "OpenAI reports that an internal version of GPT-5.5, with a purpose-built scaffold, found a proof of a known asymptotic result on off-diagonal Ramsey numbers, then verified in Lean. The announcement appeared with the model launch on April 23, 2026; the full proof is in a separate PDF now linked. Page rechecked November 12, 2026: the text matches.",
    "field": "mathematics",
    "ai_technology": "GPT-5.5 (internal version)",
    "organization": "OpenAI",
    "source_url": "https://openai.com/index/introducing-gpt-5-5/",
    "source_type": "blog",
    "publication_date": "2026-04-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://cdn.openai.com/pdf/6dc7175d-d9e7-4b8d-96b8-48fe5798cd5b/Ramsey.pdf",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "GPT-5.5 found a proof of an asymptotic fact about off-diagonal Ramsey numbers, later verified in Lean.",
      "unknown": "Not stated in the reviewed material",
      "ai_role": "An internal version of GPT-5.5 with a custom harness found the proof.",
      "evidence_urls": [
        "https://openai.com/index/introducing-gpt-5-5/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-03-vibe-physics",
    "title": "Theoretical physics with Claude: the Sudakov shoulder in the C parameter",
    "summary": "Physicist Matthew Schwartz (Harvard) guided Claude Opus 4.5 through a resummation calculation for e+e− collisions, producing a paper with a new factorization theorem and testable predictions against data in about two weeks. Schwartz checked every step; the model claimed checks it had not done and made errors.",
    "field": "physics",
    "ai_technology": "Claude Opus 4.5",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/research/vibe-physics",
    "source_type": "blog",
    "publication_date": "2026-03-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2601.02484",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Results are validated against EVENT2 Monte Carlo simulations, with analytic curves matching the histograms; the paper does not describe a comparison with experimental data, and the author checked each step, with GPT and Gemini cross-checking the written calculations.",
      "unknown": "The paper does not describe comparison with experimental data (the author mentions a follow-up project for this); the author states that LLMs cannot yet do original theoretical physics autonomously and that the work needed expert evaluation.",
      "ai_role": "Claude Opus 4.5, run through Claude Code, performed the calculations including the factorization derivation, numerical analysis, simulations and manuscript preparation, under physicist supervision.",
      "evidence_urls": [
        "https://www.anthropic.com/research/vibe-physics",
        "https://arxiv.org/abs/2601.02484"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "source_type": "press release",
    "id": "2026-03-gpt52-gravitoni",
    "title": "GPT-5.2 Pro extends a scattering-amplitude result to gravitons",
    "summary": "Building on work on gluons, GPT-5.2 Pro helped derive formulas for nonzero single-minus gravitational amplitudes in a special semi-collinear regime. The authors report analytical verification and checks against known physical limits; this is a theoretical preprint without experimental confirmation.",
    "field": "physics",
    "ai_technology": "GPT-5.2 Pro",
    "organization": "OpenAI; and academic collaborators",
    "source_url": "https://openai.com/index/extending-single-minus-amplitudes-to-gravitons/",
    "publication_date": "2026-03-04",
    "paper_url": "https://cdn.openai.com/pdf/graviton.pdf",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The preprint derives explicit formulas for single-minus graviton tree amplitudes in the half-collinear regime and shows they follow from symmetry principles and recursion relations; no experimental test is described.",
      "unknown": "The page calls it a small step towards reconciling quantum mechanics with general relativity.",
      "ai_role": "Researchers used GPT-5.2 Pro to help find the result; the paper is authored by named physicists on behalf of OpenAI.",
      "evidence_urls": [
        "https://openai.com/index/extending-single-minus-amplitudes-to-gravitons/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-gpt52-gluoni",
    "title": "GPT-5.2 proposes a formula for nonzero gluon amplitudes",
    "summary": "GPT-5.2 conjectured a formula for gluon scattering amplitudes with a single negative helicity in a special semi-collinear regime. An internal model produced a proof that the authors analytically verified; the announcement describes a preprint, not experimental confirmation.",
    "field": "physics",
    "ai_technology": "GPT-5.2 Pro",
    "organization": "OpenAI; and academic collaborators",
    "source_url": "https://openai.com/index/new-result-theoretical-physics/",
    "source_type": "press release",
    "publication_date": "2026-02-13",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2602.12176",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The arXiv abstract reports that the gluon formula satisfies several consistency conditions, including Weinberg's soft theorem, and that the formula is restricted to half-collinear configurations; no experimental test is described.",
      "unknown": "Not reported (the abstract states no limitations beyond the restricted half-collinear and complexified-momentum setting).",
      "ai_role": "GPT-5.2 conjectured the formula; an internal model produced a proof that the authors verified analytically.",
      "evidence_urls": [
        "https://arxiv.org/abs/2602.12176"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-02-deep-think-ricerca",
    "title": "Gemini Deep Think on research problems in mathematics, physics, and computer science",
    "summary": "The Aletheia mathematical agent scored up to 90% on IMO-ProofBench Advanced and independently solved four open Erdős questions; work with experts covered 18 research problems, including a disproof of a submodular optimization conjecture. Its score on doctoral-level problems falls to about 38–46%; no result is claimed as a breakthrough, and the papers were submitted to journals.",
    "field": "mathematics",
    "ai_technology": "Gemini Deep Think, Aletheia agent",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/accelerating-mathematical-and-scientific-discovery-with-gemini-deep-think/",
    "source_type": "blog",
    "publication_date": "2026-02-11",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2602.03837",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Aletheia was evaluated on IMO-ProofBench Advanced (about 90%, human experts graded), an internal PhD-level benchmark (about 46%), a semi-autonomous run over 700 open Erdős problems (four autonomous solutions), and expert collaborations; Level 2 papers were submitted to journals.",
      "unknown": "The article states that no Level 3 or Level 4 results are claimed; the abstract states no limitations, and the PhD-level score falls to about 38-46%.",
      "ai_role": "Aletheia, a math research agent built on Gemini Deep Think, generated, verified and revised proofs and used search, with varying autonomy across projects.",
      "evidence_urls": [
        "https://deepmind.google/blog/accelerating-mathematical-and-scientific-discovery-with-gemini-deep-think/",
        "https://arxiv.org/abs/2602.03837"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-02-congettura-fel",
    "title": "Fel's conjecture proved and verified in Lean",
    "summary": "Axiom Prover generated a proof of Fel's conjecture on syzygies of numerical semigroups, verified in Lean 4 with Mathlib, without hints, lemmas, or manual corrections after the run began. This is a case study and an unreviewed preprint, with no data on attempts or time taken.",
    "field": "mathematics",
    "ai_technology": "Axiom Prover, Lean",
    "organization": "Axiom Math",
    "source_url": "https://axiommath.ai/research/proof-of-concept",
    "source_type": "blog",
    "publication_date": "2026-02-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2602.03716",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The proof of Fel's conjecture was generated by AxiomProver with no hints, lemmas or manual corrections after kickoff and checked by Lean 4 with Mathlib; the formalization is publicly available and the mathematical results appear in a human-written arXiv preprint.",
      "unknown": "The authors state the limiting factor is often the quality of the shared library (Mathlib), and the post describes a single case study with no claim about automating research.",
      "ai_role": "AxiomProver translated the statement into Lean, generated intermediate lemmas and assembled the complete Lean proof.",
      "evidence_urls": [
        "https://axiommath.ai/research/proof-of-concept",
        "https://arxiv.org/abs/2602.03716"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-02-gpt5-cfps",
    "title": "GPT-5 reduces the cost of cell-free protein synthesis",
    "summary": "A cycle between GPT-5 and a robotic lab explored over 36,000 compositions and reduced protein-production cost by 40% compared with the previous baseline. This is an experimental result in one CFPS setup; it does not establish the same reduction for other proteins, scales, or facilities.",
    "field": "biology",
    "ai_technology": "GPT-5",
    "organization": "OpenAI; and Ginkgo Bioworks",
    "source_url": "https://openai.com/index/gpt-5-lowers-protein-synthesis-cost/",
    "source_type": "press release",
    "publication_date": "2026-02-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Over six rounds in a cloud laboratory, the system tested more than 36,000 reaction compositions on 580 plates and reached a 40% lower protein production cost (57% lower reagent cost).",
      "unknown": "Not stated in the reviewed material",
      "ai_role": "GPT-5 proposed and analyzed experiments in a closed loop with Ginkgo Bioworks automation.",
      "evidence_urls": [
        "https://openai.com/index/gpt-5-lowers-protein-synthesis-cost/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-dinotracker",
    "title": "DinoTracker: AI classifies dinosaur footprints",
    "summary": "An unsupervised model trained on nearly 2,000 fossil footprints and millions of simulated variations identifies eight features of variation and agrees with experts about 90% of the time. Some tracks over 200 million years old resemble bird feet, which could suggest bird origins tens of millions of years earlier. The interpretation is open: they could be early dinosaurs whose feet happened to look similar.",
    "field": "other",
    "ai_technology": "DinoTracker (unsupervised model)",
    "organization": "Helmholtz-Zentrum Berlin; and University of Edinburgh",
    "source_url": "https://www.ed.ac.uk/news/ai-sheds-light-on-mysterious-dinosaur-footprints",
    "source_type": "press release",
    "publication_date": "2026-01-27",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2527222122",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An unsupervised variational autoencoder was trained on 1,974 fossil footprints; after labeling, its groupings agreed with published expert identifications in 80-93% of cases (the press release says about 90%).",
      "unknown": "The interpretation is open: the tracks could be early dinosaurs whose feet resembled birds' feet; the study states no limitations section.",
      "ai_role": "An unsupervised disentangled variational autoencoder found eight shape-variation features without prior labels, and predicted which dinosaur made each track.",
      "evidence_urls": [
        "https://www.ed.ac.uk/news/ai-sheds-light-on-mysterious-dinosaur-footprints",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1073/pnas.2527222122&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-01-anomalymatch-hubble",
    "title": "AnomalyMatch finds over 1,300 unusual objects in the Hubble archive",
    "summary": "The neural network scanned about 100 million cutouts from the Hubble Legacy Archive in roughly two and a half days: nearly 1,400 anomalies were flagged and over 1,300 confirmed, including more than 800 not previously documented in the literature (mergers, lenses, and jellyfish galaxies). AI only ranks candidates; the two counts are not explained, and no false-positive rate is given.",
    "field": "physics",
    "ai_technology": "AnomalyMatch (neural network)",
    "organization": "ESA; and ESA-Hubble",
    "source_url": "https://esahubble.org/news/heic2603/",
    "source_type": "press release",
    "publication_date": "2026-01-27",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202555512",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The neural network screened about 100 million Hubble Legacy Archive cutouts, and the authors personally inspected the top-ranked sources; more than 1,300 were confirmed as true anomalies, of which more than 800 were not previously documented; no false-positive rate is given.",
      "unknown": "The release gives no false-positive rate, does not state how many candidates were inspected, notes that several dozen objects defied classification, and says expert review is still required.",
      "ai_role": "AnomalyMatch, a neural network, screened the archive and ranked candidate anomalies for expert inspection.",
      "evidence_urls": [
        "https://esahubble.org/news/heic2603/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-01-claude-biology-labs",
    "title": "Claude in biology labs: Biomni, MozzareLLM, and others",
    "summary": "Anthropic describes three cases: Biomni (Stanford) completed a GWAS analysis in 20 minutes in an initial trial and suggested new transcription factors in embryonic development; MozzareLLM (MIT and Whitehead) interprets gene clusters from CRISPR screens; a third lab is still testing. This is an Anthropic account without independent evaluation; some figures are estimates.",
    "field": "biology",
    "ai_technology": "Claude (Biomni, MozzareLLM)",
    "organization": "Anthropic",
    "source_url": "https://www.anthropic.com/news/accelerating-scientific-research",
    "source_type": "blog",
    "publication_date": "2026-01-15",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Biomni's case studies (GWAS analysis, cloning design blind-evaluated against a postdoc, embryonic-tissue cell analysis) were checked by the Biomni team, citing a bioRxiv preprint; MozzareLLM was assessed qualitatively by its lab, with no quantitative benchmark.",
      "unknown": "The article says the validation is not independent, some figures are estimates (such as the three-week human comparison), and Biomni cannot yet do everything out of the box.",
      "ai_role": "A Claude-powered agent (Biomni) analysed GWAS and embryonic cell data and suggested transcription factors; a Claude-powered system (MozzareLLM) interprets gene clusters from CRISPR screens.",
      "evidence_urls": [
        "https://www.anthropic.com/news/accelerating-scientific-research"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-12-ipernucleo-doppio-lambda",
    "title": "A double-Lambda hypernucleus of boron-13 found with deep learning",
    "summary": "Neural networks selected images from the J-PARC E07 emulsion plates; microscopy and kinematic analysis then confirmed an event as a double-Lambda hypernucleus of boron-13. The discovery comes from analysis of 0.2% of the emulsions and is based on a single event; published in Nature Communications.",
    "field": "physics",
    "ai_technology": "Deep learning for nuclear-emulsion analysis",
    "organization": "RIKEN; and J-PARC E07 collaboration",
    "source_url": "https://www.riken.jp/en/news_pubs/research_news/pr/2025/20251218_1/index.html",
    "source_type": "press release",
    "publication_date": "2025-12-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-66517-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Neural networks pre-selected candidate images from 0.2% of the J-PARC E07 emulsion data; microscopy and detailed kinematic analysis then confirmed one event as a boron-13 double-Lambda hypernucleus.",
      "unknown": "Only one event is confirmed; the estimate of more than 2,000 events in the full dataset is an extrapolation from the detection rate, and the article has no limitations section.",
      "ai_role": "Deep learning networks trained to recognize double-strangeness signatures selected candidate images for microscopic examination.",
      "evidence_urls": [
        "https://www.riken.jp/en/news_pubs/research_news/pr/2025/20251218_1/index.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-noaa-modelli-ai",
    "title": "NOAA puts three AI-based global models into operation",
    "summary": "AIGFS (retrained from GraphCast) uses up to 99.7% less computing than GFS (a 16-day forecast in about 40 minutes); AIGEFS has skill comparable to GEFS with 9% of the computing; and HGEFS (62 hybrid members) outperforms GEFS and AIGEFS on most metrics. Cyclone intensity worsened in AIGFS v1.0; the announcement gives no numerical scores.",
    "field": "climate",
    "ai_technology": "AIGFS, AIGEFS, and HGEFS (GraphCast-derived)",
    "organization": "NOAA",
    "source_url": "https://www.noaa.gov/news-release/noaa-deploys-new-generation-of-ai-driven-global-weather-models",
    "source_type": "press release",
    "publication_date": "2025-12-17",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 1,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The release describes initial or early testing and claims skill comparable to GEFS and better ensemble performance, but gives no metrics, verification datasets or time periods.",
      "unknown": "The release states that AIGFS v1.0 degraded tropical cyclone intensity forecasts, that AIGEFS ensemble spread is still being improved, and that HGEFS hurricane intensity is being improved.",
      "ai_role": "AI global models: AIGFS forecasts with up to 99.7% less compute; AIGEFS is a 31-member AI ensemble with 9% of the GEFS computing; HGEFS combines 31 GEFS and 31 AIGEFS members.",
      "evidence_urls": [
        "https://www.noaa.gov/news-release/noaa-deploys-new-generation-of-ai-driven-global-weather-models"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2025-12-gpt5-laboratorio",
    "title": "GPT-5 and molecular cloning: 79-fold greater efficiency",
    "summary": "In a cycle with Red Queen Bio, GPT-5 optimized a cloning protocol, introducing RecA and gp32 as a new mechanism: 2.6-fold improvement with the enzymatic reaction, 36-fold with transformation, and 79-fold when combined. This applies to the model system used, with scientists carrying out the protocols; reported by OpenAI.",
    "field": "biology",
    "ai_technology": "GPT-5",
    "organization": "OpenAI",
    "source_url": "https://openai.com/index/accelerating-biological-research-in-the-wet-lab/",
    "source_type": "blog",
    "publication_date": "2025-12-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Scientists carried out GPT-5-modified cloning protocols over several rounds; the final protocol recovered 79 times more sequence-verified clones for a fixed amount of input DNA than the baseline.",
      "unknown": "The work used a benign experimental system with a limited task scope.",
      "ai_role": "GPT-5 reasoned about the protocol and proposed modifications, including a RecA and gp32 mechanism; humans only ran the protocol and uploaded data.",
      "evidence_urls": [
        "https://openai.com/index/accelerating-biological-research-in-the-wet-lab/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-sandia-idrogeno",
    "title": "Graph neural networks identify oxides for hydrogen production",
    "summary": "A model of oxygen-vacancy properties selected twelve oxides to test for thermochemical hydrogen production; about 80% were active in experiments. Some were comparable to or better than the baseline only under certain reactor conditions; further study and optimization are needed.",
    "field": "materials",
    "ai_technology": "Graph neural networks for crystal defects",
    "organization": "Sandia National Laboratories; and HydroGEN collaborators",
    "source_url": "https://energy.sandia.gov/news/sandia-team-uses-ai-ml-models-to-predict-useful-water-splitting-and-hydrogen-generation-materials/",
    "source_type": "press release",
    "publication_date": "2025-12-15",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D5MH01566A",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "Twelve metal oxides selected by the model for oxygen-vacancy properties were tested experimentally for thermochemical hydrogen production; about 80% were effective, and some matched or beat the state-of-the-art material only under certain reactor conditions.",
      "unknown": "The method so far covers only thermochemical hydrogen materials, the model needs more diverse training data and more precise calculations, and the page gives no baseline numbers.",
      "ai_role": "A graph neural network trained on first-principles calculations predicted oxygen-vacancy formation energies and screened materials from existing databases.",
      "evidence_urls": [
        "https://energy.sandia.gov/news/sandia-team-uses-ai-ml-models-to-predict-useful-water-splitting-and-hydrogen-generation-materials/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-gpt52-apprendimento",
    "title": "GPT-5.2 Pro and an open problem about learning curves",
    "summary": "GPT-5.2 Pro solved an open problem in statistical learning theory (learning curves are monotonic in the Gaussian case with known mean and unknown standard deviation), then extended it to higher dimensions. The authors checked the proof, including with outside experts. Reported by OpenAI.",
    "field": "mathematics",
    "ai_technology": "GPT-5.2 Pro",
    "organization": "OpenAI",
    "source_url": "https://openai.com/index/gpt-5-2-for-science-and-math/",
    "source_type": "blog",
    "publication_date": "2025-12-11",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2512.10220",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors verified and transcribed the model's proofs of monotonicity of the maximum likelihood estimator's KL divergence for Gaussian and Gamma families, with no outside expert review described.",
      "unknown": "The guarantees cover only the listed well-specified parametric settings, and the reverse-KL result relies on a folklore trick; the page shows only the abstract.",
      "ai_role": "All results were derived by variants of GPT-5.2 Pro, with humans prompting for further results and verifying the proofs.",
      "evidence_urls": [
        "https://arxiv.org/abs/2512.10220"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-colture-resistenti-calore",
    "title": "Heat-resistant crops: a hybrid photosynthesis enzyme",
    "summary": "At Michigan State University, AlphaFold predicted the 3D structure of glycerate kinase in a crop and a heat-loving alga. Replacing three flexible loops with rigid loops from the alga produced a hybrid that remained stable up to 65 °C. The enzyme has not yet been tested in whole plants.",
    "field": "biology",
    "ai_technology": "AlphaFold",
    "organization": "Google DeepMind; and Michigan State University",
    "source_url": "https://deepmind.google/blog/engineering-more-resilient-crops-for-a-warming-climate/",
    "source_type": "blog",
    "publication_date": "2025-12-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Hybrid glycerate kinase enzymes with three loops replaced by rigid loops from a heat-loving alga were tested in the lab, and one stayed stable up to 65 °C; whole-plant testing has not been done.",
      "unknown": "It is not yet known whether the hybrid enzyme improves heat resilience in a whole plant; the page does not say whether the AlphaFold-predicted structures were experimentally validated.",
      "ai_role": "AlphaFold predicted the enzyme's 3D structures, which were then used in molecular simulations to identify loops to modify.",
      "evidence_urls": [
        "https://deepmind.google/blog/engineering-more-resilient-crops-for-a-warming-climate/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-11-apob100-colesterolo",
    "title": "The structure of the LDL protein apoB100",
    "summary": "At the University of Missouri, Zachary Berndsen and Keith Cassidy used AlphaFold to predict the structure of apoB100, the scaffold of LDL particles, and refined it using cryo-electron microscopy data. The result is a cage-like shell with a ribbon belt. The post gives no figures such as resolution and no clinical validation.",
    "field": "medicine",
    "ai_technology": "AlphaFold",
    "organization": "Google DeepMind; and University of Missouri",
    "source_url": "https://deepmind.google/blog/revealing-a-key-protein-behind-heart-disease/",
    "source_type": "blog",
    "publication_date": "2025-11-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08467-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Cryo-EM images were refined with AlphaFold2 predictions; the resulting apoB100 structure was compared with more than 200 intramolecular cross-links and showed close agreement, with subnanometre resolution in most regions.",
      "unknown": "The article states that clinical applications will take time and the structure's implications are not yet realized; it gives no resolution figures or clinical validation, and the abstract states that resolution is subnanometre only in most regions.",
      "ai_role": "AlphaFold2 generated the initial atomic-resolution structure predictions, which were refined against cryo-EM data.",
      "evidence_urls": [
        "https://deepmind.google/blog/revealing-a-key-protein-behind-heart-disease/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-024-08467-w&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-11-api-vitellogenina",
    "title": "Healthier bees: AlphaFold predicts vitellogenin's structure",
    "summary": "With the Norwegian University of Life Sciences, predicted structures of honeybee vitellogenin were confirmed by electron microscopy in July 2025; AlphaFold 3 helped study its interactions with other molecules. Early evidence suggests breeding times for resilient bees could fall from months to weeks, a still preliminary claim.",
    "field": "biology",
    "ai_technology": "AlphaFold, AlphaFold 3",
    "organization": "Google DeepMind; and Norwegian University of Life Sciences",
    "source_url": "https://deepmind.google/blog/breeding-healthier-and-stronger-honeybees/",
    "source_type": "blog",
    "publication_date": "2025-11-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-58575-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "AlphaFold-predicted vitellogenin structures were confirmed by electron microscopy in July 2025, per the post; the linked paper reports a cryo-EM structure of native honey bee vitellogenin, and the breeding-time claim comes from early trials.",
      "unknown": "The breeding-time benefit (months to weeks) is preliminary and not cited to a publication, and implications for other egg-laying species are possibilities, not established findings.",
      "ai_role": "AlphaFold 3 predicted the vitellogenin structures and helped study how it binds other molecules inside the bee.",
      "evidence_urls": [
        "https://deepmind.google/blog/breeding-healthier-and-stronger-honeybees/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-025-58575-y&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-11-popeve",
    "title": "popEVE: diagnosing rare developmental disorders",
    "summary": "Applied to about 30,000 patients with severe undiagnosed developmental disorders, it led to a diagnosis in about one-third of cases and identified variants in 123 previously unknown genes, 25 of which were independently confirmed by other labs. This is preliminary clinical use; further checks are needed before adoption, and the page gives no detailed metrics.",
    "field": "medicine",
    "ai_technology": "popEVE (variant-effect prediction model)",
    "organization": "Harvard Medical School",
    "source_url": "https://hms.harvard.edu/news/new-artificial-intelligence-model-could-speed-rare-disease-diagnosis",
    "source_type": "press release",
    "publication_date": "2025-11-24",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41588-025-02400-1",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "popEVE was tested on documented variants and case studies, and applied to about 30,000 undiagnosed patients, leading to a diagnosis in about one-third; 123 novel gene candidates were identified, and 25 were independently confirmed by other labs.",
      "unknown": "The press article says popEVE will need further verification for safety and accuracy before wide clinical adoption, and gives no accuracy metrics; it is not yet in routine clinical use.",
      "ai_role": "A deep generative model extending EVE with a protein language model, producing a calibrated disease-likelihood score for each variant.",
      "evidence_urls": [
        "https://hms.harvard.edu/news/new-artificial-intelligence-model-could-speed-rare-disease-diagnosis",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41588-025-02400-1&resultType=core&format=json"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2025-11-gpt5-nesterov",
    "title": "GPT-5 helps solve a 40-year-old problem about Nesterov's accelerated gradient",
    "summary": "Ernest Ryu (UCLA) proved why Nesterov's accelerated gradient method, introduced in 1983, remains stable despite the extra momentum that makes it faster. GPT-5's turning point was a rewrite of the equations: not correct as given, but with a structure Ryu then developed and proved himself, over about twelve hours across three days and after a dozen approaches. OpenAI's account says the model created no new mathematical tools and often produced plausible arguments that failed verification.",
    "field": "mathematics",
    "ai_technology": "GPT-5",
    "organization": "OpenAI; and University of California, Los Angeles",
    "source_url": "https://openai.com/index/gpt-5-mathematical-discovery/",
    "source_type": "blog",
    "publication_date": "2025-11-24",
    "added_date": "2026-11-12",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2510.23513",
    "relevance": {
      "impact": 3,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Not reported",
      "unknown": "OpenAI states GPT-5 often produced plausible but flawed arguments; its restructuring suggestion was not correct as written. The proof was developed and written by Ryu. Whether peer review will confirm it is not reported.",
      "ai_role": "Per OpenAI, GPT-5 suggested approaches and restructured equations; Ryu checked each idea and wrote the final proof. About twelve hours over three days and nearly a dozen approaches. GPT-5 is not a co-author.",
      "evidence_urls": [
        "https://openai.com/index/gpt-5-mathematical-discovery/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2025-11-gpt5-esperimenti",
    "title": "Early experiments in accelerating science with GPT-5",
    "summary": "A collection of case studies with universities and national labs in mathematics, physics, biology, computer science, astronomy, and materials. Examples: it helped complete a proof of an Erdős problem with Sawhney and Sellke, reconstructed the SL(2,R) symmetry of the Kerr wave equation, and proposed an immunology mechanism later confirmed by Unutmaz's lab. Reported by OpenAI; experts supervised the work.",
    "field": "multidisciplinary",
    "ai_technology": "GPT-5 and GPT-5 Pro",
    "organization": "OpenAI",
    "source_url": "https://openai.com/index/accelerating-science-gpt-5/",
    "source_type": "blog",
    "publication_date": "2025-11-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2511.16072",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "The OpenAI page lists case studies, for example a mannose rescue experiment in T cells whose results matched GPT-5 predictions, and Erdos Problem 848, where authors turned a GPT-5 density estimate into a full proof.",
      "unknown": "OpenAI says the cases are curated illustrations, not a systematic sample. GPT-5 can invent citations, mechanisms or proofs, and expert oversight is essential. One case reproduced an earlier published proof without citing it.",
      "ai_role": "GPT-5 suggested ideas, mechanisms, literature links and experiments; scientists set questions, verified results and ran lab work. OpenAI says the model is not autonomous.",
      "evidence_urls": [
        "https://openai.com/index/accelerating-science-gpt-5/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-11-kosmos",
    "title": "Kosmos: an AI scientist for autonomous discovery",
    "summary": "Each run reads about 1,500 papers and executes 42,000 lines of analysis code; outside scientists judged 79.4% of its conclusions accurate. Of seven discoveries, three reproduce results already obtained by humans and four are new. Limitations include spurious correlations, a cost of about $200 per run, and a six-month work gain estimated by beta users.",
    "field": "multidisciplinary",
    "ai_technology": "Kosmos (agents with a structured world model)",
    "organization": "Edison Scientific",
    "source_url": "https://advances.edisonscientific.com/research/announcing-kosmos/",
    "source_type": "blog",
    "publication_date": "2025-11-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2511.02824",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Independent scientists judged 79.4% of Kosmos's conclusions accurate; three discoveries reproduced findings from preprints or unpublished manuscripts that Kosmos did not access at runtime; the four new discoveries are wet-lab validation in progress, with one checked in an independent scRNA-seq dataset.",
      "unknown": "The authors state that Kosmos chases statistically significant but irrelevant findings, that about one in five statements is not accurate, and that the human time-saving estimates are intrinsically suspect.",
      "ai_role": "An AI scientist runs about 12-hour cycles of data analysis and literature search (about 1,500 papers and 42,000 lines of code per run), linked by a structured world model, and generates traceable reports; humans set the objectives.",
      "evidence_urls": [
        "https://advances.edisonscientific.com/research/announcing-kosmos/",
        "https://arxiv.org/abs/2511.02824"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2025-11-alphaevolve-tao",
    "title": "AlphaEvolve on 67 math problems with Terence Tao",
    "summary": "Tested on 67 problems in analysis, combinatorics, geometry, and number theory, AlphaEvolve matched the best known solutions in most cases and improved them in several; in some cases it generalized finite instances into formulas valid for every input. Preprint not yet peer reviewed; source checked was only the team's article.",
    "field": "mathematics",
    "ai_technology": "AlphaEvolve",
    "organization": "Google DeepMind; Terence Tao, and Javier Gomez-Serrano",
    "source_url": "https://arxiv.org/abs/2511.02864",
    "source_type": "paper",
    "publication_date": "2025-11-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The system was run on 67 problems in analysis, combinatorics, geometry and number theory and compared with the best known solutions. It rediscovered the best known solutions in most cases and improved several.",
      "unknown": "The abstract and page state no explicit limitations, and no verification details or failure cases are given.",
      "ai_role": "An LLM-guided evolutionary coding agent proposed, tested and refined algorithmic constructions, sometimes combined with Deep Think and AlphaProof.",
      "evidence_urls": [
        "https://arxiv.org/abs/2511.02864"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-10-deepsomatic",
    "title": "DeepSomatic: a neural network for somatic variants in tumors",
    "summary": "For insertions and deletions, it reaches 90% with Illumina data versus 80% for the best other tool, and over 80% with PacBio versus under 50%. In eight pediatric leukemia samples, it recovered known variants and flagged ten new ones that have not been independently validated. The tool and CASTLE dataset are open.",
    "field": "medicine",
    "ai_technology": "DeepSomatic (convolutional neural network)",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/using-ai-to-identify-genetic-variants-in-tumors-with-deepsomatic/",
    "source_type": "blog",
    "publication_date": "2025-10-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-025-02839-x",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "DeepSomatic was benchmarked against SomaticSniper, MuTect2, Strelka2 and ClairS on the CASTLE cell-line dataset with held-out chromosomes and genomes, across Illumina, PacBio and other platforms.",
      "unknown": "The benchmark uses research cell lines rather than patient tumors, the glioblastoma result rests on one sample, and the 10 new leukemia variants are not stated to be validated.",
      "ai_role": "A convolutional neural network applied to sequencing data converted to images separated germline, somatic and error signals to call tumor variants.",
      "evidence_urls": [
        "https://research.google/blog/using-ai-to-identify-genetic-variants-in-tumors-with-deepsomatic/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41587-025-02839-x&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-10-c2s-scale",
    "title": "C2S-Scale: a model-generated cancer hypothesis confirmed in cells",
    "summary": "The 27-billion-parameter single-cell model, built on Gemma with Yale, predicted that silmitasertib would increase antigen presentation only in an immune-positive context. In human neuroendocrine cell models with low-dose interferon, the increase was about 50%. Validation is in cells only; the authors call this ‘a first step.’",
    "field": "medicine",
    "ai_technology": "C2S-Scale 27B (Gemma)",
    "organization": "Google DeepMind; and Yale",
    "source_url": "https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/",
    "source_type": "blog",
    "publication_date": "2025-10-15",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2025.04.14.648850",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The model's prediction that silmitasertib boosts antigen presentation with low-dose interferon was tested experimentally in human neuroendocrine cell models not seen in training, with a roughly 50% synergistic increase reported.",
      "unknown": "The article calls this an early first step, validation is in cell models only, and clinical use is not shown.",
      "ai_role": "The 27B single-cell language model C2S-Scale ran a dual-context virtual screen of over 4,000 drugs and generated the hypothesis.",
      "evidence_urls": [
        "https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/",
        "https://www.biorxiv.org/content/10.1101/2025.04.14.648850"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-enterololin",
    "title": "Enterololin: AI helps clarify an antibiotic and its mechanism",
    "summary": "An AI screen of over 10,000 molecules identified enterololin, and DiffDock predicted its binding to LolCDE within minutes, confirmed by resistant E. coli mutants, RNA-seq, and CRISPR. The mechanism was clarified in about six months versus 18–24 with traditional methods. In mice, in Crohn's-like inflammation models, healing was faster than with vancomycin. Early stage; no human testing.",
    "field": "medicine",
    "ai_technology": "Machine-learning screening and DiffDock",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2025/ai-maps-how-new-antibiotic-targets-gut-bacteria-1003",
    "source_type": "press release",
    "publication_date": "2025-10-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41564-025-02142-0",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Enterololin was tested in mouse Crohn's-like inflammation models against vancomycin, with resistant mutant sequencing, RNA-seq and CRISPR knockdown used to clarify its mechanism.",
      "unknown": "The source states it is early-stage development, no human data are reported, and clinical trials are not yet started.",
      "ai_role": "DiffDock predicted binding of enterololin to the LolCDE complex within minutes, guiding lab experiments on the mechanism.",
      "evidence_urls": [
        "https://news.mit.edu/2025/ai-maps-how-new-antibiotic-targets-gut-bacteria-1003",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41564-025-02142-0&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-fluidodinamica-singolarita",
    "title": "New families of unstable singularities in fluid equations",
    "summary": "Physics-informed neural networks trained to near machine precision revealed a new family of unstable singularities (blow-ups) in three fluid-dynamics equations, with a clear pattern between blow-up speed and instability order for the incompressible porous-medium and Boussinesq equations. Work with Brown, NYU, and Stanford: a step toward computer-assisted proofs, not a proof; the Navier–Stokes problem remains open.",
    "field": "mathematics",
    "ai_technology": "Physics-informed neural networks",
    "organization": "Google DeepMind; Brown University, NYU, and Stanford",
    "source_url": "https://deepmind.google/blog/discovering-new-solutions-to-century-old-problems-in-fluid-dynamics/",
    "source_type": "blog",
    "publication_date": "2025-09-18",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2509.14185",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Physics-informed neural networks trained to near machine precision found unstable singularities, and the accuracy was assessed numerically against the governing equations.",
      "unknown": "The source states the initial conditions must be tuned with extreme precision, the scaling formula is empirical, Navier-Stokes remains open, and supplementary information is not yet available.",
      "ai_role": "Neural networks trained on the equation residuals, with a high-precision optimizer, located new families of unstable singularities, with humans supplying the mathematical insight.",
      "evidence_urls": [
        "https://deepmind.google/blog/discovering-new-solutions-to-century-old-problems-in-fluid-dynamics/",
        "https://arxiv.org/abs/2509.14185"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-09-evo-fagi",
    "title": "First bacteriophage genomes designed by a genome language model",
    "summary": "The Arc Institute group used Evo to design genomes resembling that of phage ΦX174: of 285 designs, 16 yielded functional phages (67–392 mutations from the closest natural one; 13 had never-seen mutations). They infect only two E. coli strains; a cocktail overcame resistance in three evolved strains. bioRxiv preprint; peer review is unconfirmed.",
    "field": "biology",
    "ai_technology": "Evo (genome language model)",
    "organization": "Arc Institute",
    "source_url": "https://arcinstitute.org/news/hie-king-first-synthetic-phage",
    "source_type": "press release",
    "publication_date": "2025-09-17",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2025.09.12.675911",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Of 285 Evo-designed genomes, 16 were synthesized and gave functional phages in E. coli, and these were tested for fitness, host range and, for three resistant strains, a phage cocktail.",
      "unknown": "The source states host range was limited to E. coli C and W, all 16 phages were based on the phiX174 template, and clinical use is not shown.",
      "ai_role": "Evo, fine-tuned on Microviridae sequences, generated whole phage genomes resembling phiX174.",
      "evidence_urls": [
        "https://arcinstitute.org/news/hie-king-first-synthetic-phage",
        "https://www.biorxiv.org/content/10.1101/2025.09.12.675911"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-ligo-deep-loop-shaping",
    "title": "Deep Loop Shaping reduces LIGO control noise",
    "summary": "The method, described in Science and tested at the Livingston site, reduced noise 30–100-fold in the observatory's most unstable feedback loop, in collaboration with LIGO, Caltech, and GSSI. The result covers one control loop; extending it to all loops and adding hundreds of events per year are projections.",
    "field": "physics",
    "ai_technology": "Deep Loop Shaping (reinforcement-learning control)",
    "organization": "Google DeepMind; LIGO, Caltech, and GSSI",
    "source_url": "https://deepmind.google/blog/using-ai-to-perceive-the-universe-in-greater-depth/",
    "source_type": "blog",
    "publication_date": "2025-09-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adw1291",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The controller was trained in simulation and then tested on the real LIGO Livingston system, where noise in the most unstable feedback loop fell 30 to 100 times and stayed stable over prolonged periods.",
      "unknown": "The source states gains come from one control loop, applying the method to all loops and extra detections per year are described as potential or projected.",
      "ai_role": "A reinforcement learning controller with frequency-domain rewards replaced linear control design for the mirror feedback loop.",
      "evidence_urls": [
        "https://deepmind.google/blog/using-ai-to-perceive-the-universe-in-greater-depth/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.adw1291&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-additivi-lnmo",
    "title": "New combinations of additives for LNMO batteries found with machine learning",
    "summary": "A model trained on data for 28 additives predicted the performance of 125 new combinations and identified formulations that outperformed the initial ones in tests. The study concerns high-voltage LNMO batteries; the Argonne announcement, accessed via Newswise, does not quantify the improvement or describe industrial-scale validation.",
    "field": "materials",
    "ai_technology": "Machine learning on molecular features of additives",
    "organization": "Argonne National Laboratory",
    "source_url": "https://www.newswise.com/doescience/do-batteries-need-medicine/?article_id=834113",
    "source_type": "press release",
    "publication_date": "2025-08-26",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-57961-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Electrolyte additive combinations proposed by a model trained on 28 additives were tested in Gr||LNMO cells, and the ML-suggested mixtures outperformed the original set.",
      "unknown": "The source states the model used only 28 additives, gives no quantitative performance figures or top additive names, and does not describe industrial-scale validation.",
      "ai_role": "A machine learning model linked additive descriptors to battery performance and predicted the performance of 125 new combinations.",
      "evidence_urls": [
        "https://www.newswise.com/doescience/do-batteries-need-medicine/?article_id=834113",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-025-57961-w&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-gpt4b-retro",
    "title": "GPT-4b micro: Yamanaka factor variants with over 50-fold higher expression",
    "summary": "OpenAI and Retro Biosciences used a specialized model to design Yamanaka factor variants that produced over 50-fold higher expression of reprogramming markers in vitro than controls, with results replicated across multiple donors, cell types, and delivery methods, and pluripotency confirmed. Reported by OpenAI; the model is not publicly available.",
    "field": "biology",
    "ai_technology": "GPT-4b micro",
    "organization": "OpenAI; and Retro Biosciences",
    "source_url": "https://openai.com/index/accelerating-life-sciences-research-with-retro-biosciences/",
    "source_type": "blog",
    "publication_date": "2025-08-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Designed Yamanaka factor variants were tested in vitro across multiple donors, cell types and delivery methods, with over 50-fold higher expression of reprogramming markers and pluripotency confirmed relative to controls.",
      "unknown": "The page gives no detailed protocol or data, the model is not publicly available, and the results are reported by OpenAI only.",
      "ai_role": "A specialized GPT-4b micro model designed Yamanaka factor variants.",
      "evidence_urls": [
        "https://openai.com/index/accelerating-life-sciences-research-with-retro-biosciences/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-antibiotici-generativi",
    "title": "Antibiotics designed with generative AI against gonorrhea and MRSA",
    "summary": "Over 36 million possible compounds were screened computationally: candidate NG1 killed resistant Neisseria gonorrhoeae in a dish and in a mouse model, while DN1 cleared a MRSA skin infection in mice. Only 2 of 80 fragment candidates could be synthesized; these are preclinical results.",
    "field": "medicine",
    "ai_technology": "Generative AI for molecule design",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2025/using-generative-ai-researchers-design-compounds-kill-drug-resistant-bacteria-0814",
    "source_type": "press release",
    "publication_date": "2025-08-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2025.07.033",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Generative models screened more than 10^7 chemical fragments in silico against N. gonorrhoeae or S. aureus, plus unconstrained design. Of 24 synthesized compounds, seven showed selective antibacterial activity; two leads reduced bacterial burden in mouse infection models.",
      "unknown": "Mouse results only; no human testing is stated. The paper's abstract was read, not the full text.",
      "ai_role": "Genetic algorithms and variational autoencoders generated fragment-based and de novo compounds; researchers synthesized and tested selected compounds.",
      "evidence_urls": [
        "https://www.cell.com/cell/abstract/S0092-8674(25)00855-4"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Original institutional announcement checked for the specific results, AI contribution and limitations cited in the antibiotic dossier or mathematical follow-up pilot. Full-paper reanalysis, external replication and an exhaustive follow-up search were not performed."
    },
    "relationships": [
      {
        "type": "extends",
        "target_id": "2020-02-halicin",
        "reason": "The source names this earlier screening result as program background for moving to de novo generative design. New compounds and experiments; not a clinical follow-up or replication of the earlier candidate.",
        "source_urls": [
          "https://news.mit.edu/2025/using-generative-ai-researchers-design-compounds-kill-drug-resistant-bacteria-0814"
        ]
      },
      {
        "type": "extends",
        "target_id": "2023-05-abaucina",
        "reason": "The source names this earlier screening result as program background for moving to de novo generative design. New compounds and experiments; not a clinical follow-up or replication of the earlier candidate.",
        "source_urls": [
          "https://news.mit.edu/2025/using-generative-ai-researchers-design-compounds-kill-drug-resistant-bacteria-0814"
        ]
      }
    ]
  },
  {
    "id": "2025-08-bci-linguaggio-interiore",
    "title": "Decoding inner speech with a brain-computer interface",
    "summary": "In four people with severe motor and speech impairments, researchers found imagined-speech signals in the motor cortex and decoded them with machine learning. They also tested filters and a trigger phrase to limit unintended decoding. This is a proof of principle with implanted electrodes: free, rapid inner speech cannot be decoded accurately, and attempted speech remains more accurate.",
    "field": "medicine",
    "ai_technology": "Brain-computer interface and machine learning",
    "organization": "Stanford University; and collaborators",
    "source_url": "https://news.stanford.edu/stories/2025/08/study-inner-speech-decoding-device-patients-paralysis",
    "source_type": "press release",
    "publication_date": "2025-08-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2025.06.015",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Imagined-speech signals were recorded from motor cortex in four participants with implanted electrodes and decoded in real time, with filters and a trigger phrase tested to limit unintended decoding.",
      "unknown": "The source states this is a proof of principle, free inner speech cannot yet be decoded accurately, and attempted speech remains more accurate.",
      "ai_role": "Machine learning decoded the imagined-speech neural signals.",
      "evidence_urls": [
        "https://news.stanford.edu/stories/2025/08/study-inner-speech-decoding-device-patients-paralysis",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.cell.2025.06.015&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-alphaearth",
    "title": "AlphaEarth Foundations: a 10-meter map of Earth",
    "summary": "A model combines optical, radar, 3D laser, and climate-simulation data into one representation for 10-meter squares, using 16 times less storage and with 24% lower average error than the compared models. The Google Earth Engine dataset has over 1.4 trillion embeddings per year and has been tried by more than 50 organizations. Benefits with LLM agents remain prospective.",
    "field": "climate",
    "ai_technology": "AlphaEarth Foundations",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/",
    "source_type": "blog",
    "publication_date": "2025-07-30",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2507.22291",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Compared against traditional methods and other AI mapping systems on land-use identification and surface-property estimation; average error 24% lower than the models tested, storage 16 times lower. Over 50 organizations tested the dataset; no sample sizes given.",
      "unknown": "The page states no limitations; it says the team is still exploring time-based capabilities and that embeddings may be more useful combined with reasoning LLM agents.",
      "ai_role": "The model turns satellite and Earth observation data into compact embeddings; scientists and partners use them to create maps and classify ecosystems.",
      "evidence_urls": [
        "https://deepmind.google/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/",
        "https://arxiv.org/abs/2507.22291"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-virtual-lab",
    "title": "Virtual Lab: AI agents design nanobodies against SARS-CoV-2",
    "summary": "A group of AI agents led by a principal agent designed 92 nanobodies: two bound to spike protein from recent variants in lab tests, and humans wrote about 1% of the conversation. Only 2 of 92 bound; the date is the Nature publication date given on the page.",
    "field": "biology",
    "ai_technology": "Multiple AI agents (Virtual Lab)",
    "organization": "CZ Biohub; and Stanford University",
    "source_url": "https://biohub.org/news/with-no-need-for-sleep-or-food-ai-built-scientists-get-the-job-done-quickly/",
    "source_type": "press release",
    "publication_date": "2025-07-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09442-9",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The Virtual Lab designed 92 nanobodies; experiments (multiplexed ELISA) showed a range of functional binders. Two showed improved binding to JN.1 or KP.3 while keeping strong binding to the ancestral spike protein.",
      "unknown": "The two are called candidates for further investigation. Neutralization and clinical use not reported in the abstract. Full text is paywalled; counts of tested designs beyond 92 and binding values were not read.",
      "ai_role": "An LLM principal investigator agent guided LLM scientist agents through meetings, with human feedback; they built a pipeline using ESM, AlphaFold-Multimer and Rosetta. Humans ran the experiments.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-025-09442-9"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-polimeri-mit",
    "title": "An autonomous platform speeds research on polymer mixtures",
    "summary": "The platform tests up to 700 mixtures a day (96 at a time), guided by a genetic algorithm: hundreds of mixtures outperformed the individual components, and the best retained 73% enzyme activity, 18% more than the best component. Application is limited to enzyme thermal stability, replenishment is still manual; published in Matter.",
    "field": "materials",
    "ai_technology": "Genetic algorithm with robotics",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2025/new-system-dramatically-speeds-polymer-materials-search-0728",
    "source_type": "press release",
    "publication_date": "2025-07-28",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.matt.2025.102336",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An autonomous platform tested up to 700 polymer mixtures a day, and the best mixtures retained 73% enzyme activity, 18% more than the best component, in a retrospective analysis of the results.",
      "unknown": "The source states application is limited to enzyme thermal stability and replenishment is still manual.",
      "ai_role": "A genetic algorithm guided the choice of polymer mixtures tested by the robotic platform.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.matt.2025.102336&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-autobot",
    "title": "AutoBot: robotic optimization of halide perovskites",
    "summary": "A closed loop with robotics and Bayesian optimization explored a four-parameter synthesis space of over 5,000 combinations, and the model made predictions after sampling less than 1%; humidity during deposition was the most influential factor. This is process optimization, not a new material; the page gives no accuracy figure.",
    "field": "materials",
    "ai_technology": "Bayesian optimization with robotics",
    "organization": "Lawrence Berkeley National Laboratory (Molecular Foundry)",
    "source_url": "https://foundry.lbl.gov/2025/07/25/accelerating-experiment-design-through-ai-driven-robot/",
    "source_type": "press release",
    "publication_date": "2025-07-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/aenm.202502294",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "A closed loop of robotic synthesis and Bayesian optimization explored over 5,000 combinations, and predictions were validated with manual and in situ photoluminescence measurements.",
      "unknown": "The page gives no model accuracy metrics and no explicit limitations; the wider application to other semiconductors is forward-looking.",
      "ai_role": "Bayesian optimization chose which experiments to run and identified humidity as the most influential factor after sampling under 1% of the space.",
      "evidence_urls": [
        "https://foundry.lbl.gov/2025/07/25/accelerating-experiment-design-through-ai-driven-robot/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-aeneas",
    "title": "Aeneas: restoration, dating, and provenance of Latin inscriptions",
    "summary": "The model restores texts with 73% top-20 accuracy for gaps up to ten characters (58% when length is unknown), attributes provenance with 72% accuracy across 62 Roman provinces, and dates inscriptions within 13 years; it was trained on over 176,000 inscriptions. In a study with 23 historians on 60 inscriptions, results were best when parallels and predictions were used together. Published in Nature.",
    "field": "humanities",
    "ai_technology": "Aeneas",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/discover/blog/aeneas-transforms-how-historians-connect-the-past/",
    "source_type": "blog",
    "publication_date": "2025-07-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09292-5",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Corpus of 176,861 Latin inscriptions. Aeneas: 72% geographical accuracy over 62 provinces; dating within 13 years on average; restoration character error rate 40.5% with known length. 23 historians on 60 inscriptions, solo versus with Aeneas parallels and predictions.",
      "unknown": "The paper notes a small dataset, survival bias, only about 5% of inscriptions with images, and weaker results where data are scarce. Training data include earlier editors' restorations. Other languages are suggested, not tested.",
      "ai_role": "Aeneas restored text, attributed provinces, estimated dates and retrieved parallels; historians evaluated outputs and combined them with their own knowledge.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-025-09292-5"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-imo-oro",
    "title": "Gemini Deep Think reaches gold-medal level at the 2025 IMO",
    "summary": "An advanced version of Gemini Deep Think solved 5 of 6 problems at the 2025 IMO, scoring 35 out of 42 and working in natural language within the 4.5-hour limit. IMO coordinators assessed and certified the score, while clarifying that they do not validate the system.",
    "field": "mathematics",
    "ai_technology": "Gemini Deep Think",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/",
    "source_type": "blog",
    "publication_date": "2025-07-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "An advanced Gemini Deep Think version solved the official 2025 IMO problems within the time limit, and IMO coordinators graded and certified the 35 of 42 score with the same criteria as for students.",
      "unknown": "The IMO did not validate the system or model, and this version is not yet public and was first given to trusted testers.",
      "ai_role": "Gemini Deep Think worked end-to-end in natural language, exploring multiple solution paths in parallel.",
      "evidence_urls": [
        "https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-06-alphagenome",
    "title": "AlphaGenome: predicting the effects of DNA variants",
    "summary": "The model reads sequences up to 1 million base pairs and predicts thousands of gene-regulation properties at single-letter resolution. It outperformed the best external models in 22 of 24 sequence evaluations and matched or outperformed the best in 24 of 26 variant evaluations. Stated limits: distant regulatory elements; not validated for personal genomes or clinical use.",
    "field": "biology",
    "ai_technology": "AlphaGenome",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/",
    "source_type": "blog",
    "publication_date": "2025-06-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2025.06.25.661532",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "AlphaGenome was evaluated on sequence and variant-effect benchmarks against external models, outperforming in 22 of 24 and matching or exceeding in 24 of 26 evaluations, plus a TAL1 case study.",
      "unknown": "The source states it is not validated for personal genomes or clinical use, distant regulatory elements remain a challenge, and cell-specific patterns need improvement.",
      "ai_role": "A model takes up to 1 million DNA base pairs and predicts thousands of genomic tracks and variant effects.",
      "evidence_urls": [
        "https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/",
        "https://www.biorxiv.org/content/10.1101/2025.06.25.661532"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-06-skala",
    "title": "Skala: a deep-learning exchange-correlation functional",
    "summary": "It comes close to the 1 kcal/mol threshold on the W4-17 benchmark (0.85 kcal/mol on the single-reference subset) and is competitive with leading hybrid functionals on GMTKN55 at lower cost. It was trained on about 150,000 high-accuracy energy differences, roughly 100 times the previous dataset size. Accuracy is shown only within the training data's chemical space; it was not yet public at the time.",
    "field": "chemistry",
    "ai_technology": "Skala (neural network for DFT)",
    "organization": "Microsoft Research",
    "source_url": "https://www.microsoft.com/en-us/research/blog/breaking-bonds-breaking-ground-advancing-the-accuracy-of-computational-chemistry-with-deep-learning/",
    "source_type": "blog",
    "publication_date": "2025-06-18",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2506.14665",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Skala was evaluated on the W4-17 atomization benchmark, reaching 0.85 kcal/mol on the single-reference subset, and on GMTKN55, where it was competitive with hybrid functionals.",
      "unknown": "Accuracy is claimed only within the training data chemical space, expansion to broader chemistry is at an early stage, and the functional was not yet public.",
      "ai_role": "A deep-learning exchange-correlation functional learned from about 150,000 high-accuracy energy differences.",
      "evidence_urls": [
        "https://www.microsoft.com/en-us/research/blog/breaking-bonds-breaking-ground-advancing-the-accuracy-of-computational-chemistry-with-deep-learning/",
        "https://arxiv.org/abs/2506.14665"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-06-rotoli-mar-morto",
    "title": "Dead Sea Scrolls redated with AI and radiocarbon",
    "summary": "The Enoch tool dated the writing of 135 manuscripts with about 79% agreement with palaeographers and roughly 30-year precision, following new radiocarbon dating; many scrolls appear older than expected. The 79% measures agreement with palaeographers, not accuracy against known dates; the announcement gives no details on training data.",
    "field": "humanities",
    "ai_technology": "Enoch (palaeographic dating model)",
    "organization": "University of Groningen; Pisa, KU Leuven, and University of Southern Denmark",
    "source_url": "https://www.sdu.dk/en/om-sdu/fakulteterne/naturvidenskab/nyheder-2025/dead-sea-scrolls",
    "source_type": "press release",
    "publication_date": "2025-06-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pone.0323185",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The Enoch dates for 135 manuscripts, following new radiocarbon dating, were compared with palaeographers' judgments, with about 79% agreement.",
      "unknown": "The source states the 79% measures agreement with palaeographers rather than accuracy against known dates, that about one in five datings disagreed, and that training-data details are not given.",
      "ai_role": "Enoch, a writing-style model trained on radiocarbon-dated samples, dated the manuscripts.",
      "evidence_urls": [
        "https://www.sdu.dk/en/om-sdu/fakulteterne/naturvidenskab/nyheder-2025/dead-sea-scrolls",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1371/journal.pone.0323185&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-06-rentosertib",
    "title": "Rentosertib: AI-designed drug reports Phase IIa results in pulmonary fibrosis",
    "summary": "In the GENESIS-IPF study of 71 patients with idiopathic pulmonary fibrosis at 22 centers (placebo versus three doses, 12 weeks), the 60 mg/day group had a mean FVC change of +98.4 mL versus −20.3 mL for placebo. Groups were small, so larger studies are needed; biomarker analysis was exploratory.",
    "field": "medicine",
    "ai_technology": "Pharma.AI (generative AI for targets and molecules)",
    "organization": "Insilico Medicine",
    "source_url": "https://insilico.com/news/tnrecuxsc1-insilico-announces-nature-medicine-publi",
    "source_type": "press release",
    "publication_date": "2025-06-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-025-03743-2",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Phase 2a, multicenter, double-blind, randomized, placebo-controlled, 71 patients for 12 weeks. Primary endpoint (patients with at least one treatment-emergent adverse event) was similar across arms: 70.6% to 83.3%. Treatment-related serious adverse events were low and comparable.",
      "unknown": "Efficacy in larger and longer trials is not established by this phase 2a study.",
      "ai_role": "Generative AI was used to discover the TNIK target and the small-molecule inhibitor, per the paper's title and abstract.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41591-025-03743-2"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2025-05-robin-amd",
    "title": "Robin: ripasudil as a candidate for dry macular degeneration",
    "summary": "FutureHouse's multi-agent system generated hypotheses, experiments, and analyses and identified ripasudil, already used for glaucoma, as a candidate against dry macular degeneration. Lab testing of ten candidates found that a ROCK inhibitor increases phagocytosis by retinal pigment epithelial cells; human researchers carried out the experiments. Announced May 20, 2025, and published in Nature May 19, 2026. Preclinical data.",
    "field": "medicine",
    "ai_technology": "Robin (Crow, Falcon, and Finch agents)",
    "organization": "FutureHouse",
    "source_url": "https://www.futurehouse.org/research/demonstrating-end-to-end-scientific-discovery-with-robin-a-multi-agent-system",
    "source_type": "blog",
    "publication_date": "2025-05-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2505.13400",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Human researchers lab-tested ten candidates proposed by Robin, and a ROCK inhibitor was found to increase phagocytosis by retinal pigment epithelial cells.",
      "unknown": "The source calls ripasudil a candidate, not a proven treatment, and the mechanism proposed by Robin is not stated to have been tested.",
      "ai_role": "A multi-agent system generated hypotheses, chose experiments, analyzed data and drafted main figures.",
      "evidence_urls": [
        "https://www.futurehouse.org/research/demonstrating-end-to-end-scientific-discovery-with-robin-a-multi-agent-system",
        "https://arxiv.org/abs/2505.13400"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-microsoft-discovery-refrigerante",
    "title": "Microsoft Discovery: a PFAS-free refrigerant in about 200 hours",
    "summary": "Microsoft researchers used AI models and high-performance simulations to identify a prototype PFAS-free immersion coolant for data centers in about 200 hours, then synthesized it in under four months; initial tests of key properties match predictions. Microsoft describes it as an experiment still under analysis.",
    "field": "materials",
    "ai_technology": "Microsoft Discovery (agentic orchestration and high-performance computing)",
    "organization": "Microsoft",
    "source_url": "https://azure.microsoft.com/en-us/blog/transforming-rd-with-agentic-ai-introducing-microsoft-discovery/",
    "source_type": "blog",
    "publication_date": "2025-05-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "A prototype PFAS-free coolant was synthesized, and some of its primary properties were tested and matched the AI predictions.",
      "unknown": "Microsoft calls it only an experiment still under analysis, and only some properties have been tested.",
      "ai_role": "AI models and HPC simulations in Microsoft Discovery identified the candidate in about 200 hours.",
      "evidence_urls": [
        "https://azure.microsoft.com/en-us/blog/transforming-rd-with-agentic-ai-introducing-microsoft-discovery/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-alphaevolve",
    "title": "AlphaEvolve: an agent for designing algorithms",
    "summary": "An evolutionary agent based on Gemini and automated evaluators. It recovered an average of 0.7% of Google's computing resources, sped up a Gemini training kernel by 23%, and found an algorithm using 48 multiplications for 4×4 complex matrices. Across more than 50 open math problems, it matched the state of the art in about 75% and improved it in about 20%.",
    "field": "mathematics",
    "ai_technology": "AlphaEvolve (Gemini Flash and Pro)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/",
    "source_type": "blog",
    "publication_date": "2025-05-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2506.13131",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Results were checked by automated evaluators, and the agent was applied to over 50 open math problems and deployed tasks, matching state of the art in about 75% and improving it in about 20%.",
      "unknown": "The source states the approach applies only to problems whose solutions can be checked automatically, and broader access is not yet available.",
      "ai_role": "An ensemble of Gemini models proposed code in an evolutionary loop with automated evaluators.",
      "evidence_urls": [
        "https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/",
        "https://arxiv.org/abs/2506.13131"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-04-aardvark",
    "title": "Aardvark Weather: end-to-end forecasting without a physical model",
    "summary": "A single AI model replaces the entire forecasting chain: it starts from satellite and station observations and produces global and local forecasts in minutes on a desktop using about one-tenth of the data; the page says it beats the US GFS on many variables. The announcement gives no metrics; it is not operational and depends on ERA5. Published in Nature.",
    "field": "climate",
    "ai_technology": "Aardvark Weather",
    "organization": "University of Cambridge; Alan Turing Institute, Microsoft Research, and ECMWF",
    "source_url": "https://www.eng.cam.ac.uk/node/7921",
    "source_type": "press release",
    "publication_date": "2025-04-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-08897-0",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Forecasts were compared with the operational NWP baseline (GFS) and with NWS and an end-to-end system; the Nature paper reports the global forecasts beat the baseline for several variables and lead times.",
      "unknown": "The source states it is not operational, depends on ERA5, and that hurricane, air quality and other applications are proposed rather than tested.",
      "ai_role": "A single model ingested satellite and station observations and produced global and local forecasts.",
      "evidence_urls": [
        "https://www.eng.cam.ac.uk/node/7921",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-08897-0&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-04-astar-solare",
    "title": "Seven organic-solar-cell molecules designed with generative models",
    "summary": "Generative and predictive models, refined with experts and experiments, suggested new non-fullerene acceptors for organic solar cells. Seven molecules were synthesized and tested, with efficiencies close to predictions; the news item gives no efficiency values or advantage over the best existing materials.",
    "field": "materials",
    "ai_technology": "Generative and predictive molecule models",
    "organization": "A*STAR; National University of Singapore, and Elix",
    "source_url": "https://research.a-star.edu.sg/articles/highlights/man-machine-team-up-boosts-solar-materials-search/",
    "source_type": "press release",
    "publication_date": "2025-04-10",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acsenergylett.4c02086",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Seven generated non-fullerene acceptor molecules were synthesized and tested, with power conversion efficiencies close to the model predictions.",
      "unknown": "The source gives no efficiency values, no test methods and no advantage over the best existing materials.",
      "ai_role": "A generative model and a predictive model, refined with experiments, proposed the molecules.",
      "evidence_urls": [
        "https://research.a-star.edu.sg/articles/highlights/man-machine-team-up-boosts-solar-materials-search/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-04-microns",
    "title": "MICrONS: a map of one cubic millimeter of mouse visual cortex",
    "summary": "A map with over half a billion connections and about 75,000 reconstructed neurons from roughly 28,000 sections, produced over nine years by more than 150 researchers. AI traced every neuron outline, and the result requires human checking, much of which is already done. This is about one-thousandth of a mouse connectome, and cell counts vary within the text.",
    "field": "biology",
    "ai_technology": "AI neuron segmentation",
    "organization": "Princeton University; Allen Institute, and Baylor College of Medicine",
    "source_url": "https://engineering.princeton.edu/news/2025/04/09/scientists-map-vision-centers-mammalian-brain",
    "source_type": "press release",
    "publication_date": "2025-04-09",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-08790-w",
    "relevance": {
      "impact": 5,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Not reported",
      "unknown": "The source states the map is about one-thousandth of a mouse connectome, the result requires human proofreading (much already done), and the connectome is not necessarily the neural code.",
      "ai_role": "AI traced the outline of every neuron through the electron microscopy slices.",
      "evidence_urls": [
        "https://engineering.princeton.edu/news/2025/04/09/scientists-map-vision-centers-mammalian-brain",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-08790-w&resultType=core&format=json"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2025-03-euclid-lenti",
    "title": "Euclid: 500 candidate strong gravitational lenses",
    "summary": "The first catalog of 500 candidate galaxy–galaxy lenses, almost all previously unknown, was selected with AI models and then reviewed by citizen scientists and experts. These are candidates, not confirmed lenses; 7,000 by the late-2026 release and about 100,000 by mission end are projections.",
    "field": "physics",
    "ai_technology": "Image-classification models",
    "organization": "ESA; and Euclid Consortium",
    "source_url": "https://www.esa.int/ESA_Multimedia/Images/2025/03/Strong_gravitational_lenses_captured_by_Euclid",
    "source_type": "press release",
    "publication_date": "2025-03-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2503.15324",
    "relevance": {
      "impact": 3,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "AI candidates were inspected by citizen scientists and then vetted and modelled by experts, system by system, to produce 497 candidate lenses in the first catalogue.",
      "unknown": "The page calls these candidates, not confirmed lenses, gives no error or false-positive rates, and the future counts are projections.",
      "ai_role": "Deep learning models performed the initial sweep for candidate lenses.",
      "evidence_urls": [
        "https://www.esa.int/ESA_Multimedia/Images/2025/03/Strong_gravitational_lenses_captured_by_Euclid",
        "https://arxiv.org/abs/2503.15324"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2025-03-ai-scientist-workshop",
    "title": "The AI Scientist-v2: a fully generated paper passes a workshop review",
    "summary": "A paper produced end to end by the system received a mean score of 6.33 at the ICBINB workshop at ICLR 2025, above the average acceptance threshold. Humans only supplied the topic and selected three papers to submit, and the accepted one was withdrawn. Only one of three cleared the threshold; it was a workshop, not the main conference, and there was no meta-review.",
    "field": "other",
    "ai_technology": "The AI Scientist-v2",
    "organization": "Sakana AI",
    "source_url": "https://sakana.ai/ai-scientist-first-publication/",
    "source_type": "blog",
    "publication_date": "2025-03-12",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Three AI-generated papers went through double-blind review at an ICLR 2025 workshop, and one scored 6.33 on average, above the acceptance threshold.",
      "unknown": "The source states it was a workshop rather than the main track, there was no meta-review, and the authors' own review found none of the papers met the main-track bar.",
      "ai_role": "The system generated the papers end to end from a human-supplied topic, and humans chose which three to submit.",
      "evidence_urls": [
        "https://sakana.ai/ai-scientist-first-publication/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2025-02-ecmwf-aifs",
    "title": "ECMWF puts AI forecasts (AIFS) into operation",
    "summary": "The European center operationalized AIFS Single on February 25, 2025, alongside the physical IFS model. ECMWF reports gains of up to 20% on tropical-cyclone tracks and about 1,000 times lower energy use. It is deterministic, at 28 km versus 9 km for IFS.",
    "field": "climate",
    "ai_technology": "AIFS (machine-learning model)",
    "organization": "ECMWF",
    "source_url": "https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational",
    "source_type": "press release",
    "publication_date": "2025-02-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Not reported",
      "unknown": "The page does not describe the verification methods or datasets, the grid is 28 km versus 9 km for IFS, and ensemble forecasting is a future step.",
      "ai_role": "A machine learning model trained on past weather evolution projects the same initial conditions as the IFS model forward every six hours.",
      "evidence_urls": [
        "https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2025-02-corrdiff",
    "title": "CorrDiff: kilometer-scale weather forecasts with generative AI",
    "summary": "A diffusion model takes atmospheric data from 25-kilometer to 2-kilometer scale and also predicts variables absent from the inputs, such as radar reflectivity. The paper appeared the same day in Communications Earth and Environment; the initial model was tuned on Taiwan data with the Central Weather Administration, and Taiwan's disaster-reduction center uses it for alerts. NVIDIA claims 500 times faster and 10,000 times more energy-efficient than high-resolution numerical forecasting on CPU: company figures, with no accuracy comparison on the page.",
    "field": "climate",
    "ai_technology": "CorrDiff (diffusion model)",
    "organization": "NVIDIA",
    "source_url": "https://blogs.nvidia.com/blog/earth-2-ai-high-resolution-forecasts/",
    "source_type": "blog",
    "publication_date": "2025-02-24",
    "added_date": "2026-11-12",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43247-025-02042-5",
    "relevance": {
      "impact": 3,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Not reported",
      "unknown": "The article gives no accuracy metrics or accuracy comparison and does not explicitly list limitations; the speed and energy figures are company claims.",
      "ai_role": "A diffusion model downscales atmospheric data from 25 km to 2 km and predicts variables missing from the input.",
      "evidence_urls": [
        "https://blogs.nvidia.com/blog/earth-2-ai-high-resolution-forecasts/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2025-02-bioemu",
    "title": "BioEmu-1: ensembles of protein structures with deep learning",
    "summary": "An open-source model that generates thousands of protein structures per hour on a GPU and reproduces the equilibrium distribution of a molecular-dynamics simulation using 10,000 to 100,000 times fewer GPU-hours. Predicted folding free energies correlate well with experimental values. The authors call it ‘a first step.’",
    "field": "biology",
    "ai_technology": "BioEmu-1",
    "organization": "Microsoft Research",
    "source_url": "https://www.microsoft.com/en-us/research/blog/exploring-the-structural-changes-driving-protein-function-with-bioemu-1/",
    "source_type": "blog",
    "publication_date": "2025-02-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2024.12.05.626885",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Generated ensembles were compared with a molecular dynamics simulation of protein G using 10,000 to 100,000 times fewer GPU hours, and predicted folding free energies were correlated with experimental values.",
      "unknown": "The authors state they are aware of limitations but the post does not list them, and it is presented as an early step.",
      "ai_role": "A generative deep learning model produces thousands of protein structures per hour on a GPU.",
      "evidence_urls": [
        "https://www.microsoft.com/en-us/research/blog/exploring-the-structural-changes-driving-protein-function-with-bioemu-1/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-02-ai-co-scientist",
    "title": "AI co-scientist: a multi-agent system for generating hypotheses",
    "summary": "A system based on Gemini 2.0 that generates hypotheses and research proposals. In the lab, three drugs it suggested for repurposing against acute myeloid leukemia reduced cancer-cell viability, and all suggested treatments for liver fibrosis showed activity. Expert assessments covered only 11 targets; fibrosis results awaited a Stanford report.",
    "field": "medicine",
    "ai_technology": "AI co-scientist (Gemini 2.0, multi-agent)",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/",
    "source_type": "blog",
    "publication_date": "2025-02-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2502.18864",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Elo auto-evaluation over 203 research goals; 15 expert-curated goals; blinded experts rated 11 goals (preference rank 2.36). In vitro on four AML cell lines: binimetinib, pacritinib, cerivastatin reduced viability; KIRA6 IC50 10 nM in KG-1a versus 180 nM in TK6. Seven combinations tested.",
      "unknown": "The paper calls validation preliminary and in vitro only. KIRA6 was far less effective in MOLM-13 and HL-60. Expert evaluation was small and subjective; Elo is not ground truth. Liver fibrosis and AMR details are summarized only.",
      "ai_role": "A Gemini-based multi-agent system generated, debated and ranked hypotheses; experts selected candidates and prioritized lab experiments.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-026-10644-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-02-evo2",
    "title": "Evo 2: the largest AI model for biology",
    "summary": "Trained on more than 9.3 trillion nucleotides from over 128,000 genomes, it reads sequences up to 1 million nucleotides, flags harmful mutations, and designs genetic elements and genomes as long as those of simple bacteria. In BRCA1 tests it exceeds 90% accuracy in distinguishing benign from potentially pathogenic variants (a result limited to that gene). Preprint; human pathogens were excluded from training.",
    "field": "biology",
    "ai_technology": "Evo 2 (genome language model)",
    "organization": "Arc Institute",
    "source_url": "https://arcinstitute.org/news/evo2",
    "source_type": "press release",
    "publication_date": "2025-02-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2025.02.18.638918",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The model was tested on BRCA1 variants, where it reached over 90% accuracy distinguishing benign from potentially pathogenic mutations, and on identifying genetic changes affecting protein function and fitness.",
      "unknown": "The source reports the BRCA1 result for one gene only and states no other limitations beyond human pathogens being excluded from training.",
      "ai_role": "A genome language model trained on 9.3 trillion nucleotides flags harmful mutations and designs genetic elements.",
      "evidence_urls": [
        "https://arcinstitute.org/news/evo2"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-01-kras-ai-quantistico",
    "title": "Two molecules against KRAS from a hybrid AI and quantum process",
    "summary": "A process combining quantum and classical computing with Chemistry42 selected 15 candidates; two showed activity against mutated KRAS variants in living cells. Published in Nature Biotechnology, this is a proof of principle, with no significant quantum advantage or demonstrated superiority over compounds obtained by classical methods.",
    "field": "medicine",
    "ai_technology": "Chemistry42 and hybrid quantum-classical models",
    "organization": "University of Toronto; and Insilico Medicine",
    "source_url": "https://www.utoronto.ca/news/ai-quantum-computing-used-target-undruggable-cancer-protein",
    "source_type": "press release",
    "publication_date": "2025-01-27",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-024-02526-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Fifteen candidates from the hybrid process were tested in the lab, and two strongly targeted mutated KRAS variants in living cells.",
      "unknown": "The source states it is a proof-of-principle study with no significant quantum advantage and no demonstrated superiority over classically designed molecules.",
      "ai_role": "Generative AI (Chemistry42) and a quantum-classical process designed molecules, narrowed to 15 candidates.",
      "evidence_urls": [
        "https://www.utoronto.ca/news/ai-quantum-computing-used-target-undruggable-cancer-protein"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-masif-neosuperfici",
    "title": "MaSIF designs binders for protein–drug complexes",
    "summary": "MaSIF designed proteins that recognize three complexes containing progesterone, venetoclax, or actinonin, then validated them in the lab. A venetoclax-inducible system activated the antitumor activity of CAR T cells in vitro; this is a proof of concept without clinical validation.",
    "field": "biology",
    "ai_technology": "MaSIF (deep learning on molecular surfaces)",
    "organization": "EPFL; and AITHYRA",
    "source_url": "https://actu.epfl.ch/news/a-deep-learning-pipeline-for-controlling-protein-3/",
    "source_type": "press release",
    "publication_date": "2025-01-18",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08435-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "MaSIF-designed binders were validated in the lab against three drug-bound complexes, and a venetoclax-inducible system switched on CAR-T tumor-killing activity in vitro.",
      "unknown": "The source states the CAR-T work is in vitro only, with no clinical validation, and binding sites are hard to predict.",
      "ai_role": "MaSIF generated surface fingerprints and searched for complementary protein surfaces to design binders.",
      "evidence_urls": [
        "https://actu.epfl.ch/news/a-deep-learning-pipeline-for-controlling-protein-3/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-mattergen",
    "title": "MatterGen: materials design with a generative model",
    "summary": "A diffusion model trained on 608,000 stable materials. Compound TaCr2O6, generated with a constraint of 200 GPa bulk modulus, was synthesized and measured at 169 GPa (relative error under 20%), but with Ta–Cr disorder differing from the proposed structure. One experimental example.",
    "field": "materials",
    "ai_technology": "MatterGen (diffusion model)",
    "organization": "Microsoft Research",
    "source_url": "https://www.microsoft.com/en-us/research/blog/mattergen-a-new-paradigm-of-materials-design-with-generative-ai/",
    "source_type": "blog",
    "publication_date": "2025-01-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-08628-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The generated TaCr2O6 structure was synthesized and its bulk modulus measured at 169 GPa against the 200 GPa target, with relative error under 20%.",
      "unknown": "The source reports one experimental example, with Ta-Cr disorder differing from the proposed structure, and cites no separate publication for this compound.",
      "ai_role": "MatterGen, a diffusion model conditioned on a 200 GPa bulk modulus, generated the structure.",
      "evidence_urls": [
        "https://www.microsoft.com/en-us/research/blog/mattergen-a-new-paradigm-of-materials-design-with-generative-ai/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-antiveleno-proteine",
    "title": "Computer-designed proteins neutralize snake toxins",
    "summary": "Small, stable proteins designed with AI bind and neutralize three-finger toxins from cobras and mambas in the lab; in mice, they yield 80–100% survival against lethal neurotoxin doses, depending on the toxin. Laboratory and animal studies only; traditional antivenoms remain central.",
    "field": "medicine",
    "ai_technology": "AI protein design (RFdiffusion)",
    "organization": "University of Washington Institute for Protein Design",
    "source_url": "https://www.ipd.uw.edu/2025/01/neutralizing-deadly-snake-toxins/",
    "source_type": "press release",
    "publication_date": "2025-01-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08393-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Lab tests showed the designed proteins neutralized multiple subtypes of three-finger toxins; in mice they gave complete protection from lethal neurotoxin exposure, with 80-100% survival depending on the toxin and antitoxin.",
      "unknown": "The page does not mention human trials, toxin types other than elapid three-finger toxins, and traditional antivenoms remain central to treatment.",
      "ai_role": "Designed small, stable proteins on the computer that bind key regions of the toxins.",
      "evidence_urls": [
        "https://www.ipd.uw.edu/2025/01/neutralizing-deadly-snake-toxins/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-praim-mammografia",
    "title": "PRAIM: AI finds more breast cancers in German screening",
    "summary": "In a prospective study of over 460,000 women at twelve German sites, AI-assisted reading detected 6.7 cancers per thousand exams versus 5.7 without AI. Recall rates for further assessment were 37.4 versus 38.3 per thousand. This observational study does not show a survival benefit, and estimates of workload reduction with automated reading come from simulations.",
    "field": "medicine",
    "ai_technology": "AI support for mammogram reading (Vara)",
    "organization": "Universität zu Lübeck; UKSH, and Vara",
    "source_url": "https://www.uni-luebeck.de/aktuelles/pressemitteilung/artikel/ki-verbessert-brustkrebserkennung.html",
    "source_type": "press release",
    "publication_date": "2025-01-07",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-024-03408-6",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "A prospective study of over 460,000 women at twelve German sites (2021-2023) compared AI-assisted reading with standard double reading by radiologists: 6.7 vs 5.7 cancers per 1,000 exams.",
      "unknown": "No survival benefit is shown; workload and skipped-review figures come from simulations, and the effect on prognosis is still to be studied.",
      "ai_role": "AI read about half of the mammograms, marking cases as normal or suspicious, with the rest read by double reading.",
      "evidence_urls": [
        "https://www.uni-luebeck.de/aktuelles/pressemitteilung/artikel/ki-verbessert-brustkrebserkennung.html"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2024-12-gencast",
    "title": "GenCast: ensemble weather forecasts with a diffusion model",
    "summary": "A 0.25° model that outperformed ECMWF's ENS ensemble on 97.2% of 1,320 variable-and-lead-time combinations (99.8% beyond 36 hours) and produces a 15-day ensemble in about 8 minutes on a TPU. Physical models remain essential because they provide training data and initial conditions.",
    "field": "climate",
    "ai_technology": "GenCast (diffusion model)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/gencast-predicts-weather-and-the-risks-of-extreme-conditions-with-sota-accuracy/",
    "source_type": "blog",
    "publication_date": "2024-12-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08252-9",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "Forecasts were compared with ECMWF ENS on held-out 2019 data (trained up to 2018) across 1,320 variable and lead-time combinations; GenCast was more accurate on 97.2% of targets.",
      "unknown": "Forecasts are especially uncertain beyond a few days, the wind-power analysis is only a proof of principle, and physics-based models remain needed for training data and initial conditions.",
      "ai_role": "A diffusion model generates an ensemble of 50 or more 15-day weather trajectories from the current weather state.",
      "evidence_urls": [
        "https://deepmind.google/blog/gencast-predicts-weather-and-the-risks-of-extreme-conditions-with-sota-accuracy/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-alphaqubit",
    "title": "AlphaQubit: a neural decoder for quantum error correction",
    "summary": "A Transformer-based decoder trained on data from a 49-qubit Sycamore processor made 6% fewer errors than tensor-network methods and 30% fewer than correlated matching in the largest experiments. It is too slow for real-time error correction on fast superconducting processors.",
    "field": "physics",
    "ai_technology": "AlphaQubit (Transformer)",
    "organization": "Google DeepMind; and Google Quantum AI",
    "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/alphaqubit-quantum-error-correction/",
    "source_type": "blog",
    "publication_date": "2024-11-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08148-8",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "On Sycamore data, logical error per round was 2.901e-2 (distance 3) and 2.748e-2 (distance 5), versus 3.028e-2 and 2.915e-2 for a tensor-network decoder. Simulations to distance 11 (241 qubits); correlated matching was 1.25 to 1.4 times worse.",
      "unknown": "Throughput is slower than the 1 microsecond per round target. Training at distance 11 needs more data. Accuracy beyond distance 11 and logical computations remain future work. Larger-distance results are simulated.",
      "ai_role": "A recurrent transformer neural network decodes surface-code syndromes to predict logical errors; it is the method under test.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-024-08148-8"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-ocx24",
    "title": "OCx24: experimental catalysts for evaluating prediction models",
    "summary": "A dataset from Meta FAIR, the University of Toronto, and VSParticle covers over 600 mixed-metal catalysts made from 13 elements and tested under the same conditions, alongside analysis of over 19,000 materials. Hydrogen-evolution predictions flagged platinum-containing alloys as promising, although platinum was absent from training data. Fewer than 25% of produced catalysts matched the target; this was a pilot.",
    "field": "chemistry",
    "ai_technology": "Catalyst prediction models (Open Catalyst)",
    "organization": "Meta AI; University of Toronto, and VSParticle",
    "source_url": "https://ai.meta.com/blog/open-catalyst-simulations-experiments/",
    "source_type": "blog",
    "publication_date": "2024-11-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2411.11783",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Computational predictions were compared with lab results: 441 gas diffusion electrodes tested in zero-gap electrolysis and 572 characterized samples, alongside DFT-verified adsorption energies on about 20,000 materials.",
      "unknown": "Platinum was ranked as a top hydrogen-evolution candidate but was not experimentally tested, and the abstract lists no explicit limitations.",
      "ai_role": "AI-accelerated relaxations (685 million) fed a data-driven Sabatier volcano that flagged platinum-containing alloys as promising.",
      "evidence_urls": [
        "https://arxiv.org/abs/2411.11783"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-evo-science",
    "title": "Evo: a foundation model for DNA, from molecules to genomes",
    "summary": "Published in Science, Evo (Arc Institute, Stanford, and UC Berkeley) designed EvoCas9-1, a CRISPR system sharing only about 73% of its sequence with CRISPR-Cas9 but with comparable activity after eleven designs, and new IS200/IS605 transposons that cut and paste DNA. The team calls this the beginning and conducted an ethics and safety study.",
    "field": "biology",
    "ai_technology": "Evo (DNA foundation model)",
    "organization": "Arc Institute; Stanford, and UC Berkeley",
    "source_url": "https://arcinstitute.org/news/evo-science",
    "source_type": "press release",
    "publication_date": "2024-11-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.ado9336",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Evo was prompted to generate CRISPR systems; EvoCas9-1 was obtained after testing eleven designs, validated in the lab by an experimental biology subteam.",
      "unknown": "The page gives no success rates or error analysis, and an ethics and safety investigation addressed potential risks rather than efficacy.",
      "ai_role": "Evo, a DNA-trained foundation model, designed the protein and RNA components of EvoCas9-1 and new IS200/IS605 transposons.",
      "evidence_urls": [
        "https://arcinstitute.org/news/evo-science"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-flood-hub-globale",
    "title": "Seven-day flood forecasts with reliability similar to earlier five-day forecasts",
    "summary": "The new model, trained on 15,980 stations (up from 5,680), gives seven-day reliability comparable to the previous model's five-day forecast. Flood Hub covers over 100 countries and about 700 million people (up from 460 million). About 250,000 virtual points are unverified and intended for experts only; flash, urban, and coastal floods remain out of scope. The linked March 2024 Nature paper describes the predecessor model, also covered by the catalog's March entry; the November performance and coverage update is reported in this Google Research announcement.",
    "field": "climate",
    "ai_technology": "Hydrological forecasting model",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/a-flood-forecasting-ai-model-trained-and-evaluated-globally/",
    "source_type": "blog",
    "publication_date": "2024-11-11",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07145-1",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Gauged basins were checked with standard cross-validation, and ungauged locations were checked against Sentinel-1 radar inundation images; reliability was compared using Nash-Sutcliffe efficiency distributions across 3,079 basins.",
      "unknown": "Ungauged validation covers only a fraction of locations, flash, urban and coastal floods are out of scope, and virtual-gauge forecasts are unverified.",
      "ai_role": "An LSTM network takes meteorological time series and static catchment attributes, adds input from DeepMind's medium-range forecast model, and outputs a probabilistic streamflow distribution.",
      "evidence_urls": [
        "https://research.google/blog/a-flood-forecasting-ai-model-trained-and-evaluated-globally/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-robot-mobili-chimica",
    "title": "Mobile autonomous robots for exploratory synthetic chemistry",
    "summary": "Mobile robots move samples between a synthesis platform, a UPLC-MS spectrometer, and a benchtop NMR, while a heuristic decision-maker evaluates data and chooses reactions. Demonstrated on structural diversification, supramolecular chemistry, and photochemical synthesis. The AI is heuristic and there is no speed or accuracy metric; the announcement says robots will not have a ‘Eureka’ moment.",
    "field": "chemistry",
    "ai_technology": "Heuristic decision-making with mobile robots",
    "organization": "University of Liverpool",
    "source_url": "https://news.liverpool.ac.uk/?p=123346",
    "source_type": "press release",
    "publication_date": "2024-11-06",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08173-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Mobile robots were demonstrated on three exploratory chemistry problems (structural diversification, supramolecular chemistry, photochemical synthesis); the page reports no reaction counts or accuracy metrics, only that the AI made the same or similar decisions as a human researcher.",
      "unknown": "The page reports no quantitative accuracy metrics, only two robots were used, and the system lacks a 'Eureka' moment.",
      "ai_role": "A heuristic decision-maker processes analytical data and autonomously decides whether a reaction proceeds to the next step.",
      "evidence_urls": [
        "https://news.liverpool.ac.uk/?p=123346"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-10-flywire-connettoma",
    "title": "FlyWire: the complete map of neurons in the Drosophila brain",
    "summary": "Connectome of the adult Drosophila brain: nearly 140,000 neurons, over 15 million synaptic connections, and over 8,000 cell types. The page says deep neural networks segmented images and extracted neurons, but automated segmentation had millions of errors and required about 33 years of human correction across over 50 labs and citizen scientists.",
    "field": "biology",
    "ai_technology": "Deep neural networks for segmentation",
    "organization": "MRC Laboratory of Molecular Biology; Princeton, and FlyWire",
    "source_url": "https://mrclmb.ac.uk/?p=4875",
    "source_type": "press release",
    "publication_date": "2024-10-02",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07558-y",
    "relevance": {
      "impact": 5,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The connectome was checked against an earlier partial map, and connectomes were found to be highly consistent between animals.",
      "unknown": "About 0.5% of neurons have developmental wiring defects, connection signs were predicted rather than measured, and the dataset was not built for a specific medical application.",
      "ai_role": "Deep neural networks segmented electron microscopy images and extracted neurons; the first automated pass had millions of errors and was corrected by humans.",
      "evidence_urls": [
        "https://mrclmb.ac.uk/?p=4875"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-09-txgnn",
    "title": "TxGNN: repurposing existing drugs for rare diseases",
    "summary": "The model covers about 8,000 drugs across 17,080 diseases and is nearly 50% better than competitors at identifying candidates and 35% more accurate on contraindications. It was tested on 1.2 million medical records and on three rare diseases excluded from training. Computational predictions only, with no laboratory or clinical confirmation.",
    "field": "medicine",
    "ai_technology": "TxGNN (graph neural network)",
    "organization": "Harvard Medical School",
    "source_url": "https://hms.harvard.edu/news/researchers-harness-ai-repurpose-existing-drugs-treatment-rare-diseases",
    "source_type": "press release",
    "publication_date": "2024-09-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-024-03233-x",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The model was checked against about 8,000 drugs and 17,080 diseases, validated on 1.2 million medical records, and asked to find drugs for three rare diseases excluded from training; it was compared with leading drug-repurposing models.",
      "unknown": "Identified therapies would require further evaluation for dosing and delivery, and there is no laboratory or clinical confirmation.",
      "ai_role": "Proposed existing drugs for thousands of diseases, predicted contraindications and side effects, and gave a rationale for each prediction.",
      "evidence_urls": [
        "https://hms.harvard.edu/news/researchers-harness-ai-repurpose-existing-drugs-treatment-rare-diseases"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-09-nazca-geoglifi",
    "title": "Nazca: AI helps discover 303 new figurative geoglyphs",
    "summary": "An IBM model flagged 1,309 candidates; about a quarter were ground-checked over roughly six months, finding 303 new figurative geoglyphs among 430 already known and increasing the discovery rate 16-fold. Another 968 candidates remain to be checked; the study had been accepted by PNAS at the time of the announcement.",
    "field": "humanities",
    "ai_technology": "Neural network for aerial images",
    "organization": "Yamagata University; and IBM Research",
    "source_url": "https://www.yamagata-u.ac.jp/en/information/info/20240924/",
    "source_type": "press release",
    "publication_date": "2024-09-24",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2407652121",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Field teams ground-checked about a quarter of the 1,309 AI-flagged candidates on foot over roughly six months, finding 303 new figurative geoglyphs.",
      "unknown": "968 candidates remain unsurveyed, training data was limited, and the study was accepted but not yet published in PNAS at the time of the announcement.",
      "ai_role": "An IBM model that works with few training samples analyzed aerial geospatial imagery to flag likely geoglyph areas.",
      "evidence_urls": [
        "https://www.yamagata-u.ac.jp/en/information/info/20240924/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-solventi-batterie-flusso",
    "title": "Solvent mixtures for flow batteries found through an AI–experiment cycle",
    "summary": "PNNL and Argonne identified solvent mixtures that dissolve three times more than a compound proposed for flow batteries, exploring less than 10% of about 2,000 combinations. The model receives experimental data and guides subsequent experiments; the result concerns solubility, without demonstrating performance of a complete battery.",
    "field": "chemistry",
    "ai_technology": "Machine learning in a loop with high-throughput experiments",
    "organization": "Pacific Northwest National Laboratory; and Argonne National Laboratory",
    "source_url": "https://www.pnnl.gov/news-media/ai-guided-experiments-speed-scientific-discovery",
    "source_type": "press release",
    "publication_date": "2024-09-24",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-47070-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Active learning guided solubility measurements of 2,1,3-benzothiadiazole in binary solvents: 218 measurements from 2,101 candidates found 18 mixtures above 6.20 M; best 6.50 M. On a 98-solvent set, Bayesian optimization needed 17 ± 11 trials versus about 50 ± 27 random.",
      "unknown": "Tested one solute, binary solvents only; solubility is not the only property and viscosity, conductivity and stability were not optimized. Model accuracy on 40 test solvents was R2 = 0.63. Real battery performance not reported.",
      "ai_role": "A Gaussian process model with an expected-improvement rule chose which solvents to measure; a robotic platform prepared samples and NMR measured solubility; humans selected features and candidate lists.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41467-024-47070-5"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-alphaproteo",
    "title": "AlphaProteo: AI-designed protein binders",
    "summary": "A system for designing high-affinity protein binders, tested against seven targets, achieved 88% laboratory success on BHRF1 and bindings 3 to 300 times stronger than the best existing methods. It produced no valid binders for TNFα; strong binding is only a first step.",
    "field": "biology",
    "ai_technology": "AlphaProteo",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/alphaproteo-generates-novel-proteins-for-biology-and-health-research/",
    "source_type": "blog",
    "publication_date": "2024-09-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2409.08022",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Binders were designed against seven target proteins and tested in the lab, with 3 to 300 times stronger binding and higher experimental success rates than the best existing methods.",
      "unknown": "Per-target success rates are not given on the abstract page and no limitations are listed.",
      "ai_role": "A family of machine learning models for protein design that generates the binders.",
      "evidence_urls": [
        "https://arxiv.org/abs/2409.08022"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-chief",
    "title": "CHIEF: detecting cancer and predicting survival from slides",
    "summary": "Trained on 15 million patches and 60,000 slides and tested on more than 19,400 slides from 32 datasets and 24 hospitals: it detects cancer with about 94% accuracy across 15 datasets, predicts mutations in 54 genes with over 70% accuracy, and improves survival prediction by 8% across 17 institutions. Retrospective validation; not trained on rare diseases or premalignant tissue.",
    "field": "medicine",
    "ai_technology": "CHIEF (pathology foundation model)",
    "organization": "Harvard Medical School",
    "source_url": "https://hms.harvard.edu/news/new-artificial-intelligence-tool-cancer",
    "source_type": "press release",
    "publication_date": "2024-09-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07894-z",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Retrospective validation on more than 19,400 slides from 32 datasets at 24 hospitals; cancer detection was about 94% accurate across datasets.",
      "unknown": "Validation beyond these datasets is still needed, rare and non-cancerous conditions and pre-malignant tissue were not tested, and gene prediction accuracy varies by gene and cancer type.",
      "ai_role": "Read digital tumor slides to detect cancer, predict gene mutations and survival, and generate heat maps of tissue features for pathologists.",
      "evidence_urls": [
        "https://hms.harvard.edu/news/new-artificial-intelligence-tool-cancer"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-08-bci-parlato-ucdavis",
    "title": "A speech neuroprosthesis reaches 97.5% accuracy after calibration",
    "summary": "A man with ALS used a brain-computer interface with machine learning to turn attempted speech into text and synthetic voice. After further training, the system maintained 97.5% accuracy; the study reports more than 248 hours of spontaneous conversation over 32 weeks. Result from one participant with an invasive implant; still experimental.",
    "field": "medicine",
    "ai_technology": "Neural speech decoding with machine learning",
    "organization": "UC Davis Health; and BrainGate",
    "source_url": "https://www.ucdavis.edu/news/brain-computer-interface-allows-man-als-speak",
    "source_type": "press release",
    "publication_date": "2024-08-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1056/NEJMoa2314132",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "One participant with ALS in the BrainGate2 trial completed 84 sessions over 32 weeks with over 248 hours of self-paced use; word accuracy was measured per session, with 97.5% maintained after continued data collection.",
      "unknown": "The page does not state the study's limitations, and the result comes from one participant with an invasive implant.",
      "ai_role": "Machine learning decoded neural activity into phonemes and then words in real time, with continuous updates, and software generated the voice.",
      "evidence_urls": [
        "https://www.ucdavis.edu/news/brain-computer-interface-allows-man-als-speak"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-alphaproof-imo",
    "title": "AlphaProof and AlphaGeometry 2: silver-medal level at the Math Olympiad",
    "summary": "The two systems solved 4 of 6 problems at IMO 2024, scoring 28 of 42 points, one point below the gold threshold. Problems were manually translated into formal language, and some required up to three days of computation instead of the official 4.5 hours.",
    "field": "mathematics",
    "ai_technology": "AlphaProof (Lean) and AlphaGeometry 2",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/ai-solves-imo-problems-at-silver-medal-level/",
    "source_type": "blog",
    "publication_date": "2024-07-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09833-y",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The systems solved official IMO 2024 problems, graded under IMO rules by Prof. Timothy Gowers and Dr Joseph Myers, reaching 28 of 42 points; problems were manually translated into formal language first.",
      "unknown": "Two combinatorics problems remained unsolved, translation was manual, and some solutions took up to three days.",
      "ai_role": "AlphaProof solved three problems with formal proofs, and AlphaGeometry 2 proved the geometry problem within 19 seconds of formalization.",
      "evidence_urls": [
        "https://deepmind.google/blog/ai-solves-imo-problems-at-silver-medal-level/"
      ],
      "test_type": "prospective or independent check"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Original institutional announcement checked for the specific results, AI contribution and limitations cited in the antibiotic dossier or mathematical follow-up pilot. Full-paper reanalysis, external replication and an exhaustive follow-up search were not performed."
    },
    "relationships": [
      {
        "type": "extends",
        "target_id": "2024-01-alphageometry",
        "reason": "AlphaGeometry 2 is explicitly described as an improved AlphaGeometry version; this link concerns the geometry component, not attribution of all four IMO solutions.",
        "source_urls": [
          "https://deepmind.google/blog/ai-solves-imo-problems-at-silver-medal-level/"
        ]
      }
    ],
    "follow_up": [
      {
        "id": "alphaproof-nature-methodology",
        "kind": "publication",
        "date": "2025-11-12",
        "date_precision": "day",
        "note": "The updated original announcement dates publication of the AlphaProof methodology in Nature to 12 November 2025. This journal transition is not an independent replication and does not apply to every component of the combined system.",
        "source_urls": [
          "https://deepmind.google/blog/ai-solves-imo-problems-at-silver-medal-level/"
        ],
        "recorded_date": "2026-11-16"
      }
    ],
    "follow_up_review": {
      "date": "2026-11-16",
      "scope": "Bounded check of the original announcements cited by the events recorded here, including explicit links to later program work. This is not an exhaustive literature search or ongoing monitoring commitment."
    }
  },
  {
    "id": "2024-07-neuralgcm",
    "title": "NeuralGCM: a hybrid physics-and-learning model for weather and climate",
    "summary": "At 1.4° resolution, the model outperformed ECMWF-ENS in 95% of cases for forecasts from 2 to 15 days; at 2.8°, its mean 1980–2020 temperature error was 0.25 °C versus 0.75 °C for atmosphere-only models. It is more than 3,500 times faster than X-SHiELD on a TPU. It models only the atmosphere and is not yet a complete climate model.",
    "field": "climate",
    "ai_technology": "NeuralGCM (differentiable dynamical core and neural networks)",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/fast-accurate-climate-modeling-with-neuralgcm/",
    "source_type": "blog",
    "publication_date": "2024-07-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07744-y",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Evaluated with the WeatherBench 2 benchmark against ECMWF-ENS (better in 95% of 2-15 day cases at 1.4 degrees) and against AMIP climate baselines for 1980-2020 temperature (0.25 vs 0.75 degrees C error).",
      "unknown": "The model covers only the atmosphere and is not yet a complete climate model; oceans and the carbon cycle are future goals.",
      "ai_role": "A neural network learns small-scale physics such as cloud formation, replacing hand-built parameterizations, coupled to a physics-based fluid dynamics solver.",
      "evidence_urls": [
        "https://research.google/blog/fast-accurate-climate-modeling-with-neuralgcm/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-06-esm3-esmgfp",
    "title": "ESM3 and esmGFP: a fluorescent protein designed by a language model",
    "summary": "ESM3 designed esmGFP, a fluorescent protein of 229 amino acids with 96 mutations relative to the closest known protein (about 58% similarity). In a second round of 96 designs, some proteins glowed as brightly as natural GFPs. The comparison with ‘more than 500 million years of evolution’ is an analogy-based estimate.",
    "field": "biology",
    "ai_technology": "ESM3",
    "organization": "EvolutionaryScale",
    "source_url": "https://www.evolutionaryscale.ai/blog/esm3-release",
    "source_type": "blog",
    "publication_date": "2024-06-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2024.07.01.600583",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In two rounds of 96 designs, lab measurements of fluorescence found some proteins as bright as natural GFPs; esmGFP differs by 96 mutations and is about 58% similar to the closest natural protein.",
      "unknown": "The page lists no formal limitations; the designed B8 was about 50 times dimmer and slower to mature than natural GFPs.",
      "ai_role": "ESM3 generated new GFP candidates prompted with structural residues from natural GFP.",
      "evidence_urls": [
        "https://www.evolutionaryscale.ai/blog/esm3-release"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-elefanti-nomi",
    "title": "African elephants address one another with calls resembling names",
    "summary": "About 470 calls from 101 callers to 117 recipients in Kenya: a model trained only on acoustic structure identifies the intended recipient, and playback experiments show elephants respond more to calls addressed to them. The model confirms a previously suspected pattern; more data are needed to isolate the ‘names’.",
    "field": "biology",
    "ai_technology": "Acoustic classification model",
    "organization": "Colorado State University; Save the Elephants and ElephantVoices",
    "source_url": "https://newsmediarelations.colostate.edu/2024/06/10/elephants-have-names-for-each-other-like-people-do-new-study-shows/",
    "source_type": "press release",
    "publication_date": "2024-06-10",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41559-024-02420-w",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "About 470 calls from 101 callers to 117 receivers were analyzed, and playback experiments showed elephants respond more to calls addressed to them.",
      "unknown": "More data are needed to isolate the names, it is unknown whether elephants name other things, and naming was not universal.",
      "ai_role": "A model trained only on acoustic structure identified the intended recipient of each call, supporting the name-like finding.",
      "evidence_urls": [
        "https://newsmediarelations.colostate.edu/2024/06/10/elephants-have-names-for-each-other-like-people-do-new-study-shows/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-ampsphere",
    "title": "AMPSphere: antimicrobial peptides from the global microbiome",
    "summary": "Machine learning identified 863,498 candidate sequences in microbial genomes and metagenomes. Of 100 synthesized peptides, 63 inhibited at least one of the eleven strains tested; some showed activity in animal models. The catalog contains candidates, not hundreds of thousands of clinically validated antibiotics.",
    "field": "medicine",
    "ai_technology": "Machine learning for antimicrobial peptides",
    "organization": "University of Pennsylvania; and collaborators",
    "source_url": "https://www.pennmedicine.org/news/penn-expert-uses-ai-to-uncover-potential-antibiotics-in-microbial-dark-matter",
    "source_type": "press release",
    "publication_date": "2024-06-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2024.05.013",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "100 predicted peptides were synthesized and tested in vitro and in vivo against drug-resistant pathogens and human gut commensals; 63 inhibited pathogens.",
      "unknown": "The abstract states no limitations and does not name the strains or animal models used.",
      "ai_role": "Machine learning predicted antimicrobial peptides across 63,410 metagenomes and 87,920 genomes, producing a catalog of 863,498 non-redundant candidates.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.cell.2024.05.013&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-aurora",
    "title": "Aurora: a foundation model for the atmosphere",
    "summary": "A 1.3-billion-parameter model for high-resolution weather and air-pollution forecasting: it matches or exceeds CAMS on 74% of targets and matches or beats GraphCast on 94%. The roughly 5,000-fold speedup over the IFS system is an estimate; GraphCast remains slightly better up to five days.",
    "field": "climate",
    "ai_technology": "Aurora (foundation model)",
    "organization": "Microsoft Research",
    "source_url": "https://www.microsoft.com/en-us/research/blog/introducing-aurora-the-first-large-scale-foundation-model-of-the-atmosphere/",
    "source_type": "blog",
    "publication_date": "2024-06-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2405.13063",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "Forecasts were compared with CAMS for air pollution (Jun-Nov 2022, matched or beat 74% of targets) and with GraphCast (matched or beat 94% of targets) over 2022 evaluation periods.",
      "unknown": "GraphCast was slightly better on specific humidity up to five days, and the page has no dedicated limitations section.",
      "ai_role": "A 1.3-billion-parameter foundation model, pretrained on over a million hours of simulations, forecasts weather and air pollution at 0.1 degrees.",
      "evidence_urls": [
        "https://www.microsoft.com/en-us/research/blog/introducing-aurora-the-first-large-scale-foundation-model-of-the-atmosphere/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-05-alphafold-3",
    "title": "AlphaFold 3: structure and interactions of life’s molecules",
    "summary": "It predicts joint structures of proteins, nucleic acids, small molecules and ions. DeepMind and Isomorphic Labs report at least 50% greater accuracy than existing methods for protein interactions with other molecules. Free for non-commercial research through the AlphaFold Server.",
    "field": "biology",
    "ai_technology": "AlphaFold 3",
    "organization": "Google DeepMind; and Isomorphic Labs",
    "source_url": "https://blog.google/innovation-and-ai/products/google-deepmind-isomorphic-alphafold-3-ai-model/",
    "source_type": "blog",
    "publication_date": "2024-05-08",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07487-w",
    "relevance": {
      "impact": 5,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Higher accuracy than docking tools for protein–ligand interactions, than nucleic-acid-specific predictors for protein–nucleic acid interactions, and than AlphaFold-Multimer v2.3 for antibody–antigen prediction.",
      "unknown": "Accuracy on new experimental targets beyond the benchmarks is not described in the abstract.",
      "ai_role": "A diffusion-based architecture predicting joint structures of proteins, nucleic acids, small molecules, ions and modified residues.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-024-07487-w"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-05-chemcrow",
    "title": "ChemCrow: a language-model agent with chemistry tools",
    "summary": "An agent with 18 tools and GPT-4 planned and carried out the synthesis of DEET and three thiourea organocatalysts on a cloud platform, and guided discovery of a chromophore absorbing at 336 nm against a 369 nm target; evaluated on 14 cases by four expert chemists. Hallucinations, limited tasks and dual-use risks; 12 of 18 tools were released.",
    "field": "chemistry",
    "ai_technology": "ChemCrow (GPT-4 with tools)",
    "organization": "ChemCrow group (Nature Machine Intelligence)",
    "source_url": "https://www.nature.com/articles/s42256-024-00832-8",
    "source_type": "paper",
    "publication_date": "2024-05-08",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Expert and LLM judgments were used to evaluate the agent, which planned and executed syntheses of an insect repellent and three organocatalysts and guided discovery of a chromophore.",
      "unknown": "The abstract says LLMs struggle with chemistry and lack external knowledge, and GPT-4 as evaluator could not distinguish clearly wrong completions from ChemCrow's performance.",
      "ai_role": "An LLM agent with 18 expert-designed chemistry tools planned and executed syntheses and guided molecular discovery; the page does not name the underlying model.",
      "evidence_urls": [
        "https://arxiv.org/abs/2304.05376"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-capodogli-alfabeto",
    "title": "Sperm whales: a possible ‘phonetic alphabet’ in their calls",
    "summary": "Analysis of nearly 9,000 codas from the eastern Caribbean clan, collected over more than ten years, proposes a combinatorial structure involving rhythm, tempo, rubato and ornamentation. Published in Nature Communications; communicative function remains unknown, data cover one clan, and the page does not specify machine-learning methods.",
    "field": "biology",
    "ai_technology": "Machine-learning analysis of whale calls",
    "organization": "Project CETI; and MIT CSAIL",
    "source_url": "https://www.projectceti.org/blog-posts/sperm-whale-phonetic-alphabet-proposed-for-the-first-time",
    "source_type": "press release",
    "publication_date": "2024-05-07",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-47221-8",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Nearly 9,000 codas from the eastern Caribbean clan, collected over more than ten years, were analyzed for context-sensitive features (rubato, ornamentation) and categorical rhythm and tempo.",
      "unknown": "The communicative function of many codas is unknown, whether the features encode intent needs future field research, and the duality-of-patterning link is only a possibility.",
      "ai_role": "Not reported; the page does not specify what machine learning did in the analysis.",
      "evidence_urls": [
        "https://www.projectceti.org/blog-posts/sperm-whale-phonetic-alphabet-proposed-for-the-first-time"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-04-laser-organici",
    "title": "Closed-loop, distributed discovery of organic laser emitters",
    "summary": "In a two-month campaign, more than 1,000 candidates were synthesized and tested at six laboratories across five sites on three continents, identifying at least 21 high-performance candidates as gain materials from a space of more than 150,000 experiments. These are gain candidates, not confirmed lasers; the page gives no laser metrics; source is a consortium press release.",
    "field": "materials",
    "ai_technology": "Closed-loop optimization with autonomous laboratories",
    "organization": "Acceleration Consortium (University of Toronto)",
    "source_url": "https://acceleration.utoronto.ca/news/published-in-the-journal-science-new-paper-delocalized-asynchronous-closed-loop-discovery-of-organic-laser-emitters-demonstrates-the-acs-global-leadership-in-self-driving-labs",
    "source_type": "press release",
    "publication_date": "2024-04-30",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adk9227",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Over 1,000 candidates were synthesized and tested in a two-month distributed campaign across six teams in five labs on three continents, identifying at least 21 high-performance gain candidates.",
      "unknown": "The commercialization speed goal was not met, the candidates are gain materials rather than confirmed lasers, and the page gives no laser metrics.",
      "ai_role": "A closed-loop AI reviewed data, selected molecules to synthesize and test, and predicted the next best experiment; a cloud platform coordinated the labs.",
      "evidence_urls": [
        "https://acceleration.utoronto.ca/news/published-in-the-journal-science-new-paper-delocalized-asynchronous-closed-loop-discovery-of-organic-laser-emitters-demonstrates-the-acs-global-leadership-in-self-driving-labs"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-eth-ppar",
    "title": "Protein-surface-designed molecules activate PPARs",
    "summary": "A generative model designed molecules from the three-dimensional surfaces of PPAR proteins: synthesized in the laboratory, they increased activity and were tested by Roche. They demonstrate the method and are not intended for commercial development; the tests do not establish safety or efficacy in humans.",
    "field": "medicine",
    "ai_technology": "Protein-surface-guided generative AI",
    "organization": "ETH Zurich; and Roche",
    "source_url": "https://ethz.ch/en/news-and-events/eth-news/news/2024/04/ai-designs-new-drugs-based-on-protein-structures.html",
    "source_type": "press release",
    "publication_date": "2024-04-24",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-47613-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "ETH chemists synthesized the AI-designed molecules in the lab, and Roche ran a variety of tests showing they are stable and non-toxic.",
      "unknown": "The article does not state limitations, the molecules were not developed toward market, and the tests do not establish safety or efficacy in humans.",
      "ai_role": "A generative model designed molecules from scratch matched to a protein's three-dimensional surface, without human intervention.",
      "evidence_urls": [
        "https://ethz.ch/en/news-and-events/eth-news/news/2024/04/ai-designs-new-drugs-based-on-protein-structures.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-opencrispr",
    "title": "OpenCRISPR-1: a genome editor designed by language models",
    "summary": "Profluent built an atlas of 5.1 million CRISPR-Cas proteins and generated a Cas9-like editor with 403 mutations relative to SpCas9. Of 48 generated sequences, the best (OpenCRISPR-1) achieved 55.7% on-target editing versus 48.3% for SpCas9, and 0.32% off-target editing versus 6.1%. Human cells in the laboratory only; company data; the page does not indicate peer review.",
    "field": "biology",
    "ai_technology": "Protein language models",
    "organization": "Profluent",
    "source_url": "https://www.profluent.bio/media/editing-the-human-genome-with-ai",
    "source_type": "press release",
    "publication_date": "2024-04-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2024.04.22.590591",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "Of 48 selected generated sequences, the best (OpenCRISPR-1) was tested in human cells: 55.7% on-target editing vs 48.3% for SpCas9, and 0.32% off-target vs 6.1%.",
      "unknown": "The company article has no explicit limitations section and does not state whether a preprint or peer review exists.",
      "ai_role": "A language model pretrained on protein data and steered to CRISPR families using a 5.1-million-protein atlas generated Cas9-like nucleases and compatible sgRNAs.",
      "evidence_urls": [
        "https://www.profluent.bio/media/editing-the-human-genome-with-ai"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-anticorpi-da-zero",
    "title": "Antibodies designed from scratch with RFdiffusion",
    "summary": "Baker’s group showed RFdiffusion can be adapted to generate antibodies: in laboratory tests, designs bound influenza virus and other targets as intended. Proof-of-concept study, first circulated as a preprint in March 2024; experts quoted by Nature call it promising but an early step; no affinity values reported.",
    "field": "medicine",
    "ai_technology": "RFdiffusion",
    "organization": "University of Washington Institute for Protein Design",
    "source_url": "https://www.ipd.uw.edu/2024/04/nature-a-landmark-moment-scientists-use-ai-to-design-antibodies-from-scratch/",
    "source_type": "press release",
    "publication_date": "2024-04-10",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2024.03.14.585103",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Laboratory tests confirmed that the designed antibodies bind influenza virus and other targets as intended; no affinity values are reported.",
      "unknown": "The work is described as a proof-of-concept study, and no specific limitations are listed.",
      "ai_role": "RFdiffusion was tuned to generate antibodies from scratch.",
      "evidence_urls": [
        "https://www.ipd.uw.edu/2024/04/nature-a-landmark-moment-scientists-use-ai-to-design-antibodies-from-scratch/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-semapimod-batteri-dormienti",
    "title": "AI identifies semapimod against dormant bacteria",
    "summary": "A machine-learning system selected semapimod, a known anti-inflammatory drug, for activity against dormant Gram-negative bacteria. MIT reports laboratory activity against E. coli and A. baumannii and an effect that sensitizes these bacteria to other antibiotics; the result is preclinical and does not demonstrate efficacy in patients.",
    "field": "medicine",
    "ai_technology": "Machine learning for compound screening",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2024/mit-scientists-are-using-artificial-intelligence-target-sleeper-bacteria-0408",
    "source_type": "press release",
    "publication_date": "2024-04-08",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.chembiol.2023.10.026",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Lab-only time-lapse microscopy of E. coli treated with semapimod showed activity against stationary-phase E. coli and A. baumannii and sensitization to other antibiotics.",
      "unknown": "The article states no limitations, no animal or patient testing is mentioned, and no drug combinations are named.",
      "ai_role": "AI-based high-throughput screening of known drug compounds identified semapimod for antibiotic activity.",
      "evidence_urls": [
        "https://news.mit.edu/2024/mit-scientists-are-using-artificial-intelligence-target-sleeper-bacteria-0408"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-seeds",
    "title": "SEEDS: weather ensembles generated with a diffusion model",
    "summary": "The model produces 256 members at 2° resolution every three minutes on a TPU and matches or exceeds the operational GEFS ensemble (31 members) on rank histograms, RMSE and CRPS. For the Lisbon heatwave (July 2022), with 16,384 members, none of GEFS’s 31 members was as warm as observed. One extreme case; depends on physics-based forecasts as seeds; published in Science Advances.",
    "field": "climate",
    "ai_technology": "SEEDS (diffusion model)",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/generative-ai-to-quantify-uncertainty-in-weather-forecasting/",
    "source_type": "blog",
    "publication_date": "2024-03-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.adk4489",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The ensemble was compared with the operational GEFS (31 members) using rank histograms, RMSE and CRPS, and with a 16,384-member ensemble for the Lisbon heatwave of July 2022.",
      "unknown": "The heatwave comparisons are single examples, the model depends on physics-based seed forecasts, and the page has no explicit limitations section.",
      "ai_role": "A denoising diffusion model generates large weather-forecast ensembles from one or two operational forecasts.",
      "evidence_urls": [
        "https://research.google/blog/generative-ai-to-quantify-uncertainty-in-weather-forecasting/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-03-synthemol",
    "title": "SyntheMol: generated antibiotic candidates against Acinetobacter baumannii",
    "summary": "The generative model proposed about 25,000 candidates in under nine hours from more than 130,000 synthesis building blocks; of 58 synthesized, six killed a drug-resistant A. baumannii strain and two were not toxic in mice. Mechanism unknown, efficacy in infected animals untested, and four of the six compounds were not water-soluble.",
    "field": "medicine",
    "ai_technology": "SyntheMol (generative model)",
    "organization": "Stanford Medicine; and McMaster University",
    "source_url": "https://med.stanford.edu/news/all-news/2024/03/ai-drug-development.html",
    "source_type": "press release",
    "publication_date": "2024-03-28",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42256-024-00809-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Of 58 synthesized candidates, six killed a drug-resistant A. baumannii strain in the lab, and two of those were not toxic in mice.",
      "unknown": "Four of the six compounds were not water-soluble so were not tested in mice, efficacy in infected animals is untested, and the mechanism is unknown.",
      "ai_role": "A generative model proposed about 25,000 candidate antibiotics and synthesis recipes in under nine hours from over 130,000 building blocks.",
      "evidence_urls": [
        "https://med.stanford.edu/news/all-news/2024/03/ai-drug-development.html"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Original institutional announcement checked for the specific results, AI contribution and limitations cited in the antibiotic dossier or mathematical follow-up pilot. Full-paper reanalysis, external replication and an exhaustive follow-up search were not performed."
    }
  },
  {
    "id": "2024-03-previsione-alluvioni",
    "title": "Reliable flood forecasts even where gauges are missing",
    "summary": "A model trained on 5,680 gauges improved average global forecast reliability from zero to five days: forecasts at days 4–5 were similar to or better than day-zero forecasts from GloFAS v4. Flood Hub covers more than 80 countries. F1 comparison across 2,092 basins for two-year return-period events; flash and urban floods remain to be covered.",
    "field": "climate",
    "ai_technology": "Hydrological forecasting model",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/using-ai-to-expand-global-access-to-reliable-flood-forecasts/",
    "source_type": "blog",
    "publication_date": "2024-03-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07145-1",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "F1 scores were compared with GloFAS v4 for 2-year return-period events across 2,092 watersheds (2014-2023); reliability at 4-5 day lead times was similar to or better than GloFAS at day zero.",
      "unknown": "Coverage must expand to more locations, flash and urban floods are not yet covered, and the gain is average-level rather than uniform.",
      "ai_role": "Two sequential LSTMs with a mixture density output produce probabilistic streamflow forecasts, trained on 5,680 gauges.",
      "evidence_urls": [
        "https://research.google/blog/using-ai-to-expand-global-access-to-reliable-flood-forecasts/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-02-tokamak-instabilita",
    "title": "Preventing tearing-mode instabilities in a tokamak with AI",
    "summary": "A model trained only on past experimental data predicts instabilities up to 300 ms ahead, and an AI controller adjusted plasma shape and beam power to avoid them in a real DIII-D experiment. The authors call it a proof of concept on one machine and one type of instability; paper published in Nature on February 21, 2024.",
    "field": "physics",
    "ai_technology": "Prediction and reinforcement-learning control",
    "organization": "Princeton Plasma Physics Laboratory; and Princeton University",
    "source_url": "https://mae.princeton.edu/about-mae/news/engineers-use-ai-wrangle-fusion-power-grid",
    "source_type": "press release",
    "publication_date": "2024-02-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07024-9",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The reinforcement-learning controller was tested in experiments on the DIII-D tokamak, where it kept the tearing likelihood below a threshold, including under unfavorable conditions.",
      "unknown": "The abstract does not state the prediction horizon and lists no other limitations; tearing instability is hard to forecast.",
      "ai_role": "A deep reinforcement learning controller, trained in a multimodal dynamic model of tearing instability, adjusted plasma operation to keep it stable.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-024-07024-9&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-ercolano-grand-prize",
    "title": "Vesuvius Challenge: the first Herculaneum scroll can be read",
    "summary": "The $700,000 prize went to a submission with 11 additional columns and more than 2,000 characters (about 5% of the first scroll), using models to detect ink and automated segmentation. Papyrologists evaluated 15 anonymized columns and recovered more text from the winning submission. Segmentation remains largely manual and costly.",
    "field": "humanities",
    "ai_technology": "Ink detection with vision models and automated segmentation",
    "organization": "Vesuvius Challenge",
    "source_url": "https://scrollprize.org/grandprize",
    "source_type": "press release",
    "publication_date": "2024-02-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 4,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Papyrologists independently evaluated 15 anonymized columns, and the winning submission yielded the most text for every reviewer, meeting the stated criteria.",
      "unknown": "The comprehensive study is not yet published, only rough draft transcriptions exist, segmentation is largely manual, and about 5% of the first scroll has been read.",
      "ai_role": "Models detected ink in CT scans (a TimeSformer-based model was strongest), and an automated tool produced the segmentation of the papyrus surfaces.",
      "evidence_urls": [
        "https://scrollprize.org/grandprize"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2024-01-alphageometry",
    "title": "AlphaGeometry: Olympiad geometry without human demonstrations",
    "summary": "A language model proposing constructions and a symbolic engine deriving proofs solved 25 of 30 problems in the IMO-AG-30 benchmark (the previous method solved 10; the average gold medalist score was 25.9), trained on 100 million synthetic examples. It covers geometry only and a 30-problem benchmark.",
    "field": "mathematics",
    "ai_technology": "AlphaGeometry (language model and symbolic engine)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/alphageometry-an-olympiad-level-ai-system-for-geometry/",
    "source_type": "blog",
    "publication_date": "2024-01-17",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06747-5",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "On the IMO-AG-30 benchmark of 30 Olympiad geometry problems, the system solved 25 versus 10 for the previous method, with each solution checked by computer.",
      "unknown": "It covers geometry only, roughly a third of Olympiad problems, and approaches but does not match gold-medalist performance.",
      "ai_role": "A language model proposes auxiliary constructions and a symbolic engine derives proofs, looping until a proof is found.",
      "evidence_urls": [
        "https://deepmind.google/blog/alphageometry-an-olympiad-level-ai-system-for-geometry/"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Original institutional announcement checked for the specific results, AI contribution and limitations cited in the antibiotic dossier or mathematical follow-up pilot. Full-paper reanalysis, external replication and an exhaustive follow-up search were not performed."
    },
    "follow_up": [
      {
        "id": "alphageometry-2-imo",
        "kind": "same-team extension",
        "date": "2024-07-25",
        "date_precision": "day",
        "note": "DeepMind describes AlphaGeometry 2 as an improved version used for the IMO 2024 geometry problem. The combined result also used AlphaProof; it does not belong to AlphaGeometry alone.",
        "source_urls": [
          "https://deepmind.google/blog/ai-solves-imo-problems-at-silver-medal-level/"
        ],
        "recorded_date": "2026-11-16"
      }
    ],
    "follow_up_review": {
      "date": "2026-11-16",
      "scope": "Bounded check of the original announcements cited by the events recorded here, including explicit links to later program work. This is not an exhaustive literature search or ongoing monitoring commitment."
    }
  },
  {
    "id": "2024-01-elettrolita-pnnl",
    "title": "From 32 million candidates to a new battery electrolyte",
    "summary": "With the PNNL laboratory, Microsoft screened 32.6 million materials using AI models and high-performance computing, narrowing them to 500,000, then 800, 150, and finally an electrolyte replacing some lithium with sodium (about 70% less lithium). PNNL synthesized it and built a working solid-state battery. Validation and optimization are ongoing; the post is by a Microsoft executive.",
    "field": "materials",
    "ai_technology": "Azure Quantum Elements (AI and DFT simulations)",
    "organization": "Microsoft Azure Quantum; and PNNL",
    "source_url": "https://azure.microsoft.com/en-us/blog/quantum/2024/01/09/unlocking-a-new-era-for-scientific-discovery-with-ai-how-microsofts-ai-screened-over-32-million-candidates-to-find-a-better-battery/",
    "source_type": "blog",
    "publication_date": "2024-01-09",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2401.04070",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The top candidates in the Na_xLi_(3-x)YCl6 series were synthesized and experimentally characterized for structure and conductivity; the abstract does not mention battery cell testing.",
      "unknown": "The approximately half-million materials are predictions of potential stability, other candidates are still under investigation, and no explicit limitations are listed.",
      "ai_role": "AI and physics-based models on cloud HPC screened over 32 million candidates and identified 18 promising electrolyte candidates.",
      "evidence_urls": [
        "https://arxiv.org/abs/2401.04070"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-01-flotte-in-mare",
    "title": "Satellite map of industrial activity at sea",
    "summary": "A model applied to about 2 million GB of Sentinel-1 radar imagery (2017–2021) maps vessels and infrastructure: about 75% of industrial fishing vessels are not publicly tracked, over 25% of shipping and energy activity is missing from AIS data, wind turbines more than doubled, and global fishing fell 12% during COVID. Published in Nature; ‘fishing’ classification is inferred; source is the organization’s own release.",
    "field": "other",
    "ai_technology": "Deep learning on radar imagery",
    "organization": "Global Fishing Watch; University of Wisconsin–Madison, Duke, UC Santa Barbara and SkyTruth",
    "source_url": "https://globalfishingwatch.org/press-release/new-research-harnesses-ai-and-satellite-imagery-to-reveal-the-expanding-footprint-of-human-activity-at-sea/",
    "source_type": "press release",
    "publication_date": "2024-01-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06825-8",
    "relevance": {
      "impact": 4,
      "validation": 1,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The page describes no ground-truth check, accuracy metric or comparison for the vessel and infrastructure mapping, so the main claim is not evaluated in the source.",
      "unknown": "Not all boats broadcast their position, the 'fishing' classification is inferred, and the page gives no accuracy figures for that step.",
      "ai_role": "Machine learning detected vessels and offshore infrastructure in satellite imagery and inferred which non-broadcasting vessels were likely fishing; the model type is not named.",
      "evidence_urls": [
        "https://globalfishingwatch.org/press-release/new-research-harnesses-ai-and-satellite-imagery-to-reveal-the-expanding-footprint-of-human-activity-at-sea/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2023-12-antibiotici-mrsa",
    "title": "A new class of antibiotic candidates against MRSA",
    "summary": "A deep-learning model trained on about 39,000 compounds screened about 12 million commercial compounds; of about 280 purchased and tested, two from the same class reduced MRSA burden tenfold in two mouse models, with very low toxicity to human cells. Results from plates and mice; toxicity was partly predicted computationally.",
    "field": "medicine",
    "ai_technology": "Deep learning with chemical-structure explainability",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2023/using-ai-mit-researchers-identify-antibiotic-candidates-1220",
    "source_type": "press release",
    "publication_date": "2023-12-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06887-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "About 283 compounds were bought and tested empirically; one structural class reduced bacterial burden in mouse skin and systemic thigh infection models, and cytotoxicity to human cells was measured.",
      "unknown": "The page does not list specific study limitations; next steps are chemical property analysis and potential clinical use.",
      "ai_role": "Graph neural network ensembles trained on about 39,000 compounds with measured antibacterial and human-cell toxicity predicted both properties for about 12 million compounds and flagged candidates for purchase and testing.",
      "evidence_urls": [
        "https://news.mit.edu/2023/using-ai-mit-researchers-identify-antibiotic-candidates-1220",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06887-8&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-12-coscientist",
    "title": "Coscientist: autonomous chemical research with language models",
    "summary": "A GPT-4-guided system designs, plans and runs experiments using web search, documentation, code and laboratory automation, performing six tasks including optimization of palladium cross-coupling reactions. The abstract does not make clear which tasks used real hardware; only the abstract was reviewed.",
    "field": "chemistry",
    "ai_technology": "Coscientist (GPT-4)",
    "organization": "Carnegie Mellon University; and Emerald Cloud Lab",
    "source_url": "https://www.nature.com/articles/s41586-023-06792-0",
    "source_type": "paper",
    "publication_date": "2023-12-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The system was run on six tasks, including a palladium-catalysed cross-coupling reaction optimization that the abstract reports as successful.",
      "unknown": "The abstract does not state which tasks used physical hardware and which used simulation, and states no limitations.",
      "ai_role": "GPT-4 planned and executed experiments by using web search, documentation, code execution and experimental automation tools.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06792-0&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-12-leganti-peptidi",
    "title": "Designed proteins bind peptides with the highest affinity reported",
    "summary": "Using RFdiffusion and ProteinMPNN, researchers designed proteins that bind helical peptides such as glucagon, neuropeptide Y and parathyroid hormone, with what the authors consider the highest reported affinity for a computer-generated biomolecule. They detect low concentrations of peptides in human serum and withstand heat; a parathyroid-hormone biosensor produced a 21-fold increase in bioluminescence. The press release gives no affinity values.",
    "field": "biology",
    "ai_technology": "RFdiffusion and ProteinMPNN",
    "organization": "University of Washington Institute for Protein Design",
    "source_url": "https://www.ipd.uw.edu/2023/12/ai-generates-proteins-with-exceptional-binding-strengths/",
    "source_type": "press release",
    "publication_date": "2023-12-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06953-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Designed binders were tested in the lab for peptide detection (mass spectrometry, including low-concentration peptide detection) and a bioluminescence biosensor produced a 21-fold signal increase with parathyroid hormone; the press release gives no affinity values.",
      "unknown": "The abstract does not describe how affinity was measured, does not mention serum testing, and states no limitations; the press release states no limitations.",
      "ai_role": "RFdiffusion, extended for flexible targets and partial diffusion, designed binders to helical peptides, de novo or by refining other designs.",
      "evidence_urls": [
        "https://www.ipd.uw.edu/2023/12/ai-generates-proteins-with-exceptional-binding-strengths/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06953-1&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-12-funsearch",
    "title": "FunSearch: new mathematical results with a language model",
    "summary": "A language model (PaLM 2) paired with an automatic evaluator found the largest cap sets on record in some dimensions, the biggest increase in 20 years, and bin-packing heuristics using fewer bins than established methods. The problem must first be expressed as code with an evaluation procedure.",
    "field": "mathematics",
    "ai_technology": "FunSearch (PaLM 2 and automatic evaluator)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models/",
    "source_type": "blog",
    "publication_date": "2023-12-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06924-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Programs were scored automatically by an evaluator; on the cap set problem they found larger cap sets than the best known, and on online bin packing their heuristics beat established baselines.",
      "unknown": "The abstract does not explicitly state limitations; the blog says the method needs an automatic evaluation procedure and that the bin-packing heuristics were tailored to specific data.",
      "ai_role": "A pretrained large language model (PaLM 2) generated candidate programs in an evolutionary search guided by an automatic evaluator.",
      "evidence_urls": [
        "https://deepmind.google/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06924-6&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-12-mattergen-anteprima",
    "title": "MatterGen: stable materials generated from desired properties",
    "summary": "A diffusion model generates stable, novel materials directly from desired properties: 2.9 times more stable, novel and unique structures than CDVAE, and over 250 materials with bulk modulus above 400 GPa versus 2 in the reference dataset. Results were checked with DFT calculations; experimental verification remains the ultimate test.",
    "field": "materials",
    "ai_technology": "MatterGen (diffusion model)",
    "organization": "Microsoft Research",
    "source_url": "https://www.microsoft.com/en-us/research/blog/mattergen-property-guided-materials-design/",
    "source_type": "blog",
    "publication_date": "2023-12-07",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2312.03687",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Generated structures were checked with DFT calculations and compared with CDVAE and screening baselines (2.9 times more stable, novel and unique structures than CDVAE, per the blog); experimental verification is stated as not yet done.",
      "unknown": "Limitations stated: DFT has known limitations and experimental verification remains the ultimate test; the arXiv abstract states no limitations.",
      "ai_role": "A diffusion model generated new crystal structures conditioned on desired properties.",
      "evidence_urls": [
        "https://arxiv.org/abs/2312.03687",
        "https://www.microsoft.com/en-us/research/blog/mattergen-property-guided-materials-design/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-11-gnome",
    "title": "GNoME: 2.2 million new predicted crystals",
    "summary": "A graph neural network predicted 2.2 million new crystals, including 380,000 of the most stable, raising the number of known stable materials from about 48,000 to 421,000. External laboratories synthesized 736 predicted structures. Stability is calculated and does not guarantee a material can be produced or will be useful.",
    "field": "materials",
    "ai_technology": "GNoME (Graph Networks for Materials Exploration)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/millions-of-new-materials-discovered-with-deep-learning/",
    "source_type": "blog",
    "publication_date": "2023-11-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06735-9",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Predicted stability was checked computationally with DFT, and external laboratories independently synthesized 736 of the predicted structures.",
      "unknown": "The blog states that most of the 380,000 stable predictions had not been made in a lab and that stability is judged by current scientific standards; manufacturing and usefulness remain open.",
      "ai_role": "A graph neural network trained on about 48,000 stable crystals predicted 2.2 million new crystals, including 380,000 stable ones.",
      "evidence_urls": [
        "https://deepmind.google/blog/millions-of-new-materials-discovered-with-deep-learning/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06735-9&resultType=core&format=json"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2023-11-a-lab",
    "title": "A-Lab: an autonomous laboratory for synthesizing inorganic materials",
    "summary": "In the Nature paper, the autonomous lab made compounds from computed targets over 17 days of continuous operation. The original press release reported 41 of 58 (71%), while the author’s January 19, 2026 correction reports 36 of 57: on review, 36 of 40 claimed successes were confirmed, 4 were inconclusive from diffraction alone, and ‘new’ meant new to the platform, not to science. One compound was removed because it was in the training data.",
    "field": "materials",
    "ai_technology": "A-Lab (autonomous lab with language models and active learning)",
    "organization": "Lawrence Berkeley National Laboratory; and UC Berkeley",
    "source_url": "https://www.nature.com/articles/s41586-023-06734-w",
    "source_type": "paper",
    "publication_date": "2023-11-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The autonomous lab attempted 58 target materials in 17 days of continuous operation and produced 41 novel compounds; failed syntheses were analyzed to suggest improvements.",
      "unknown": "The abstract does not name the characterization methods and states no limitations beyond the failed-synthesis analysis; the record notes an erratum published in January 2026.",
      "ai_role": "A robotic platform used computation, literature data, machine learning and active learning to plan and interpret solid-state powder syntheses.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06734-w&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-11-chroma",
    "title": "Chroma: a generative model for protein design",
    "summary": "Generate:Biomedicines published the Chroma model in Nature; it designs new proteins with requested biophysical, biological and therapeutic properties. The weights are freely available to academic and nonprofit researchers. The press release gives no experimental figures, and performance claims come from the company.",
    "field": "biology",
    "ai_technology": "Chroma (generative model)",
    "organization": "Generate:Biomedicines",
    "source_url": "https://generatebiomedicines.com/news-releases/generatebiomedicines-announces-publication-of-its-chroma-model-in-nature/",
    "source_type": "press release",
    "publication_date": "2023-11-15",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06728-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The Nature paper reports experimental characterization of 310 designed proteins and crystal structures of two designs matching the predicted structures (backbone RMSD about 1.0 Å); the press release itself describes no experiments and states no limitations.",
      "unknown": "The paper abstract states no limitations; the press release gives no experimental figures and its performance claims come from the company.",
      "ai_role": "A generative model samples protein structures and sequences and can be conditioned on specified properties or functions.",
      "evidence_urls": [
        "https://generatebiomedicines.com/news-releases/generatebiomedicines-announces-publication-of-its-chroma-model-in-nature/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06728-8&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-11-graphcast",
    "title": "GraphCast: global 10-day weather forecasts",
    "summary": "It outperforms ECMWF’s HRES system on over 90% of 1,380 variables and test forecast lead times, and produces a 10-day forecast in under a minute on a single TPU. Comparison is limited to deterministic HRES; training uses ERA5 data.",
    "field": "climate",
    "ai_technology": "GraphCast (graph neural network)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/",
    "source_type": "blog",
    "publication_date": "2023-11-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adi2336",
    "relevance": {
      "impact": 5,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "10-day forecasts of hundreds of variables at 0.25° in under a minute; outperforms the most accurate operational deterministic systems on 90% of 1,380 verification targets; supports better tropical-cyclone, atmospheric-river and extreme-temperature prediction.",
      "unknown": "Comparison with ensemble systems is not described in the abstract.",
      "ai_role": "A machine-learning model trained directly on reanalysis data.",
      "evidence_urls": [
        "https://doi.org/10.1126/science.adi2336"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-11-metnet3",
    "title": "MetNet-3: short-range neural forecasts in Google products",
    "summary": "The model forecasts precipitation, temperature, wind and dew point up to 24 hours ahead at 1–4 km resolution, learning from about 1,000 stations, and outperforms HRRR and ENS in several regions. It operates in Google Search and the app in the United States and 27 European countries, with precipitation forecasts up to 12 hours. Preprint, no numerical values in the text and one case study.",
    "field": "climate",
    "ai_technology": "MetNet-3",
    "organization": "Google Research; and Google DeepMind",
    "source_url": "https://research.google/blog/metnet-3-a-state-of-the-art-neural-weather-model-available-in-google-products/",
    "source_type": "blog",
    "publication_date": "2023-11-01",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2306.06079",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "MetNet-3 was benchmarked against HRRR and ENS over the contiguous US for forecasts up to 24 hours, scored with CRPS; a single January 2019 case study is also shown.",
      "unknown": "The post has no limitations section; it notes operational coverage limited to the contiguous US and parts of Europe, hyperlocal wind results depend on test-station input, and the case study is a single example.",
      "ai_role": "A neural model forecasts precipitation, temperature, wind and dew point up to 24 hours ahead, learning from about 1,000 stations.",
      "evidence_urls": [
        "https://arxiv.org/abs/2306.06079",
        "https://research.google/blog/metnet-3-a-state-of-the-art-neural-weather-model-available-in-google-products/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-10-meg-immagini",
    "title": "Meta AI: near-real-time image decoding from brain activity",
    "summary": "A three-part system (image encoder, brain encoder, decoder) reconstructs viewed images from magnetoencephalography signals at each time point, with signals better aligned to DINOv2 than the other models compared. Images remain imperfect; fMRI decodes better; the page gives no numerical metrics; preprint.",
    "field": "biology",
    "ai_technology": "Magnetoencephalography decoder",
    "organization": "Meta AI",
    "source_url": "https://ai.meta.com/blog/brain-ai-image-decoding-meg-magnetoencephalography/",
    "source_type": "blog",
    "publication_date": "2023-10-18",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2310.19812",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "MEG recordings from healthy volunteers viewing images were decoded at each time point; several pretrained image models were compared (signals aligned best with DINOv2), and MEG decoding was compared with fMRI. Participant and image counts not stated on the blog.",
      "unknown": "Generated images are imperfect: high-level features are often kept but low-level features are often wrong. MEG decoding is less precise than fMRI decoding.",
      "ai_role": "AI models built the image representations and generated the output images; volunteers supplied the brain recordings.",
      "evidence_urls": [
        "https://ai.meta.com/blog/brain-ai-image-decoding-meg-magnetoencephalography/",
        "https://arxiv.org/abs/2310.19812"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-10-ercolano-prime-lettere",
    "title": "Vesuvius Challenge: first letters read from an unopened Herculaneum scroll",
    "summary": "For the first time, a complete word (‘porphyras’, purple) was read from inside a never-opened carbonized scroll; a model trained on crack patterns yielded 13 letters independently and unanimously annotated by papyrologists. The letters are faint and lack context; other readings are not ruled out.",
    "field": "humanities",
    "ai_technology": "Machine-learning models for ink detection",
    "organization": "Vesuvius Challenge",
    "source_url": "https://scrollprize.org/firstletters",
    "source_type": "press release",
    "publication_date": "2023-10-12",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "A model trained on crack patterns produced 13 letters, which papyrologists independently annotated with varying confidence; the page links only to code and no paper.",
      "unknown": "The letters are faint and lack context, word boundaries are unclear, and the page states other readings are not ruled out and the papyrological team was still investigating.",
      "ai_role": "Machine-learning models trained on ink crackle patterns (and a second model with pseudo-labeling) read letters from the scroll.",
      "evidence_urls": [
        "https://scrollprize.org/firstletters"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2023-10-evescape",
    "title": "EVEscape: predicting viral variants before they emerge",
    "summary": "Starting from January 2020, the tool predicted SARS-CoV-2 mutations with accuracy similar to experimental antibody-binding assays and ranked their prevalence better; it generalizes to HIV and influenza. Retrospective validation; prospective performance is not reported.",
    "field": "biology",
    "ai_technology": "EVEscape",
    "organization": "Harvard Medical School; and University of Oxford",
    "source_url": "https://hms.harvard.edu/news/ai-tool-can-help-forecast-viral-outbreaks",
    "source_type": "press release",
    "publication_date": "2023-10-11",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06617-0",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "EVEscape was trained only on pre-2020 sequences and compared against high-throughput experimental scans of SARS-CoV-2 variation that later emerged, a retrospective validation; prospective performance is not reported.",
      "unknown": "The abstract states no limitations; the page says testing on Lassa and Nipah is ongoing and vaccine or therapy use against future variants is a future application.",
      "ai_role": "A deep learning model of historical viral sequences, combined with biophysical and structural information, scored the escape potential of mutations.",
      "evidence_urls": [
        "https://hms.harvard.edu/news/ai-tool-can-help-forecast-viral-outbreaks",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06617-0&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-09-alphamissense",
    "title": "AlphaMissense: cataloging missense genetic mutations",
    "summary": "It classified 89% of the 71 million possible missense variants: 32% as likely pathogenic and 57% as likely benign, with 90% precision on a database of known variants. Predictions are not intended for direct clinical use.",
    "field": "biology",
    "ai_technology": "AlphaMissense",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/a-catalogue-of-genetic-mutations-to-help-pinpoint-the-cause-of-diseases/",
    "source_type": "blog",
    "publication_date": "2023-09-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adg7492",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The model was validated on genetic and experimental benchmarks, and Genomics England cross-referenced its predictions with previously aggregated human variant pathogenicity data and found them consistent. The reported comparison uses previously collected pathogenicity and variant data; an independent data source is distinct from a new independent experimental replication.",
      "unknown": "The model does not predict structural or stability effects, the blog says predictions are not designed for direct clinical use, and the abstract describes no clinical validation.",
      "ai_role": "AlphaMissense, an AlphaFold adaptation fine-tuned on human and primate variant population frequency data, classified missense variants and released predictions for all possible human substitutions.",
      "evidence_urls": [
        "https://deepmind.google/blog/a-catalogue-of-genetic-mutations-to-help-pinpoint-the-cause-of-diseases/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.adg7492&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-08-bci-parlato-stanford",
    "title": "A neuroprosthesis translates attempted speech at 62 words per minute",
    "summary": "For a participant with ALS, intracortical electrodes and an AI algorithm converted attempted speech into text at 62 words per minute. With a 125,000-word vocabulary, the error rate was 23.8%, versus 9.1% for only 50 words. Proof of concept in one person, with an invasive implant and experimental device.",
    "field": "medicine",
    "ai_technology": "Neural network for phoneme decoding and language model",
    "organization": "Stanford Medicine; and BrainGate",
    "source_url": "https://med.stanford.edu/news/all-news/2023/08/brain-implant-speech-als.html",
    "source_type": "press release",
    "publication_date": "2023-08-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06377-x",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "One participant with ALS attempted speech in real time; word error rates were measured at 50-word and 125,000-word vocabularies, with decoding at 62 words per minute.",
      "unknown": "The abstract states no limitations; the results come from one participant, the large-vocabulary error rate (23.8%) is substantial and decoding is well below conversational speed.",
      "ai_role": "An AI algorithm decoded spiking activity from intracortical microelectrode arrays into text.",
      "evidence_urls": [
        "https://med.stanford.edu/news/all-news/2023/08/brain-implant-speech-als.html",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06377-x&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-08-bci-parlato-avatar",
    "title": "Woman with paralysis communicates through a brain-controlled avatar",
    "summary": "For a participant with stroke-related paralysis, an electrode array on the cortex and AI algorithms translated attempted speech into text, synthetic voice and avatar movements. Text reached nearly 80 words per minute after individual training. Experimental study in one person; the system requires a physical connection to computers.",
    "field": "medicine",
    "ai_technology": "Neural speech decoding and AI facial animation",
    "organization": "UC San Francisco; and UC Berkeley",
    "source_url": "https://engineering.berkeley.edu/news/2023/08/novel-brain-implant-helps-paralyzed-woman-speak-using-a-digital-avatar/",
    "source_type": "press release",
    "publication_date": "2023-08-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06443-4",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "One clinical-trial participant with paralysis attempted silent speech; decoding was measured in real time at a median of 78 words per minute with a 25% median word error rate.",
      "unknown": "The abstract does not explicitly state limitations; the single-participant study and 25% word error rate are the main caveats it implies.",
      "ai_role": "Deep-learning models decoded text from surface speech-cortex signals, synthesized personalized speech and animated a facial avatar.",
      "evidence_urls": [
        "https://engineering.berkeley.edu/news/2023/08/novel-brain-implant-helps-paralyzed-woman-speak-using-a-digital-avatar/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06443-4&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-08-peptidi-ominini",
    "title": "Neanderthal and Denisovan peptides reconstructed as antibiotic candidates",
    "summary": "AI analysis of proteins from extinct hominins identified sequences that were then synthesized and tested against pathogens in culture and in mice. Some peptides showed antimicrobial activity; these are preclinical results, with no demonstration of safety or efficacy in humans.",
    "field": "medicine",
    "ai_technology": "AI for molecular de-extinction",
    "organization": "University of Pennsylvania",
    "source_url": "https://www.engineering.upenn.edu/stories/for-a-new-generation-of-antibiotics-scientists-are-bringing-extinct-molecules-back-to-life-and-discovering-the-hidden-genetics-of-immunity-along-the-way/",
    "source_type": "press release",
    "publication_date": "2023-08-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.chom.2023.07.001",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Fragments were tested for antimicrobial activity in vitro, and representative fragments reduced A. baumannii infection in mouse skin abscess and thigh infection models.",
      "unknown": "The abstract does not explicitly state limitations; the findings rest on in vitro work and two mouse models rather than human trials.",
      "ai_role": "A random forest model (panCleave) predicted protease cleavage sites across proteomes to find antimicrobial fragments in modern human and extinct hominin proteins.",
      "evidence_urls": [
        "https://www.engineering.upenn.edu/stories/for-a-new-generation-of-antibiotics-scientists-are-bringing-extinct-molecules-back-to-life-and-discovering-the-hidden-genetics-of-immunity-along-the-way/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.chom.2023.07.001&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-08-scie-di-condensazione",
    "title": "Avoiding aircraft contrails with AI forecasts",
    "summary": "In a trial with American Airlines and Breakthrough Energy, pilots flew 70 flights over six months using AI forecasts to avoid forming contrails, reducing them by 54% compared with flights without the forecasts. Flights that avoided contrails used 2% more fuel; across an airline’s flights, the impact could fall to 0.3% if only a small share need rerouting.",
    "field": "climate",
    "ai_technology": "AI contrail-formation forecasts and satellite imagery",
    "organization": "Google Research; American Airlines and Breakthrough Energy",
    "source_url": "https://blog.google/technology/ai/ai-airlines-contrails-climate-change/",
    "source_type": "blog",
    "publication_date": "2023-08-08",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Google, American Airlines and Breakthrough Energy flew 70 pilot flights over six months using AI forecasts to avoid contrail-forming altitudes; satellite imagery reported a 54% contrail reduction.",
      "unknown": "The article does not state a control group, the baseline for the 54% figure, statistical methods or peer review; fuel impact (2% more fuel, 0.3% airline-wide) is an estimate.",
      "ai_role": "AI contrail forecasts guided pilots to avoid altitudes likely to form contrails.",
      "evidence_urls": [
        "https://blog.google/technology/ai/ai-airlines-contrails-climate-change/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-08-masai-mammografia",
    "title": "MASAI: AI-assisted mammography screening in a randomized trial",
    "summary": "In the initial analysis of 80,033 women, AI-assisted screening detected 244 cancers versus 203 with standard double reading and reduced the reading workload by 44%. A Lund update on January 30, 2026, after more than two years and about 106,000 women, reports higher sensitivity without more false positives. The comparison is in a Swedish setting; it does not demonstrate reduced mortality.",
    "field": "medicine",
    "ai_technology": "AI for mammogram triage and marking",
    "organization": "Lund University; and Unilabs",
    "source_url": "https://www.lu.se/artikel/ai-stodd-modell-saker-mammografiscreening",
    "source_type": "press release",
    "publication_date": "2023-08-02",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/S0140-6736(25)02464-X",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "A prospective randomized single-site trial compared AI-supported screening (about 40,003 women) with standard double reading (about 40,030 women), measuring cancers detected, false positives and radiologist workload.",
      "unknown": "The Lund article states limits: single site, no mortality outcomes, and the main analysis awaits two-year follow-up; the January 2026 update could not be verified on the page read.",
      "ai_role": "AI flagged higher-risk mammography exams for two physicians or marked suspicious findings for a single reader.",
      "evidence_urls": [
        "https://www.lu.se/artikel/ai-stodd-modell-saker-mammografiscreening"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2023-07-rfdiffusion",
    "title": "RFdiffusion: designing new proteins with a diffusion model",
    "summary": "A guided diffusion model generates tailored protein structures, such as proteins that bind tightly to another molecule, and outperforms previous methods across a wide range of problems. The post says only that ‘a useful fraction’ of generated designs work as intended, without figures; the software was released in March 2023.",
    "field": "biology",
    "ai_technology": "RFdiffusion",
    "organization": "University of Washington Institute for Protein Design",
    "source_url": "https://www.ipd.uw.edu/2023/07/ai-tools-are-designing-entirely-new-proteins-that-could-transform-medicine/",
    "source_type": "press release",
    "publication_date": "2023-07-12",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06415-8",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Designs were characterized experimentally (symmetric assemblies, metal-binding proteins, protein binders), and a cryo-EM structure of a designed binder was nearly identical to the design model; the abstract reports no overall success rates.",
      "unknown": "The abstract states no explicit limitations; the press post gives no success rates, sample sizes or benchmark numbers and says only that a useful fraction of designs work.",
      "ai_role": "A generative diffusion model fine-tuned from RoseTTAFold generated protein backbones for monomer, binder, oligomer, enzyme scaffold and motif designs.",
      "evidence_urls": [
        "https://www.ipd.uw.edu/2023/07/ai-tools-are-designing-entirely-new-proteins-that-could-transform-medicine/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06415-8&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-07-pangu-weather",
    "title": "Pangu-Weather: medium-range forecasts outperform the operational ECMWF model",
    "summary": "According to the authors, it is the first AI model to outperform ECMWF’s IFS at medium range: five-day Z500 RMSE of 296.7 versus 333.7, a global 24-hour forecast in 1.4 seconds on a GPU, and mean cyclone-track error of 120 km at three days versus 162 km. Source reviewed: Huawei Cloud blog, not the Nature paper (July 6, 2023); it underestimates extremes.",
    "field": "climate",
    "ai_technology": "Pangu-Weather",
    "organization": "Huawei Cloud",
    "source_url": "https://www.huaweicloud.com/intl/en-us/about/blogs/20230707.html",
    "source_type": "blog",
    "publication_date": "2023-07-07",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06185-3",
    "relevance": {
      "impact": 5,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Trained on 39 years of global data; stronger deterministic results on reanalysis data than ECMWF's operational IFS in all tested variables; also evaluated on extreme weather and ensemble forecasts.",
      "unknown": "Comparisons are initialized with reanalysis data, as the abstract states; performance from operational analyses is not given there.",
      "ai_role": "Three-dimensional deep networks with Earth-specific priors and hierarchical temporal aggregation.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-023-06185-3"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-07-nowcastnet",
    "title": "NowcastNet: extreme-precipitation forecasts up to three hours ahead",
    "summary": "A model combining neural networks and physical constraints forecasts precipitation up to three hours ahead; 62 meteorologists preferred it in 71% of 2,400 evaluated events. Tsinghua reports it was already deployed on the National Meteorological Center’s operational SWAN 3.0 platform. Expert preference does not measure reduced damage or guarantee accuracy for every event.",
    "field": "climate",
    "ai_technology": "NowcastNet",
    "organization": "Tsinghua University / National Meteorological Center / UC Berkeley",
    "source_url": "https://www.thss.tsinghua.edu.cn/info/1023/2133.htm",
    "source_type": "press release",
    "publication_date": "2023-07-07",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06184-4",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "62 forecasters from 23 meteorological stations ran retrospective and prospective checks on about 2,400 extreme-precipitation events and ranked NowcastNet first in 71% of cases against leading methods.",
      "unknown": "No limitations of NowcastNet are stated; the abstract reports no verification scores beyond expert preference, and the page says the team plans to extend the approach to other physics problems.",
      "ai_role": "A neural network combined with physics-based precipitation evolution forecasts extreme precipitation up to three hours ahead from radar data.",
      "evidence_urls": [
        "https://www.thss.tsinghua.edu.cn/info/1023/2133.htm",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06184-4&resultType=core&format=json"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2023-06-llnl-cogsim-ignizione",
    "title": "CogSim helps predict fusion ignition at the NIF",
    "summary": "Neural networks, Bayesian inference and physics simulations helped assess, before the experiment, the NIF design that produced 3.15 MJ of fusion energy from 2.05 MJ of laser energy delivered to the target on December 5, 2022. Experts designed the experiment; AI estimated its outcome and uncertainties. This is not the energy balance of the entire facility; the study was later published in Science in 2025.",
    "field": "physics",
    "ai_technology": "CogSim (neural networks, transfer learning and Bayesian inference)",
    "organization": "Lawrence Livermore National Laboratory",
    "source_url": "https://lmf.llnl.gov/news/high-performance-computing-ai-cognitive-simulation-helped-llnl-conquer-fusion-ignition",
    "source_type": "press release",
    "publication_date": "2023-06-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adm8201",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The CogSim team predicted the December 5, 2022 shot outcome before the experiment (about 50.2% chance of exceeding break-even), and the result of 3.15 MJ from 2.05 MJ matched the prediction after peer review and independent diagnostics.",
      "unknown": "The page says the approach does not guarantee success and that techniques still have much to prove; one good prediction does not validate a model, and testing on more experiments was planned for 2023.",
      "ai_role": "A generative model combining simulations, deep learning, experimental data and Bayesian statistics estimated the outcome and uncertainties of the NIF design before the shot.",
      "evidence_urls": [
        "https://lmf.llnl.gov/news/high-performance-computing-ai-cognitive-simulation-helped-llnl-conquer-fusion-ignition",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.adm8201&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-06-alphadev",
    "title": "AlphaDev: reinforcement learning discovers faster sorting routines",
    "summary": "Working at the assembly-instruction level, the system found sorting routines up to 70% faster for sequences of 3–5 elements and about 1.7% faster for sequences over 250,000 elements; they were adopted in the LLVM libc++ standard library. A hashing algorithm 30% faster was added to Abseil. The largest gains are for short sequences.",
    "field": "computer science",
    "ai_technology": "AlphaDev (reinforcement learning)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/alphadev-discovers-faster-sorting-algorithms/",
    "source_type": "blog",
    "publication_date": "2023-06-07",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06004-9",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "AlphaDev found sort 3 and sort 5 routines with fewer instructions than human benchmarks, matched sort 4, and had lower latency on variable sorts. Libc++ versions were up to 70% faster for five elements and about 1.7% faster above 250,000 elements.",
      "unknown": "Hashing result (30% faster) not seen in the text read. Search is at assembly level; the paper reports optimality proof only for sort 3 (17 instructions). Results cover tested CPUs: ARMv8, Intel Skylake, AMD Zen 2.",
      "ai_role": "AlphaDev, an AlphaZero-based reinforcement learning agent with a transformer encoder, built assembly programs instruction by instruction, rewarded for correctness and latency; the team translated results to C++.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-023-06004-9"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-05-abaucina",
    "title": "Abaucin: a targeted antibiotic against Acinetobacter baumannii",
    "summary": "A model trained on about 7,500 compounds screened 6,680 new compounds in under two hours; of 240 tested in the laboratory, nine were antibiotics, including abaucin, which was effective against wound infections in mice. Activity is highly narrow (no effect on the other bacteria tested) and still preclinical.",
    "field": "medicine",
    "ai_technology": "Machine learning for compound screening",
    "organization": "MIT; and McMaster University",
    "source_url": "https://news.mit.edu/2023/using-ai-scientists-combat-drug-resistant-infections-0525",
    "source_type": "press release",
    "publication_date": "2023-05-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41589-023-01349-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "About 7,500 molecules were screened in vitro, a neural network predicted new compounds that were tested in the laboratory (about 240 tested, nine antibiotics), and abaucin controlled an A. baumannii infection in a mouse wound model.",
      "unknown": "The abstract states no explicit limitations; the press release says activity is highly narrow (no effect on the other bacteria tested) and the compound remains preclinical.",
      "ai_role": "A neural network trained on screening results predicted structurally new compounds active against A. baumannii.",
      "evidence_urls": [
        "https://news.mit.edu/2023/using-ai-scientists-combat-drug-resistant-infections-0525",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41589-023-01349-8&resultType=core&format=json"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Original institutional announcement checked for the specific results, AI contribution and limitations cited in the antibiotic dossier or mathematical follow-up pilot. Full-paper reanalysis, external replication and an exhaustive follow-up search were not performed."
    },
    "relationships": [
      {
        "type": "extends",
        "target_id": "2020-02-halicin",
        "reason": "The source explicitly traces this A. baumannii screening study to the earlier halicin program. Distinct candidate and experiment; not independent confirmation of halicin.",
        "source_urls": [
          "https://news.mit.edu/2023/using-ai-scientists-combat-drug-resistant-infections-0525"
        ]
      }
    ],
    "follow_up": [
      {
        "id": "generative-design-program-extension",
        "kind": "same-team extension",
        "date": "2025-08-14",
        "date_precision": "day",
        "note": "MIT identifies halicin and abaucin as earlier screening results motivating generative design of new NG1/DN1 candidates. The later study does not establish clinical efficacy of these earlier candidates.",
        "source_urls": [
          "https://news.mit.edu/2025/using-generative-ai-researchers-design-compounds-kill-drug-resistant-bacteria-0814"
        ],
        "recorded_date": "2026-11-16"
      }
    ],
    "follow_up_review": {
      "date": "2026-11-16",
      "scope": "Bounded check of the original announcements cited by the events recorded here, including explicit links to later program work. This is not an exhaustive literature search or ongoing monitoring commitment."
    }
  },
  {
    "id": "2023-05-exominer-69",
    "title": "ExoMiner and multiplicity validate 69 new exoplanets",
    "summary": "ExoMiner V1.2 combines deep learning with information about other signals in the same star system to validate 69 planets in Kepler data. This is statistical validation of transit signals, not new direct observation; the result is distinct from the 301 planets validated in 2021.",
    "field": "physics",
    "ai_technology": "ExoMiner V1.2 / multiplicity boost",
    "organization": "USRA; NASA and SETI Institute",
    "source_url": "https://newsroom.usra.edu/discovery-of-69-new-exoplanets-using-machine-learning/",
    "source_type": "press release",
    "publication_date": "2023-05-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2305.02470",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "ExoMiner V1.2 with multiplicity information was applied to Kepler data and validated 69 planets, compared against baseline classifiers and prior validation work.",
      "unknown": "The arXiv page states no limitations, and the statistical validation and false-positive tests are not described on the page read.",
      "ai_role": "A deep-learning classifier (ExoMiner V1.2) combined with information about other signals in the same star system to validate transit signals.",
      "evidence_urls": [
        "https://arxiv.org/abs/2305.02470",
        "https://newsroom.usra.edu/discovery-of-69-new-exoplanets-using-machine-learning/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-05-interferenza-fotoni",
    "title": "Nonlocal quantum interference without entanglement of intrinsic properties",
    "summary": "An AI- and graph-theory-based prediction was confirmed in two experiments: four photons exhibit nonlocal interference without requiring entanglement of their intrinsic properties. The superposition concerns their generation locations; applications to metrology remain prospective.",
    "field": "physics",
    "ai_technology": "AI and graph theory for quantum experiments",
    "organization": "Max Planck Institute for the Science of Light; Nanjing University and Hefei University",
    "source_url": "https://mpl.mpg.de/news/article/where-quantum-particles-are-born-ai-reveals-new-insights-beyond-entanglement",
    "source_type": "press release",
    "publication_date": "2023-05-17",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-37228-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Four-photon nonlocal interference without entanglement of intrinsic properties was demonstrated experimentally, and the abstract reports the interference controlled by an undetected photon.",
      "unknown": "The abstract states no limitations; the press release says applications to metrology remain prospective.",
      "ai_role": "The press release says an AI- and graph-theory-based prediction was confirmed in two experiments; the Nature Communications abstract does not describe the AI prediction.",
      "evidence_urls": [
        "https://mpl.mpg.de/news/article/where-quantum-particles-are-born-ai-reveals-new-insights-beyond-entanglement",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-023-37228-y&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-05-pangenoma-umano",
    "title": "First draft of the human pangenome reference",
    "summary": "A consortium of 119 scientists from 60 institutions published in Nature a pangenome combining assembled genomes from 47 people of diverse ancestries, rather than one person. Google adapted DeepVariant to identify variants and DeepConsensus to correct sequencing errors; a long-read assembler achieved over 99.999% final accuracy. This is a draft: benefits for diagnosis and treatment remain prospective.",
    "field": "biology",
    "ai_technology": "DeepVariant and DeepConsensus",
    "organization": "Google Research; and Human Pangenome Reference Consortium",
    "source_url": "https://blog.google/innovation-and-ai/technology/health/first-pangenome-reference-nature-paper-ai/",
    "source_type": "blog",
    "publication_date": "2023-05-10",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-05896-x",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The draft pangenome from 47 assemblies was evaluated for accuracy at structural and base-pair levels, and short-read analysis was compared with GRCh38-based workflows (34% fewer small-variant errors).",
      "unknown": "The abstract states no explicit limitations; the blog says this is a draft and benefits for diagnosis and treatment remain prospective.",
      "ai_role": "Google adapted DeepVariant to call variants and DeepConsensus to correct sequencing errors in the pangenome assembly.",
      "evidence_urls": [
        "https://blog.google/innovation-and-ai/technology/health/first-pangenome-reference-nature-paper-ai/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-05896-x&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-05-superconduttore-zr-in-ni",
    "title": "New zirconium-indium-nickel superconductor identified with machine learning",
    "summary": "Predictive models and experts guided fabrication of candidates and identified a zirconium, indium and nickel alloy with a superconducting transition around 9 K. The press release reports a three-month discovery; this is not a room-temperature superconductor or a demonstration of industrial use.",
    "field": "materials",
    "ai_technology": "MITHRIL / machine learning",
    "organization": "Johns Hopkins Applied Physics Laboratory",
    "source_url": "https://www.jhuapl.edu/news/news-releases/230503-ai-discovers-novel-superconductor",
    "source_type": "press release",
    "publication_date": "2023-05-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2212.11855",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Closed-loop ML predictions were tested experimentally: candidates were fabricated and measured, and results fed back to refine the model; a Zr-In-Ni superconductor with transition around 9 K was found in four cycles.",
      "unknown": "The arXiv page states no limitations and omits synthesis, transition temperature and model accuracy details; the press release notes training-data human bias and chemically impossible generated compositions, and cites an npj Computational Materials paper that was not read.",
      "ai_role": "Machine-learning models trained on public superconductor datasets predicted and ranked candidate compositions for fabrication.",
      "evidence_urls": [
        "https://www.jhuapl.edu/news/news-releases/230503-ai-discovers-novel-superconductor",
        "https://arxiv.org/abs/2212.11855"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-05-lineardesign",
    "title": "LinearDesign: more stable mRNA sequences designed by an algorithm",
    "summary": "The algorithm, based on a parsing technique from natural-language processing, finds the most stable mRNA sequence for a given protein in about eleven minutes. According to Baidu’s press release, whose study appeared in Nature, sequences produced this way showed greater stability, protein expression and antibody response than comparators in animal experiments. Figures are company-reported; the Baidu release copy on EurekAlert was reviewed, not a baidu.com page.",
    "field": "medicine",
    "ai_technology": "LinearDesign",
    "organization": "Baidu Research",
    "source_url": "https://www.eurekalert.org/news-releases/987999",
    "source_type": "press release",
    "publication_date": "2023-05-02",
    "added_date": "2026-11-12",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06127-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "LinearDesign-designed mRNA for COVID-19 and VZV was compared with wildtype and codon-optimized sequences in animal experiments, measuring half-life, expression and antibody response; the figures are company-reported and the release describes no independent replication.",
      "unknown": "The Nature abstract states no specific study limitations and the release states none; the release says benefits such as lower doses and fewer side effects are possibilities not tested.",
      "ai_role": "A parsing-based algorithm (LinearDesign) optimized mRNA structural stability and codon usage for a given protein.",
      "evidence_urls": [
        "https://www.eurekalert.org/news-releases/987999",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06127-z&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-05-decoder-semantico",
    "title": "An MRI decoder turns brain activity into text",
    "summary": "A noninvasive decoder reconstructs continuous text from fMRI that closely, and sometimes precisely, matches the original meaning in about half of trials (after dozens of hours of training with podcasts; it also works on silent videos). Published in Nature Neuroscience; it works only with cooperative participants and is not word-for-word transcription.",
    "field": "biology",
    "ai_technology": "Semantic decoder from fMRI",
    "organization": "University of Texas at Austin",
    "source_url": "https://news.utexas.edu/2023/05/01/brain-activity-decoder-can-reveal-stories-in-peoples-minds/",
    "source_type": "press release",
    "publication_date": "2023-05-01",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41593-023-01304-9",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A non-invasive fMRI decoder generated continuous text that conveyed the meaning of perceived speech, imagined speech and silent videos; the press release says it matched the original meaning in about half of trials and works only with cooperative participants.",
      "unknown": "The abstract does not report participant numbers or semantic accuracy measures and states no explicit limitations; the press release says it is not word-for-word transcription.",
      "ai_role": "A decoder trained on fMRI recordings of cortical semantic representations generated word sequences matching the meaning of brain activity.",
      "evidence_urls": [
        "https://news.utexas.edu/2023/05/01/brain-activity-decoder-can-reveal-stories-in-peoples-minds/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41593-023-01304-9&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-04-anticorpi-modelli-linguistici",
    "title": "Language models guide improvements to seven antibodies",
    "summary": "Two rounds of laboratory evolution, using at most 20 variants per antibody, improved the affinity of four mature antibodies by up to sevenfold and three immature ones by up to 160-fold. Biochemical and pseudovirus tests; no demonstration of therapeutic benefit in humans.",
    "field": "biology",
    "ai_technology": "Protein language models",
    "organization": "Stanford University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10869273/",
    "source_type": "paper",
    "publication_date": "2023-04-24",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-023-01763-2",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Language-model-guided variants of seven antibodies were made in two lab evolution rounds (at most 20 variants per antibody) and binding was measured by biolayer interferometry; four antibodies were also checked for thermostability, polyspecificity and pseudovirus neutralization.",
      "unknown": "The authors state that affinity gains are smaller than in vivo somatic hypermutation, the method improves existing binding rather than creating new binders, and it may work less well under other selection pressures; no therapeutic benefit in humans is shown.",
      "ai_role": "General protein language models guided affinity maturation by choosing which antibody variants to test.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10869273/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41587-023-01763-2&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-04-m87-primo",
    "title": "PRIMO: a sharper image of the M87 black hole",
    "summary": "A machine-learning method trained on more than 30,000 simulated black-hole images reconstructed M87* from 2017 Event Horizon Telescope data, using the array’s full resolution: the ring is about half as wide as in the 2019 image. It fills gaps with patterns learned from simulations, without quantified uncertainty.",
    "field": "physics",
    "ai_technology": "PRIMO (dictionary learning)",
    "organization": "Georgia Institute of Technology; Institute for Advanced Study and NSF NOIRLab",
    "source_url": "https://news.research.gatech.edu/2023/04/13/sharper-look-m87-black-hole",
    "source_type": "press release",
    "publication_date": "2023-04-13",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/2041-8213/acc32d",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "The PRIMO reconstruction, built from over 30,000 simulated black-hole images, was checked for consistency with the 2017 EHT data and the 2019 image; no held-out tests are described.",
      "unknown": "The article states that generating the image requires assuming an appropriate form of the missing information and describes no quantified uncertainties; implications for the black hole are framed as future possibilities.",
      "ai_role": "A machine-learning method trained on simulated images filled gaps in the 2017 EHT data to reconstruct the M87* image.",
      "evidence_urls": [
        "https://news.research.gatech.edu/2023/04/13/sharper-look-m87-black-hole"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-02-luxsit-luciferasi",
    "title": "LuxSit: de novo luciferases active in the laboratory",
    "summary": "The group designed new luciferases and identified three active enzymes among the initial designs. The 117-residue LuxSit catalyzes blue light with the DTZ substrate; its function is verified in the laboratory, while therapeutic or industrial applications remain prospective.",
    "field": "biology",
    "ai_technology": "Family-wide hallucination and deep learning",
    "organization": "Institute for Protein Design",
    "source_url": "https://www.ipd.uw.edu/2023/02/machine-learning-generates-custom-enzymes/",
    "source_type": "press release",
    "publication_date": "2023-02-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-05696-3",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Designed luciferases were tested in activity assays with synthetic luciferin substrates; the most active design had catalytic efficiency comparable to native luciferases and high substrate selectivity.",
      "unknown": "The abstract does not state how many designs were tested or how many were active, and states no explicit limitations; therapeutic or industrial applications remain untested.",
      "ai_role": "Deep learning generated idealized scaffolds with binding pockets and matching sequences used to design artificial luciferases.",
      "evidence_urls": [
        "https://www.ipd.uw.edu/2023/02/machine-learning-generates-custom-enzymes/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-05696-3&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-02-incendi-perimetri",
    "title": "Real-time wildfire perimeters from satellite imagery",
    "summary": "A model using geostationary imagery tracks fire perimeters with updates every 10–15 minutes and, according to the page, is more accurate than similar products and the previous model based on NOAA GOES-16/17 data; shown in Search and Maps for large fires in the United States, Mexico, much of Canada and Australia. Resolution is 2 km; clouds can hide fires; figures appear only in an image.",
    "field": "climate",
    "ai_technology": "Model on geostationary satellite imagery",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/real-time-tracking-of-wildfire-boundaries-using-satellite-imagery/",
    "source_type": "blog",
    "publication_date": "2023-02-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Fire perimeters were compared with official fire scars using precision, recall and distance metrics; a comparison with an earlier GOES-16/17 model and a 2022 US test set unavailable during training are also reported.",
      "unknown": "The post states limitations: smoke and clouds hide fires, resolution is about 2 km and coarser away from nadir, and coverage is limited to large fires in the listed regions; no paper is cited.",
      "ai_role": "A model using geostationary satellite imagery tracks fire perimeters with updates every 10-15 minutes.",
      "evidence_urls": [
        "https://research.google/blog/real-time-tracking-of-wildfire-boundaries-using-satellite-imagery/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-01-seti-820-stelle",
    "title": "Deep learning searches for technosignatures around 820 nearby stars",
    "summary": "A deep-learning method analyzed 150 TB of Green Bank Telescope data from 820 stars, previously searched in 2017 without interesting signals, and found eight narrowband signals of interest with nonzero drift, present only in observations where the target was in view. Follow-up observations did not find them again; no claim of extraterrestrial origin.",
    "field": "physics",
    "ai_technology": "Deep learning on radio data",
    "organization": "SETI Institute; Breakthrough Listen and University of Toronto",
    "source_url": "https://www.eurekalert.org/news-releases/978138",
    "source_type": "press release",
    "publication_date": "2023-01-30",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2301.12670",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The deep-learning method analyzed Green Bank Telescope data from 820 stars and flagged eight narrowband candidates; the page reports no follow-up results and no injection tests, and the abstract gives no false-positive figures.",
      "unknown": "The page states no limitations; rejecting human radio interference is named as the core challenge, and the full paper's methods, which the listing says were later published in Nature Astronomy, were not read.",
      "ai_role": "A beta-Convolutional Variational Autoencoder flagged technosignature candidates in a semi-unsupervised manner.",
      "evidence_urls": [
        "https://arxiv.org/abs/2301.12670",
        "https://www.eurekalert.org/news-releases/978138"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-01-sybil-polmone",
    "title": "Sybil: lung-cancer risk prediction up to six years from a low-dose CT",
    "summary": "The model predicts risk from a single CT with a six-year C-index of 0.75 (NLST), 0.81 (MGH) and 0.80 (CGMH), and a one-year ROC area between 0.86 and 0.94. Retrospective validation on three independent datasets; trained only on smokers, with prospective testing and testing in nonsmokers as next steps.",
    "field": "medicine",
    "ai_technology": "Sybil (neural network on CT scans)",
    "organization": "MIT; Mass General Cancer Center and Chang Gung Memorial Hospital",
    "source_url": "https://news.mit.edu/2023/ai-model-can-detect-future-lung-cancer-0120",
    "source_type": "press release",
    "publication_date": "2023-01-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1200/JCO.22.01345",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The model was trained on NLST low-dose CT scans and validated on a held-out NLST set and two independent cohorts (MGH and CGMH), with six-year C-index 0.75, 0.81 and 0.80; the abstract describes no prospective testing.",
      "unknown": "The abstract states that future study is required to understand clinical applications; the press release says prospective testing and testing in nonsmokers are next steps.",
      "ai_role": "A deep learning model predicts lung-cancer risk from a single low-dose chest CT.",
      "evidence_urls": [
        "https://news.mit.edu/2023/ai-model-can-detect-future-lung-cancer-0120",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1200/JCO.22.01345&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-01-hcc-pharma-ai",
    "title": "A liver-cancer target and molecule identified with Pharma.AI",
    "summary": "Pharma.AI combined PandaOmics, Chemistry42 and an AlphaFold-predicted structure to identify a hepatocellular-carcinoma target and produce a hit molecule within 30 days of target selection, synthesizing seven compounds. A second round produced a more potent hit; an early Chemical Science study, with no clinical efficacy demonstrated.",
    "field": "medicine",
    "ai_technology": "Pharma.AI, PandaOmics, Chemistry42 and AlphaFold",
    "organization": "University of Toronto; Insilico Medicine and collaborators",
    "source_url": "https://www.utoronto.ca/news/researchers-use-ai-powered-database-design-potential-cancer-drug-30-days",
    "source_type": "press release",
    "publication_date": "2023-01-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D2SC05709C",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected compounds were synthesized and tested in biological assays: the first hit bound CDK20 (Kd about 9.2 µM) and a second AI-driven round gave a more potent compound with greater antiproliferative activity in liver cancer cells; the abstract describes no in vivo testing.",
      "unknown": "The abstract notes AlphaFold predictions have varying confidence, the first hit came from only 7 synthesized compounds, and potency values vary widely; no clinical efficacy is demonstrated.",
      "ai_role": "Pharma.AI platforms (PandaOmics, Chemistry42) used AlphaFold-predicted structures to select a liver-cancer target and generate candidate molecules.",
      "evidence_urls": [
        "https://www.utoronto.ca/news/researchers-use-ai-powered-database-design-potential-cancer-drug-30-days",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1039/D2SC05709C&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-11-esm-atlas",
    "title": "ESM Metagenomic Atlas: over 600 million protein structures",
    "summary": "Meta AI published predicted structures for almost the entire MGnify90 database (over 600 million), generated in about two weeks on roughly 2,000 GPUs with a 15-billion-parameter language model, up to 60 times faster than previous methods. These are predictions, and high-confidence structures are a much smaller subset.",
    "field": "biology",
    "ai_technology": "ESMFold (15-billion-parameter language model)",
    "organization": "Meta AI",
    "source_url": "https://ai.meta.com/blog/protein-folding-esmfold-metagenomics/",
    "source_type": "blog",
    "publication_date": "2022-11-01",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2022.07.20.500902",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The preprint reports structure prediction up to 60 times faster while maintaining accuracy versus state-of-the-art methods and an atlas of 617 million structures; the blog describes no experimental checks.",
      "unknown": "The blog lists no limitations and notes some sequences fold into disordered forms; high-confidence predictions are a subset.",
      "ai_role": "Language model (15B parameters) predicted protein structures from sequence.",
      "evidence_urls": [
        "https://ai.meta.com/blog/protein-folding-esmfold-metagenomics/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1101/2022.07.20.500902&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-10-alphatensor",
    "title": "AlphaTensor: new algorithms for matrix multiplication",
    "summary": "A reinforcement-learning agent found matrix-multiplication algorithms better than known ones for many sizes (for example, 76 multiplications instead of 80 for 4×5 by 5×5). Algorithms optimized for hardware were 10–20% faster on V100 GPUs and TPU v2. The improvement over Strassen applies in a finite field.",
    "field": "mathematics",
    "ai_technology": "AlphaTensor (based on AlphaZero)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/discovering-novel-algorithms-with-alphatensor/",
    "source_type": "blog",
    "publication_date": "2022-10-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-022-05172-4",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Algorithms are provably correct by construction and their multiplication counts are compared with known methods; hardware-targeted algorithms were benchmarked on V100 GPU and TPU v2.",
      "unknown": "The page lists no formal limitations; it notes the search space is very large and the work is specific to matrix multiplication.",
      "ai_role": "Reinforcement-learning agent searched for tensor decompositions giving new matrix-multiplication algorithms.",
      "evidence_urls": [
        "https://deepmind.google/blog/discovering-novel-algorithms-with-alphatensor/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-022-05172-4&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-10-carbonio-metastabile",
    "title": "Machine learning and synthesis clarify metastable phases of carbon",
    "summary": "An algorithm constructed metastable phase diagrams of carbon and identified structural features of n-diamond that were then verified experimentally. The group also synthesized predicted phases and compared structures under a microscope; the work concerns carbon only and does not establish useful properties of the new materials.",
    "field": "materials",
    "ai_technology": "Machine learning and atomistic simulations",
    "organization": "Argonne National Laboratory",
    "source_url": "https://www.alcf.anl.gov/news/scientists-use-machine-learning-accelerate-materials-discovery",
    "source_type": "press release",
    "publication_date": "2022-10-05",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-022-30820-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Workflow found 505 candidate carbon structures; a neural network trained on 273 phases, test error 37.1 meV/atom. Diamond-anvil laser-heating experiments on graphite with electron microscopy were used to compare predicted metastable phases, including n-diamond.",
      "unknown": "The search is not guaranteed exhaustive; the paper calls cubic-diaphite a candidate structure for n-diamond. The model had higher errors at high pressure; kinetics are only partly covered. Argonne synthesis of unreported phases was not seen in the paper text read.",
      "ai_role": "A deep neural network predicted Gibbs free energy; a support vector machine drew phase boundaries; an evolutionary search proposed structures. Humans ran the experiments.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41467-022-30820-8",
        "https://www.alcf.anl.gov/news/scientists-use-machine-learning-accelerate-materials-discovery"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-09-pink1-parkinson-alphafold",
    "title": "AlphaFold helps interpret activation of PINK1, a Parkinson’s-linked protein",
    "summary": "After experimentally determining different forms of PINK1, the WEHI group used AlphaFold to model the human protein and its dimer, obtaining structures consistent with experimental results. The predictions also helped interpret how mutations associated with early-onset Parkinson’s may affect dimerization; this does not amount to a validated therapy.",
    "field": "biology",
    "ai_technology": "AlphaFold",
    "organization": "WEHI; and Google DeepMind",
    "source_url": "https://deepmind.google/blog/targeting-early-onset-parkinsons-with-ai/",
    "source_type": "blog",
    "publication_date": "2022-09-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The WEHI group compared an AlphaFold model of human PINK1 dimer with its own experimental structure work and used it to model disease mutations, which it says confirmed their suspicions.",
      "unknown": "The article describes no specific limitations of the predictions and says plenty of questions remain.",
      "ai_role": "AlphaFold modelled human PINK1 and its dimer and the effect of mutations on dimerisation.",
      "evidence_urls": [
        "https://deepmind.google/blog/targeting-early-onset-parkinsons-with-ai/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-09-proteinmpnn",
    "title": "ProteinMPNN: protein sequences designed in about one second",
    "summary": "A neural network generates the amino-acid sequence for a given protein shape in about one second, over 200 times faster than the best previous software, without specialist tuning. A related study shows that proteins designed with ProteinMPNN fold correctly much more often, including nanoscale rings and complex assemblies. Researchers describe it as the beginning of machine learning in protein design.",
    "field": "biology",
    "ai_technology": "ProteinMPNN",
    "organization": "University of Washington Institute for Protein Design",
    "source_url": "https://www.ipd.uw.edu/2022/09/proteinmpnn-excels-at-creating-new-proteins/",
    "source_type": "press release",
    "publication_date": "2022-09-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.add2187",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Speed is compared with prior software; a related Science study tested lab-made designs, which folded as intended more often, including nanoscale rings imaged by electron microscopy.",
      "unknown": "The article gives no quantitative success rates and notes protein data is scarce, requiring trial and error.",
      "ai_role": "Neural network generated amino-acid sequences for given protein shapes.",
      "evidence_urls": [
        "https://www.ipd.uw.edu/2022/09/proteinmpnn-excels-at-creating-new-proteins/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1126/science.add2187&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-07-alphafold-200-milioni",
    "title": "AlphaFold Database: over 200 million structures",
    "summary": "The predicted-protein-structure archive, maintained with EMBL-EBI, grew from nearly 1 million to over 200 million structures, covering practically all cataloged proteins, with free access and bulk download from Google Cloud. More than 500,000 researchers in 190 countries had already used it. These are predictions, not experimental measurements.",
    "field": "biology",
    "ai_technology": "AlphaFold 2",
    "organization": "Google DeepMind; and EMBL-EBI",
    "source_url": "https://deepmind.google/discover/blog/alphafold-reveals-the-structure-of-the-protein-universe/",
    "source_type": "blog",
    "publication_date": "2022-07-28",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The post gives no accuracy metrics for the new structures; it cites two Nature AlphaFold papers and uses in experiments as evidence.",
      "unknown": "The post states no explicit limitations or caveats on prediction accuracy.",
      "ai_role": "AlphaFold predicted protein structures added to the AlphaFold Database.",
      "evidence_urls": [
        "https://deepmind.google/discover/blog/alphafold-reveals-the-structure-of-the-protein-universe/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2022-07-trews-sepsi",
    "title": "TREWS: early sepsis alerts evaluated in five hospitals",
    "summary": "A machine-learning system monitored about 590,000 patients in five hospitals, flagging sepsis risk from medical records. In the studied cohort, timely clinical confirmation of alerts was associated with lower mortality, organ failure and length of stay. Observational study: the association does not show that AI caused better outcomes.",
    "field": "medicine",
    "ai_technology": "TREWS (Targeted Real-time Early Warning System)",
    "organization": "Johns Hopkins University; and Bayesian Health",
    "source_url": "https://hub.jhu.edu/2022/07/21/artificial-intelligence-sepsis-detection/",
    "source_type": "press release",
    "publication_date": "2022-07-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-022-01894-0",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A prospective multi-site cohort of 590,736 monitored patients (five hospitals) compared outcomes of sepsis alerts confirmed by a provider within 3 hours with outcomes otherwise, adjusting for severity.",
      "unknown": "The association is observational and does not show that the AI caused better outcomes.",
      "ai_role": "Machine-learning early warning system (TREWS) flagged sepsis risk from medical records.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41591-022-01894-0&resultType=core&format=json"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2022-06-birdnet",
    "title": "BirdNET: citizen science through bird-song identification",
    "summary": "The app recognizes over 3,000 species from their calls, with more than 2.2 million people contributing since 2018; in four test cases, its data reproduced known patterns such as song distribution and the migration ranges of two species. Published in PLOS Biology; only four cases, with no identification-error figures.",
    "field": "biology",
    "ai_technology": "BirdNET (bioacoustics)",
    "organization": "Cornell Lab of Ornithology",
    "source_url": "https://news.cornell.edu/stories/2022/06/ai-powered-birdnet-app-makes-citizen-science-easier",
    "source_type": "press release",
    "publication_date": "2022-06-28",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pbio.3001670",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In four test cases where conventional research had robust answers, BirdNET data replicated known patterns such as song-type distribution and migration ranges.",
      "unknown": "No identification-error rates are reported and the authors describe the validation as the first step.",
      "ai_role": "Bird-song recognition app identified species from calls and produced citizen-science observations.",
      "evidence_urls": [
        "https://news.cornell.edu/stories/2022/06/ai-powered-birdnet-app-makes-citizen-science-easier",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1371/journal.pbio.3001670&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-06-poro-nucleare",
    "title": "Model of the human nuclear pore with AlphaFold2 and electron microscopy",
    "summary": "AlphaFold2 and ColabFold, combined with cryo-tomography and cryo-electron microscopy, produced a model of the human nuclear pore complex (about 30 proteins, about 1,000 copies, 120 MDa) covering over 90% of the complex, versus 46% for previous models. Flexible parts remain unresolved; a complete dynamic model is future work.",
    "field": "biology",
    "ai_technology": "AlphaFold2 and ColabFold",
    "organization": "EMBL",
    "source_url": "https://www.embl.org/news/science/puzzling-out-the-structure-of-a-molecular-giant/",
    "source_type": "press release",
    "publication_date": "2022-06-10",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.abm9506",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The integrative model (AlphaFold2 with cryo-ET and cryo-EM data) was validated against experimental data and covers more than 90% of the NPC core versus 46% for previous models.",
      "unknown": "Flexible peripheral parts are not modelled, no quantitative confidence scores are given, and a complete dynamic model is future work.",
      "ai_role": "AlphaFold2 and ColabFold predicted protein structures that were assembled into the nuclear pore model.",
      "evidence_urls": [
        "https://www.embl.org/news/science/puzzling-out-the-structure-of-a-molecular-giant/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1126/science.abm9506&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-05-microlensing-degenerazione",
    "title": "Machine learning suggests a more general theory of microlensing",
    "summary": "An algorithm highlighted microlensing configurations not explained by usual interpretations, leading to a theory unifying two degeneracies. The authors report a better fit to data from two dozen previous discoveries; they describe the theory as probably incomplete.",
    "field": "physics",
    "ai_technology": "Machine-learning inference",
    "organization": "UC Berkeley / The Ohio State University",
    "source_url": "https://news.berkeley.edu/2022/05/24/ai-reveals-unsuspected-math-underlying-search-for-exoplanets/",
    "source_type": "press release",
    "publication_date": "2022-05-24",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41550-022-01671-6",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The machine-learning-derived theory was checked against the datasets of two dozen earlier microlensing discoveries and fit them better than the earlier interpretations.",
      "unknown": "The authors say the broader theory is likely still incomplete, and the multi-planet case is not yet worked out.",
      "ai_role": "Algorithm highlighted microlensing configurations not explained by the usual interpretations.",
      "evidence_urls": [
        "https://news.berkeley.edu/2022/05/24/ai-reveals-unsuspected-math-underlying-search-for-exoplanets/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41550-022-01671-6&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-04-fast-petase",
    "title": "FAST-PETase: an AI-engineered enzyme degrades PET waste",
    "summary": "A machine-learning model suggested PETase mutations, then tested on 51 post-consumer containers, fibers and fabrics. FAST-PETase degrades PET below 50 °C, in some cases down to monomers within 24 hours; production at scale and environmental use still needed development.",
    "field": "biology",
    "ai_technology": "MutCompute / 3D convolutional network",
    "organization": "University of Texas at Austin",
    "source_url": "https://news.utexas.edu/2022/04/27/plastic-eating-enzyme-could-eliminate-billions-of-tons-of-landfill-waste/",
    "source_type": "press release",
    "publication_date": "2022-04-27",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-022-04599-z",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "FAST-PETase was tested on 51 post-consumer containers, fibers and fabrics, degrading PET below 50 °C, in some cases to monomers within 24 hours.",
      "unknown": "Scale-up, field use at polluted sites and cost are not demonstrated, and the work covers only PET.",
      "ai_role": "Machine-learning model suggested enzyme mutations.",
      "evidence_urls": [
        "https://news.utexas.edu/2022/04/27/plastic-eating-enzyme-could-eliminate-billions-of-tons-of-landfill-waste/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-022-04599-z&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-04-murmurazioni-curve",
    "title": "Murmurations: unexpected oscillations in elliptic-curve data",
    "summary": "Classification experiments and analysis of elliptic-curve data show oscillations in the mean of Frobenius traces that vary with rank. A numerical observation, not proof of a general law; the 2022 preprint was later published in Experimental Mathematics in 2024.",
    "field": "mathematics",
    "ai_technology": "Logistic regression and PCA",
    "organization": "Yang-Hui He; Kyu-Hwan Lee, Thomas Oliver and Alexey Pozdnyakov",
    "source_url": "https://arxiv.org/abs/2204.10140",
    "source_type": "paper",
    "publication_date": "2022-04-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/html/2204.10140v2",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Classification experiments and analysis of elliptic-curve data show oscillations in the average Frobenius trace that vary with rank.",
      "unknown": "Not reported in the arXiv abstract read.",
      "ai_role": "Machine-learning classification experiments were used to classify elliptic curves by rank.",
      "evidence_urls": [
        "https://arxiv.org/abs/2204.10140"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-03-ithaca",
    "title": "Ithaca: restoring and dating ancient Greek inscriptions",
    "summary": "A neural network reconstructs missing text in damaged Greek inscriptions, identifies their provenance and dates them: 62% accuracy in restoration, 71% in location, and dating within 30 years of reference ranges. With Ithaca, historians improved from 25% to 72% correct restorations. It is intended to assist historians.",
    "field": "humanities",
    "ai_technology": "Ithaca (deep neural network)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/discover/blog/predicting-the-past-with-ithaca/",
    "source_type": "blog",
    "publication_date": "2022-03-09",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-022-04448-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Restoration (62%), location (71%) and dating were checked on held-out dated inscriptions, and a historian study rose from 25% to 72% correct restorations with the tool.",
      "unknown": "Accuracy is far from perfect, dating is a range within about 30 years, and the model was trained only on Greek inscriptions.",
      "ai_role": "Neural network restored missing text, identified provenance and dated Greek inscriptions.",
      "evidence_urls": [
        "https://deepmind.google/discover/blog/predicting-the-past-with-ithaca/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-022-04448-z&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-02-fourcastnet",
    "title": "FourCastNet: high-resolution global forecasts with neural networks",
    "summary": "The preprint presents global forecasts at 0.25° resolution and a one-week forecast in under two seconds. Comparison with IFS is favorable at short lead times for large-scale variables and some finer-scale variables; this does not mean general superiority or operational use.",
    "field": "climate",
    "ai_technology": "Adaptive Fourier Neural Operators",
    "organization": "Jaideep Pathak; and collaborators",
    "source_url": "https://arxiv.org/abs/2202.11214",
    "source_type": "paper",
    "publication_date": "2022-02-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Forecasts at 0.25 degrees are compared with ECMWF IFS, matching it at short lead times for large-scale variables and outperforming it for some fine-scale variables such as precipitation.",
      "unknown": "Not reported in the arXiv abstract read.",
      "ai_role": "Adaptive Fourier neural operator produced global weather forecasts.",
      "evidence_urls": [
        "https://arxiv.org/abs/2202.11214"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-02-fusione-tokamak",
    "title": "Tokamak plasma control with reinforcement learning",
    "summary": "A single neural network controls the magnetic coils of Lausanne’s TCV tokamak and forms plasmas in different configurations (snowflake, ITER-like shape, double droplet). Trained mainly in simulation and then tested on the real machine. Stated limitations: plasma lasts about three seconds and the machine changes from day to day.",
    "field": "physics",
    "ai_technology": "Deep reinforcement learning",
    "organization": "Google DeepMind; and EPFL Swiss Plasma Center",
    "source_url": "https://deepmind.google/blog/accelerating-fusion-science-through-learned-plasma-control/",
    "source_type": "blog",
    "publication_date": "2022-02-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-021-04301-9",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "On the TCV tokamak the controller produced and held elongated, conventional, negative-triangularity and snowflake configurations, with accurate tracking of location, current and shape, and sustained two separate 'droplet' plasmas.",
      "unknown": "Transfer to larger or reactor-scale devices is not reported in the abstract.",
      "ai_role": "A deep reinforcement-learning architecture learned to command the full set of control coils from high-level objectives.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-021-04301-9"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-12-dingo",
    "title": "DINGO: gravitational-wave parameters in seconds",
    "summary": "The neural network estimates the mass, spin, position, inclination and distance of binary black holes in seconds rather than hours or months, with about ten days of training, and was tested on eight events from LIGO and Virgo’s O1 and O2 runs. Published in Physical Review Letters; extension to neutron stars is only promising; source reviewed was a Max Planck doctoral-program page.",
    "field": "physics",
    "ai_technology": "DINGO (Bayesian inference with neural networks)",
    "organization": "Max Planck Institute for Intelligent Systems; and Max Planck Institute for Gravitational Physics",
    "source_url": "https://imprs.is.mpg.de/news/machine-learning-decodes-tremors-of-the-universe",
    "source_type": "press release",
    "publication_date": "2021-12-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevLett.127.241103",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Parameter estimates for eight LIGO-Virgo events agree closely with standard inference codes, with inference times cut from days to about 20 seconds.",
      "unknown": "The article says extension to neutron stars and complex signals is promising but not demonstrated.",
      "ai_role": "Neural network acted as a surrogate for Bayesian posterior distributions for gravitational-wave parameters.",
      "evidence_urls": [
        "https://imprs.is.mpg.de/news/machine-learning-decodes-tremors-of-the-universe",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1103/PhysRevLett.127.241103&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-12-dm21-funzionale",
    "title": "DM21: a neural network as a density-functional theory functional",
    "summary": "A neural network represents the functional used in DFT and addresses two longstanding errors of traditional functionals: excessive electron delocalization and spin-symmetry breaking during bond breaking. Published in Science with open code; the post gives no benchmark figures and DFT remains an approximation.",
    "field": "chemistry",
    "ai_technology": "DM21 (neural network for the exchange-correlation functional)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/discover/blog/simulating-matter-on-the-quantum-scale-with-ai/",
    "source_type": "blog",
    "publication_date": "2021-12-09",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.abj6511",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "DM21 is compared with traditional functionals on benchmarks for main-group atoms and molecules and on bond-breaking and charge-localisation examples.",
      "unknown": "The page gives no numerical benchmark figures and says DFT remains an approximation.",
      "ai_role": "Neural network represents the exchange-correlation functional in density functional theory.",
      "evidence_urls": [
        "https://deepmind.google/discover/blog/simulating-matter-on-the-quantum-scale-with-ai/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1126/science.abj6511&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-12-matematica-intuizione",
    "title": "AI-guided pure mathematics: knots and representation theory",
    "summary": "With mathematicians Williamson (Sydney) and Lackenby and Juhász (Oxford), supervised-learning techniques revealed patterns leading to an unexpected link between a knot’s algebraic signature and its geometry (the ‘natural slope’), and to a new algorithm for the combinatorial invariance conjecture. The conjectures and proofs are the mathematicians’; the algorithm was checked on over 3 million examples, not proved in general.",
    "field": "mathematics",
    "ai_technology": "Supervised learning and attribution techniques",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/discover/blog/exploring-the-beauty-of-pure-mathematics-in-novel-ways/",
    "source_type": "blog",
    "publication_date": "2021-12-01",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-021-04086-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Conjectures and proofs were established by mathematicians, and the new algorithm was computationally verified on over 3 million examples.",
      "unknown": "Certain patterns still elude ML, and the representation-theory structure appears only at 100s or 1000s of vertices.",
      "ai_role": "Supervised learning revealed patterns that led to the knot and representation-theory results and a new algorithm.",
      "evidence_urls": [
        "https://deepmind.google/discover/blog/exploring-the-beauty-of-pure-mathematics-in-novel-ways/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-021-04086-x&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-11-proteine-allucinate",
    "title": "A neural network “hallucinates” new protein structures",
    "summary": "The network predicted about 2,000 new sequences as foldable; over 100 were produced in the lab, and for three, nuclear magnetic resonance and X-ray crystallography confirmed the predicted shapes. Only three proteins were characterized in detail; using the method to create useful functions (drugs, enzymes) is a future goal.",
    "field": "biology",
    "ai_technology": "Neural network for structure prediction (hallucination)",
    "organization": "University of Washington Institute for Protein Design",
    "source_url": "https://newsroom.uw.edu/news-releases/deep-learning-dreams-new-protein-structures",
    "source_type": "press release",
    "publication_date": "2021-11-30",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-021-04184-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "129 hallucinated sequences were synthesised, 27 gave monodisperse proteins, and three structures determined by X-ray crystallography or NMR matched the hallucinated models.",
      "unknown": "Only three proteins were characterised in detail, and using the method for useful functions is a future goal.",
      "ai_role": "Neural network hallucinated new protein sequences predicted to fold.",
      "evidence_urls": [
        "https://newsroom.uw.edu/news-releases/deep-learning-dreams-new-protein-structures",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-021-04184-w&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-11-exominer",
    "title": "ExoMiner: 301 new planets validated in Kepler data",
    "summary": "The ExoMiner neural network, run on the Pleiades supercomputer, validated 301 planets from the remaining Kepler candidates, adding to the 4,569 already validated; none is Earth-like or in the habitable zone. “Validated” is statistical, not confirmed; NASA’s 2026 count gives 370 for the same work, an unexplained discrepancy.",
    "field": "physics",
    "ai_technology": "ExoMiner (deep neural network)",
    "organization": "NASA Ames Research Center; and NASA JPL",
    "source_url": "https://www.nasa.gov/missions/kepler/new-deep-learning-method-adds-301-planets-to-keplers-total-count/",
    "source_type": "press release",
    "publication_date": "2021-11-23",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2111.10009",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "ExoMiner was evaluated on a test set (recall 0.936 at 99% precision in the arXiv abstract) and used to validate 301 new planets.",
      "unknown": "Validation is statistical rather than confirmation, the article gives no false-positive rates, and none of the planets is Earth-like.",
      "ai_role": "Deep learning classifier vetted transit signals from Kepler.",
      "evidence_urls": [
        "https://www.nasa.gov/missions/kepler/new-deep-learning-method-adds-301-planets-to-keplers-total-count/",
        "https://arxiv.org/abs/2111.10009"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-10-enformer",
    "title": "Enformer: gene expression predicted from DNA sequence",
    "summary": "A Transformer-based model predicts gene expression from 200,000 base pairs of input DNA, over five times more than earlier methods, and is much more accurate at predicting variant effects. Enhancer attribution scores are comparable to methods using experimental data. The post gives no numerical gain and acknowledges that linking enhancers to the genes they regulate remains an open problem.",
    "field": "biology",
    "ai_technology": "Enformer (Transformer)",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/predicting-gene-expression-with-ai/",
    "source_type": "blog",
    "publication_date": "2021-10-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-021-01252-x",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Enformer is compared with prior models on variant-effect prediction and enhancer contribution scores against methods using experimental data, with a rs11644125 case study.",
      "unknown": "Linking enhancers to the genes they regulate is described as a major unsolved problem, and many disease variants are spuriously correlated.",
      "ai_role": "Transformer model predicted gene expression from 200,000 base pairs of DNA.",
      "evidence_urls": [
        "https://deepmind.google/blog/predicting-gene-expression-with-ai/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41592-021-01252-x&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-09-nowcasting-pioggia",
    "title": "Short-term rainfall prediction with a generative model",
    "summary": "Developed with the UK Met Office, the model forecasts precipitation up to two hours ahead; over 50 expert meteorologists chose it as their first choice in 89% of cases compared with the most widely used methods. The authors cite long-range accuracy and rare, intense events as limitations.",
    "field": "climate",
    "ai_technology": "Deep generative radar model (DGMR)",
    "organization": "Google DeepMind; and Met Office",
    "source_url": "https://deepmind.google/blog/nowcasting-the-next-hour-of-rain/",
    "source_type": "blog",
    "publication_date": "2021-09-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-021-03854-z",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Over 50 Met Office meteorologists chose the model first in 89% of cases, and it was compared with PySTEPS and UNet on UK and US radar data.",
      "unknown": "No method is without limitations; long-range accuracy and rare intense events need improvement.",
      "ai_role": "Generative model forecast precipitation up to two hours ahead.",
      "evidence_urls": [
        "https://deepmind.google/blog/nowcasting-the-next-hour-of-rain/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-021-03854-z&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-09-climate-trace",
    "title": "Climate TRACE: first independent database of global emissions",
    "summary": "Using machine learning on data from over 300 satellites and 11,100 sensors (over 59 trillion bytes), emissions from oil and gas in major countries were estimated at about twice UNFCCC reports (about 1 billion tonnes more), while rice emissions in India were nearly three times the official 2016 inventory. No peer review; the page gives both September 14 and September 16, 2021.",
    "field": "climate",
    "ai_technology": "Machine learning on satellite and sensor data",
    "organization": "Climate TRACE",
    "source_url": "https://climatetrace.org/news/climate-trace-releases-first-comprehensive-independent",
    "source_type": "press release",
    "publication_date": "2021-09-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 1,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The page gives no methodology for checking the estimates; it says more than 50 outside organisations reviewed methodology and validated findings, without describing results.",
      "unknown": "The page notes granularity is still developing and that estimates are hedged and constantly evolving; no peer review is mentioned.",
      "ai_role": "Machine learning on satellite and sensor data estimated global emissions.",
      "evidence_urls": [
        "https://climatetrace.org/news/climate-trace-releases-first-comprehensive-independent"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2021-08-icenet",
    "title": "IceNet: seasonal forecasts of Arctic sea ice",
    "summary": "The model outperforms ECMWF’s SEAS5 dynamical system in seasonal forecasts of summer sea ice, especially for extreme events, and is over 2,000 times faster (under ten seconds on a GPU). SEAS5 is better at a one-month lead, and the September comparison covers nine years; published in Nature Communications.",
    "field": "climate",
    "ai_technology": "IceNet (probabilistic deep learning)",
    "organization": "British Antarctic Survey; Alan Turing Institute, and others",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8390499/",
    "source_type": "paper",
    "publication_date": "2021-08-26",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "IceNet is compared with the SEAS5 dynamical model in seasonal summer sea-ice forecasts, especially extreme events.",
      "unknown": "Not reported in the abstract read.",
      "ai_role": "Probabilistic deep-learning model forecast the next six months of sea-ice concentration.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=PMCID:PMC8390499&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-08-ins018073-fibrosi-renale",
    "title": "INS018_073: AI identifies a preclinical candidate for renal fibrosis",
    "summary": "Insilico Medicine named INS018_073 a preclinical candidate after in vitro tests and testing in a mouse model of renal fibrosis. The company reports improvements in pathology, biomarkers, and histologically observed fibrosis. The candidate was proposed for IND-enabling studies; these are company-reported preclinical results, not clinical evidence or independent confirmation.",
    "field": "medicine",
    "ai_technology": "Pharma.AI / Chemistry42",
    "organization": "Insilico Medicine",
    "source_url": "https://insilico.com/blog/pcc_kidney",
    "source_type": "blog",
    "publication_date": "2021-08-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Company-reported in vitro potency (nanomolar) and a mouse unilateral ureteral obstruction model with histology, plus a 14-day mouse safety study.",
      "unknown": "No effect sizes, doses or biomarkers are named, and no independent confirmation or peer review is stated.",
      "ai_role": "AI-driven target and compound identification (PandaOmics) produced the candidate.",
      "evidence_urls": [
        "https://insilico.com/blog/pcc_kidney"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-07-alphafold-database",
    "title": "AlphaFold Protein Structure Database: first 350,000 models",
    "summary": "Predicted structures for the human proteome (about 20,000 proteins) and 20 other organisms were published, over 350,000 models in total, hosted free by EMBL-EBI. DeepMind announced an expansion to over 100 million structures. These are predictions that complement experimental methods. Note: the page header says July 22, 2022, but the text calls CASP14 “last December” and the 2022 post refers to it as a 2021 release; date corrected to 2021.",
    "field": "biology",
    "ai_technology": "AlphaFold 2",
    "organization": "Google DeepMind; and EMBL-EBI",
    "source_url": "https://deepmind.google/discover/blog/putting-the-power-of-alphafold-into-the-worlds-hands",
    "source_type": "blog",
    "publication_date": "2021-07-22",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-021-03828-1",
    "relevance": {
      "impact": 5,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Predictions cover 98.5% of human proteins; 58% of residues have a confident prediction and 36% very high confidence. Case studies illustrate possible hypotheses.",
      "unknown": "The structures are predictions; the abstract does not report experimental confirmation of new structures.",
      "ai_role": "AlphaFold 2 applied at proteome scale, with confidence metrics built on the model.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-021-03828-1"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-07-progen-enzimi",
    "title": "ProGen generates artificial proteins with enzyme activity",
    "summary": "A protein language model generated synthetic lysozymes: two of the five characterized in cells degraded bacterial cell walls with activity comparable to egg-white lysozyme. Laboratory tests and crystallography support the result, but do not demonstrate clinical applications. The preprint dates to July 18, 2021; the Nature Biotechnology publication and UCSF announcement are from January 2023.",
    "field": "biology",
    "ai_technology": "ProGen",
    "organization": "UCSF / Salesforce Research",
    "source_url": "https://www.ucsf.edu/news/2023/01/424641/ai-technology-generates-original-proteins-scratch",
    "source_type": "press release",
    "publication_date": "2021-07-18",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1101/2021.07.18.452833",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Five model-generated lysozyme families were experimentally characterised, with activity and catalytic efficiency compared with natural lysozymes, and crystal structures reported.",
      "unknown": "Not reported in the abstract read.",
      "ai_role": "Protein language model trained on over 280 million sequences generated artificial lysozymes.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1101/2021.07.18.452833&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-07-rosettafold",
    "title": "RoseTTAFold: accessible prediction of protein structures",
    "summary": "A three-track neural network (sequence, residue interactions, and 3D structure) computes a structure in about ten minutes on a single gaming computer. Baker’s group predicted hundreds of new structures, including little-known proteins in the human genome. The web server received over 4,500 requests in the previous month. The page does not state limitations.",
    "field": "biology",
    "ai_technology": "RoseTTAFold",
    "organization": "University of Washington Institute for Protein Design",
    "source_url": "https://www.ipd.uw.edu/2021/07/rosettafold-accurate-protein-structure-prediction-accessible-to-all/",
    "source_type": "press release",
    "publication_date": "2021-07-15",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.abj8754",
    "relevance": {
      "impact": 4,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The page describes the tool's speed and hundreds of new structures but does not describe how predictions were checked.",
      "unknown": "Not reported.",
      "ai_role": "Three-track neural network predicted protein structures.",
      "evidence_urls": [
        "https://www.ipd.uw.edu/2021/07/rosettafold-accurate-protein-structure-prediction-accessible-to-all/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1126/science.abj8754&resultType=core&format=json"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2021-07-bci-parole-paralisi",
    "title": "A neuroprosthesis restores words for a man with paralysis",
    "summary": "Neural networks decoded the attempted speech of a man with stroke-related paralysis, displaying words and sentences on a screen. With a 50-word vocabulary, the system reached up to 18 words per minute and a median accuracy of 75%. Proof of principle in one person, using an implant on the brain surface and a limited vocabulary.",
    "field": "medicine",
    "ai_technology": "Neural networks for speech and language model",
    "organization": "UC San Francisco",
    "source_url": "https://womenshealth.ucsf.edu/coe/news/article/neuroprosthesis-restores-words-man-paralysis",
    "source_type": "press release",
    "publication_date": "2021-07-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1056/NEJMoa2027540",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Decoding was tested in one participant over 48 sessions using a 50-word vocabulary, reaching up to 18 words per minute and a median accuracy of 75%.",
      "unknown": "It is a proof of principle in one person with a limited vocabulary, per the press release.",
      "ai_role": "Deep-learning algorithms decoded attempted words from cortical activity.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1056/NEJMoa2027540&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-06-h01-connettoma",
    "title": "H01: reconstruction of one cubic millimeter of human cerebral cortex",
    "summary": "A sample of about 1 mm³ of human cortex was imaged at 4 nm across about 5,300 sections and automatically reconstructed: 1.4 petabytes, tens of thousands of neurons, 130 million annotated synapses, and 104 manually corrected cells. The sample is about one-millionth of a brain; reconstruction is imperfect and initial findings are preliminary.",
    "field": "biology",
    "ai_technology": "Automated segmentation of electron microscopy images",
    "organization": "Google Research; and Harvard (Lichtman Lab)",
    "source_url": "https://research.google/blog/a-browsable-petascale-reconstruction-of-the-human-cortex/",
    "source_type": "blog",
    "publication_date": "2021-06-01",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://www.biorxiv.org/content/10.1101/2021.05.29.446289",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The automated reconstruction was checked by manual proofreading of about 104 cells; the abstract says connections were validated.",
      "unknown": "Automated reconstruction is imperfect, the sample is about one-millionth of a brain, and findings are preliminary.",
      "ai_role": "Automated machine-learning pipeline reconstructed neurons and synapses from electron microscopy images.",
      "evidence_urls": [
        "https://research.google/blog/a-browsable-petascale-reconstruction-of-the-human-cortex/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1101/2021.05.29.446289&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-05-eagle-elettrocardiogrammi",
    "title": "EAGLE: AI electrocardiograms to detect reduced heart function",
    "summary": "A randomized study across 120 primary-care groups at 45 sites evaluated neural-network alerts on electrocardiograms from 22,641 adults. The intervention produced five additional diagnoses of low ejection fraction per thousand patients, without a global increase in echocardiography use. The result concerns identification of dysfunction, not improved clinical outcomes.",
    "field": "medicine",
    "ai_technology": "Convolutional neural network for electrocardiograms",
    "organization": "Mayo Clinic",
    "source_url": "https://newsnetwork.mayoclinic.org/discussion/trial-demonstrates-early-ai-guided-detection-of-heart-disease-in-routine-practice/",
    "source_type": "press release",
    "publication_date": "2021-05-06",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-021-01335-4",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A cluster-randomised trial across 120 primary-care teams (22,641 adults) compared AI-ECG alerts with usual care on new low-EF diagnoses within 90 days.",
      "unknown": "Not reported in the abstract read; the trial measured diagnosis rather than clinical outcomes.",
      "ai_role": "Neural network analysed electrocardiograms and alerted clinicians to possible low ejection fraction.",
      "evidence_urls": [
        "https://newsnetwork.mayoclinic.org/discussion/trial-demonstrates-early-ai-guided-detection-of-heart-disease-in-routine-practice/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41591-021-01335-4&resultType=core&format=json"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2021-04-combinatoria-reti-neurali",
    "title": "Combinatorial constructions and counterexamples with neural networks",
    "summary": "The deep cross-entropy method (reinforcement learning) finds constructions and counterexamples to open conjectures in extremal combinatorics and graph theory, including a Brualdi–Cao question about permanents. The counterexamples are verifiable; source checked was only the arXiv paper, with no institutional announcement.",
    "field": "mathematics",
    "ai_technology": "Deep cross-entropy (reinforcement learning)",
    "organization": "Adam Zsolt Wagner",
    "source_url": "https://arxiv.org/abs/2104.14516",
    "source_type": "paper",
    "publication_date": "2021-04-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The deep cross-entropy method found explicit constructions and counterexamples, verifiable by checking, including a refutation of a Brualdi and Cao question.",
      "unknown": "Not reported in the arXiv abstract read.",
      "ai_role": "Reinforcement learning (deep cross-entropy method) searched for constructions and counterexamples.",
      "evidence_urls": [
        "https://arxiv.org/abs/2104.14516"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-04-isaia-due-scribi",
    "title": "Two scribes in the Great Isaiah Scroll: new evidence from handwriting",
    "summary": "Independent analyses of character shapes distinguish two groups across the scroll’s 54 columns, consistent with two scribes alternating around the manuscript’s midpoint. This is statistical evidence from one artifact; the writers’ identities are unknown, and other explanations for the differences are not entirely ruled out.",
    "field": "humanities",
    "ai_technology": "Pattern recognition and Kohonen maps",
    "organization": "University of Groningen",
    "source_url": "https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0249769",
    "source_type": "paper",
    "publication_date": "2021-04-21",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pone.0249769",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Independent analyses of character shapes across 54 columns separated two groups, treated as statistical evidence from one artifact.",
      "unknown": "Not reported in the abstract read.",
      "ai_role": "Pattern recognition and AI techniques analysed handwriting features to identify writers.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1371/journal.pone.0249769&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-03-peptidi-ibm",
    "title": "IBM designs two new antimicrobial peptides with generative models",
    "summary": "Over 48 days, 20 candidates were designed, synthesized, and tested; two showed activity against several pathogens, including multidrug-resistant Klebsiella pneumoniae. In vitro and mouse tests only, with no demonstration of safety or efficacy in humans.",
    "field": "medicine",
    "ai_technology": "Generative autoencoder and CLaSS",
    "organization": "IBM Research",
    "source_url": "https://research.ibm.com/blog/ai-finds-new-peptides",
    "source_type": "blog",
    "publication_date": "2021-03-11",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41551-021-00689-x",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "20 generative-model candidates were synthesised and tested; two were active against several pathogens, with in vitro and mouse tests.",
      "unknown": "No demonstration of safety or efficacy in humans, and no human trials are mentioned.",
      "ai_role": "Generative models designed candidate antimicrobial peptides.",
      "evidence_urls": [
        "https://research.ibm.com/blog/ai-finds-new-peptides",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41551-021-00689-x&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-03-trrosetta-pfam",
    "title": "trRosetta releases over 6,300 predicted protein-family structures",
    "summary": "Over 6,300 protein structures predicted with deep learning are available in Pfam to explore families without a known structure. These are computational models with estimated quality, not newly experimentally determined structures; this result is distinct from RoseTTAFold and the AlphaFold Database.",
    "field": "biology",
    "ai_technology": "trRosetta",
    "organization": "Institute for Protein Design / EMBL-EBI",
    "source_url": "https://www.ipd.uw.edu/2021/03/trrosetta-reveals-structures-for-every-protein-family/",
    "source_type": "press release",
    "publication_date": "2021-03-03",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Pfam models are scored with lDDT (0.6 or higher considered reasonable); the article describes no independent validation.",
      "unknown": "The article does not state limitations and says more structures are to follow.",
      "ai_role": "Deep-learning trRosetta predicted structures for protein families.",
      "evidence_urls": [
        "https://www.ipd.uw.edu/2021/03/trrosetta-reveals-structures-for-every-protein-family/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2021-01-mirai-rischio-seno",
    "title": "Mirai: future breast cancer risk from mammograms",
    "summary": "Mirai, trained on over 200,000 exams, was evaluated on hospital data from the United States, Sweden, and Taiwan. In a comparison of high-risk groups at Massachusetts General Hospital, it identified nearly twice as many future diagnoses as Tyrer-Cuzick; it does not demonstrate reduced mortality.",
    "field": "medicine",
    "ai_technology": "Deep learning on mammograms",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2021/robust-artificial-intelligence-tools-predict-future-cancer-0128",
    "source_type": "press release",
    "publication_date": "2021-01-28",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/scitranslmed.aba4373",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Mirai was validated on held-out test sets from MGH, Karolinska and Chang Gung, and identified nearly twice as many future diagnoses as Tyrer-Cuzick in high-risk groups.",
      "unknown": "It does not use prior imaging and is still being validated in more diverse populations, and it does not show reduced mortality.",
      "ai_role": "Deep learning predicted future breast cancer risk from mammograms.",
      "evidence_urls": [
        "https://news.mit.edu/2021/robust-artificial-intelligence-tools-predict-future-cancer-0128",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1126/scitranslmed.aba4373&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-01-covid-radiografie-meta",
    "title": "Sequential X-rays to predict COVID-19 deterioration",
    "summary": "Models trained in part on 26,838 X-rays from 4,914 patients predict deterioration up to 96 hours ahead, outperforming specialists in some retrospective comparisons. This research is for hospital resource planning, not a clinical product or evidence of better patient outcomes.",
    "field": "medicine",
    "ai_technology": "MoCo and Transformer",
    "organization": "Meta AI / NYU Langone Health",
    "source_url": "https://ai.meta.com/blog/new-ai-research-to-help-predict-covid-19-resource-needs-from-a-series-of-x-rays/",
    "source_type": "blog",
    "publication_date": "2021-01-15",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2101.04909",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Models trained on chest X-rays are compared with specialists in some retrospective comparisons for predicting deterioration.",
      "unknown": "Not reported in the arXiv abstract read; the research is for resource planning, not a clinical product.",
      "ai_role": "Self-supervised learning (MoCo) predicted COVID-19 deterioration from sequential chest X-rays.",
      "evidence_urls": [
        "https://arxiv.org/abs/2101.04909"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-01-viral-escape-mit",
    "title": "Language models identify viral regions less prone to immune escape",
    "summary": "Analysis of influenza, HIV, and coronavirus sequences identifies regions of surface proteins less likely to mutate to escape immune responses, including the S2 subunit of spike. These are computational leads for studying possible vaccine targets, not validated vaccines.",
    "field": "biology",
    "ai_technology": "Protein-sequence language model",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2021/model-viruses-escape-immune-0114",
    "source_type": "press release",
    "publication_date": "2021-01-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.abd7331",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Language models trained on about 60,000 HIV, 45,000 influenza and 4,000 coronavirus sequences were checked against earlier findings on influenza and HIV; the SARS-CoV-2 analysis is described as not yet peer-reviewed.",
      "unknown": "Flagged SARS-CoV-2 sequences still need investigation, and it is unknown how fast SARS-CoV-2 mutates, so how long current vaccines stay effective is unclear.",
      "ai_role": "Adapted language-model methods to viral protein sequences to predict surface regions more or less likely to enable immune escape, as candidate vaccine targets.",
      "evidence_urls": [
        "https://news.mit.edu/2021/model-viruses-escape-immune-0114",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.abd7331&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-01-lenti-desi",
    "title": "1,210 gravitational-lens candidates in the DESI Legacy Surveys",
    "summary": "Applied to DR8 data from the DESI Legacy Imaging Surveys, machine learning found 1,210 strong gravitational-lens candidates. The DESI collaboration says this is about twice the number of lenses known at the time; these are candidates, not objects individually confirmed in every case.",
    "field": "physics",
    "ai_technology": "Residual neural network",
    "organization": "NSF NOIRLab; University of San Francisco, and Lawrence Berkeley National Laboratory",
    "source_url": "https://www.desi.lbl.gov/2021/01/14/desi-legacy-imaging-surveys-doubles-the-number-of-known-gravitational-lenses/",
    "source_type": "press release",
    "publication_date": "2021-01-14",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2005.04730",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Candidates were selected by applying trained networks to survey images and visually inspecting the top-ranked ones; the sources read describe no spectroscopic confirmation or held-out test.",
      "unknown": "Not reported",
      "ai_role": "Deep residual neural networks trained on known lenses and non-lenses found 1,210 strong-lens candidates in DESI Legacy Imaging Survey DR8 data.",
      "evidence_urls": [
        "https://www.desi.lbl.gov/2021/01/14/desi-legacy-imaging-surveys-doubles-the-number-of-known-gravitational-lenses/",
        "https://arxiv.org/abs/2005.04730"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-11-alphafold-casp14",
    "title": "AlphaFold wins CASP14 for protein folding",
    "summary": "AlphaFold achieved a median GDT score of 92.4 across the target set (about 1.6 Å mean error) and 87.0 on the hardest targets, which organizers judged a solution to the problem of single-protein structures. DeepMind notes that protein complexes and interactions with other molecules remain open challenges.",
    "field": "biology",
    "ai_technology": "AlphaFold 2",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/",
    "source_type": "blog",
    "publication_date": "2020-11-30",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 5,
      "validation": 4,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Blind assessment at CASP14 on recently determined protein structures: median GDT of 92.4 across targets and 87.0 on the hardest free-modelling targets, judged by the organizers as a solution to single-protein structure prediction.",
      "unknown": "Protein complexes, interactions with DNA, RNA or small molecules, and full side-chain placement remain open; not every predicted structure will be perfect.",
      "ai_role": "Attention-based neural network using evolutionary sequence alignments and residue-pair representations to predict 3D protein structure from sequence.",
      "evidence_urls": [
        "https://deepmind.google/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2020-11-cameo-gst467",
    "title": "CAMEO discovers GST467, a phase-change material with double the optical contrast",
    "summary": "CAMEO, using Bayesian active learning and closed-loop X-ray measurements, identified the Ge4Sb6Te7 alloy in 19 experimental cycles (about 10 hours), selected from 177 compositions. GST467 showed twice the optical contrast of GST225 and was demonstrated in a phase-change memory device; the result concerns this library and application, not a commercial memory.",
    "field": "materials",
    "ai_technology": "CAMEO, closed-loop Bayesian active learning",
    "organization": "NIST; University of Maryland, University of Washington, SLAC",
    "source_url": "https://www.nist.gov/news-events/news/2020/11/nist-ai-system-discovers-new-material",
    "source_type": "press release",
    "publication_date": "2020-11-24",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-020-19597-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "CAMEO ran a closed loop over 177 candidate compositions in 19 experimental cycles; GST467 was measured by X-ray diffraction, showed twice the optical contrast of the reference alloy, and was demonstrated in a phase-change memory device.",
      "unknown": "Long-term performance and larger-scale behavior are not discussed; the tenfold time reduction is an estimate, not a measured result.",
      "ai_role": "Closed-loop Bayesian active learning chose which compositions to measure next and ran the X-ray measurements.",
      "evidence_urls": [
        "https://www.nist.gov/news-events/news/2020/11/nist-ai-system-discovers-new-material",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-020-19597-w&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-10-covid-tosse-mit",
    "title": "A model analyzes coughs to detect COVID-19 infections",
    "summary": "A model trained on volunteer recordings recognized 98.5% of coughs associated with confirmed infections in the test reported by MIT. This is an early result on data collected online, with health status also self-reported; it is not a diagnosis or prospective clinical validation.",
    "field": "medicine",
    "ai_technology": "Neural networks for audio",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2020/covid-19-cough-cellphone-detection-1029",
    "source_type": "press release",
    "publication_date": "2020-10-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1109/OJEMB.2020.3026928",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Model trained on 4,256 subjects and tested on 1,064 held-out subjects from web-collected cough recordings, with results validated against official COVID-19 test outcomes; sensitivity 98.5% and specificity 94.2% overall, but 83.2% specificity in asymptomatic subjects.",
      "unknown": "Not a diagnosis for symptomatic people and does not separate Covid from flu or asthma; status was self-reported; false-positive results for healthy participants not reported; app would need approval and a larger, more diverse dataset.",
      "ai_role": "Neural networks trained on cough biomarkers (vocal cords, sentiment in speech, lung performance) combined with a muscular-degradation estimate to flag Covid-positive coughs.",
      "evidence_urls": [
        "https://news.mit.edu/2020/covid-19-cough-cellphone-detection-1029",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1109/OJEMB.2020.3026928&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-10-ferminet",
    "title": "FermiNet: quantum physics of molecules from first principles",
    "summary": "FermiNet approximates molecular ground-state electron wavefunctions from first principles. Against the coupled-cluster reference, it recovered 99.8% of the correlation energy beyond a single Slater determinant for the smallest molecules and at least 97% for bicyclobutane (30 electrons). It also modeled stretched nitrogen and hydrogen chains. These are computational results from the original work announced on 19 October 2020. The source page was updated on 22 August 2024 with a separate excited-state study; that later study's carbon-dimer error is not a 2020 result.",
    "field": "chemistry",
    "ai_technology": "FermiNet",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/ferminet-quantum-physics-and-chemistry-from-first-principles/",
    "source_type": "blog",
    "publication_date": "2020-10-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevResearch.2.033429",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "Energies compared with coupled-cluster and variational Monte Carlo reference calculations: 99.8% of the correlation energy beyond a single determinant recovered for the smallest molecules and at least 97% for bicyclobutane.",
      "unknown": "The coupled-cluster baseline scales poorly, so results apply to medium-sized molecules; the Monte Carlo method is approximate; excited-state results are within about 0.1 eV of the most demanding calculations rather than matching them.",
      "ai_role": "Neural network wavefunction, antisymmetric in its inputs, trained by minimizing energy with Monte Carlo sampling using only nuclear positions.",
      "evidence_urls": [
        "https://deepmind.google/blog/ferminet-quantum-physics-and-chemistry-from-first-principles/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-10-alberi-africa-nasa",
    "title": "Over 1.8 billion trees mapped in Africa’s drylands",
    "summary": "Analysis of high-resolution images located and measured the crowns of over 1.8 billion trees across 1.3 million km² of West Africa. Validation used manual annotations and field data; the study counts trees but does not yet measure their stored carbon.",
    "field": "climate",
    "ai_technology": "Convolutional neural network",
    "organization": "NASA",
    "source_url": "https://www.nasa.gov/centers-and-facilities/goddard/nasa-supercomputing-study-breaks-ground-for-tree-mapping-carbon-research/",
    "source_type": "press release",
    "publication_date": "2020-10-16",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-020-2824-5",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Tree detections checked against manually annotated data and field data from the region; the abstract itself does not describe the validation method.",
      "unknown": "The study counts trees but does not yet measure stored carbon; the method can mistake stones for trees and gets harder with more detail.",
      "ai_role": "Deep learning (fully convolutional network trained on about 90,000 hand-marked trees) on submetre satellite imagery mapped the crowns of over 1.8 billion trees.",
      "evidence_urls": [
        "https://www.nasa.gov/centers-and-facilities/goddard/nasa-supercomputing-study-breaks-ground-for-tree-mapping-carbon-research/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-020-2824-5&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-10-chematica",
    "title": "Chematica: computational synthesis planning for natural products",
    "summary": "The system plans multistep syntheses of complex natural products, with routes in a Turing-test study largely indistinguishable from human routes; three computer-designed syntheses were carried out in the lab: dauricine, tacamonidine, and lamellodysidine A. Based on over 100,000 rules plus AI, it is not modern generative AI.",
    "field": "chemistry",
    "ai_technology": "Chematica / Allchemy",
    "organization": "Polish Academy of Sciences; Northwestern University, IBS Center for Soft and Living Matter, and UNIST",
    "source_url": "https://www.nature.com/articles/s41586-020-2855-y",
    "source_type": "paper",
    "publication_date": "2020-10-13",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-020-2855-y",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a Turing-like test, synthesis experts largely could not distinguish the program's routes from human ones; three computer-designed natural-product syntheses were carried out in the lab.",
      "unknown": "How often its routes fail across other targets is not reported in the abstract.",
      "ai_role": "The planning program combines chemistry knowledge, data-based AI routines and causal relationships to plan several steps ahead.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-020-2855-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-07-chimico-robotico-mobile",
    "title": "A mobile robotic chemist",
    "summary": "A mobile robot ran 688 experiments in eight days in a ten-variable space, guided by a batch Bayesian algorithm, to search for photocatalysts that produce hydrogen from water; the best mixtures were six times more active than the initial formulations. This optimizes a formulation, not a new structure; only the abstract was reviewed.",
    "field": "chemistry",
    "ai_technology": "Batch Bayesian optimization and mobile robot",
    "organization": "University of Liverpool",
    "source_url": "https://www.nature.com/articles/s41586-020-2442-2",
    "source_type": "paper",
    "publication_date": "2020-07-08",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "688 autonomous experiments over eight days in a ten-variable space; the best photocatalyst mixtures were about six times more active than the initial formulations.",
      "unknown": "The abstract lists no limitations of this study; it notes that atomistic simulations cannot yet capture multi-scale complexity and that robots are hard to adopt widely across diverse samples and instruments.",
      "ai_role": "A batch Bayesian algorithm chose which experiments the mobile robot ran next, automating the researcher rather than the instruments.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-020-2442-2&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-05-covid-tc-mount-sinai",
    "title": "AI and clinical data to identify COVID-19 from CT scans",
    "summary": "In a test on 279 patients, a system combining CT scans and clinical information reached 84% sensitivity, compared with 75% for radiologists in the reported comparison. Retrospective study on data collected in China, without evidence of effectiveness in prospective clinical use.",
    "field": "medicine",
    "ai_technology": "Multimodal deep learning",
    "organization": "Mount Sinai",
    "source_url": "https://www.mountsinai.org/about/newsroom/2020/mount-sinai-first-in-us-to-use-artificial-intelligence-to-analyze-coronavirus-covid19-patients-pr",
    "source_type": "press release",
    "publication_date": "2020-05-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-020-0931-3",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Trained on 626 of 905 patients and tested on the remaining 279 against radiologists: sensitivity 84% versus 75%, AUC 0.92, and 68% of RT-PCR-positive cases with normal CT correctly identified.",
      "unknown": "Intended for use when CT scans and clinical history are both available; CT alone may have limited negative predictive value; prospective clinical use not evaluated; specificity not reported.",
      "ai_role": "Combined chest CT findings with clinical symptoms, exposure history and lab results to identify COVID-19-positive patients.",
      "evidence_urls": [
        "https://www.mountsinai.org/about/newsroom/2020/mount-sinai-first-in-us-to-use-artificial-intelligence-to-analyze-coronavirus-covid19-patients-pr",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-020-0931-3&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-05-amd-moorfields",
    "title": "An AI system predicts progression of macular degeneration",
    "summary": "The system analyzes OCT scans to predict whether the less affected eye of patients with macular degeneration will develop the wet form, outperforming specialists in the comparison described. The announcement gives no numerical metrics; preventive clinical studies were still to be planned.",
    "field": "medicine",
    "ai_technology": "Deep learning on OCT images",
    "organization": "Moorfields Eye Hospital",
    "source_url": "https://www.moorfields.nhs.uk/about-us/news-and-blogs/news/ai-to-develop-an-early-warning-system-for-amd",
    "source_type": "press release",
    "publication_date": "2020-05-18",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-020-0867-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Retrospective prediction of whether the second eye converts to wet AMD within 6 months: 80% sensitivity at 55% specificity, and better than five of six experts in the comparison described.",
      "unknown": "Preventive use needs prospective trials that were only being considered; 56% of individual eyes were false positives at the high-sensitivity setting; dataset size and generalizability not stated.",
      "ai_role": "Models built on 3D OCT scans and automatically generated tissue maps to predict conversion to wet AMD.",
      "evidence_urls": [
        "https://www.moorfields.nhs.uk/about-us/news-and-blogs/news/ai-to-develop-an-early-warning-system-for-amd",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-020-0867-7&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-05-elettrocatalizzatori-co2",
    "title": "Cu–Al electrocatalysts for CO2 found with active learning",
    "summary": "DFT and active machine learning identified Cu–Al catalysts that reduce CO2 to ethylene with over 80% Faradaic efficiency at 400 mA/cm² (about 66% for pure copper) and 55% half-cell energy efficiency at 150 mA/cm². Synthesized and tested in the lab; the abstract does not state how many candidates were screened.",
    "field": "chemistry",
    "ai_technology": "Active machine learning and DFT",
    "organization": "University of Toronto",
    "source_url": "https://www.nature.com/articles/s41586-020-2242-8",
    "source_type": "paper",
    "publication_date": "2020-05-13",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Cu-Al catalysts synthesized and tested in the lab: over 80% Faradaic efficiency for ethylene at 400 mA/cm2 versus about 66% for copper, and 55% half-cell energy efficiency at 150 mA/cm2; in situ X-ray absorption examined the copper environment.",
      "unknown": "The abstract does not state how many candidates were screened or list explicit limitations; full text not read.",
      "ai_role": "DFT calculations combined with active machine learning identified the Cu-Al catalysts.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-020-2242-8&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-04-slac-regolazione",
    "title": "LCLS magnet tuning with machine learning",
    "summary": "A physics-informed Gaussian process using historical data tunes the accelerator’s 24 LCLS quadrupole magnets. The authors estimate tuning is about three to five times faster, with beams tending to be more intense; this is a result on the real machine, limited to this part of the tuning process.",
    "field": "physics",
    "ai_technology": "Gaussian-process optimization",
    "organization": "SLAC National Accelerator Laboratory",
    "source_url": "https://www6.slac.stanford.edu/news/2020-04-29-new-machine-learning-method-streamlines-particle-accelerator-operations",
    "source_type": "press release",
    "publication_date": "2020-04-29",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevLett.124.124801",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Applied to the LCLS X-ray free-electron laser; the abstract reports the Bayesian optimizer as sample-efficient and outperforming existing optimizers, and the press release estimates tuning three to five times faster.",
      "unknown": "Speed and beam-quality gains are estimates; the method covers only part of the tuning process; results from three other accelerators are not reported.",
      "ai_role": "Gaussian process model predicted the effect of magnet adjustments, chose which to try, and was trained on past machine data.",
      "evidence_urls": [
        "https://www6.slac.stanford.edu/news/2020-04-29-new-machine-learning-method-streamlines-particle-accelerator-operations",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1103/PhysRevLett.124.124801&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-04-vetri-reti-grafo",
    "title": "Graph networks to understand glass dynamics",
    "summary": "The networks predict particle mobility in simulated glasses; perturbing the network reveals growing spatial correlations as the glass transition is approached. Evidence comes from simulations: the learned correlation length does not translate directly into a measurable physical quantity.",
    "field": "physics",
    "ai_technology": "Graph neural networks",
    "organization": "DeepMind",
    "source_url": "https://deepmind.google/blog/towards-understanding-glasses-with-graph-neural-networks/",
    "source_type": "blog",
    "publication_date": "2020-04-06",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41567-020-0842-8",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Not reported",
      "tested": "Networks trained on simulated glasses predicted particle mobility and were checked against simulation ground truth, outperforming physics-inspired and state-of-the-art baselines (96% correlation at short timescales, 64% at relaxation time).",
      "unknown": "The network's correlation length cannot be directly converted into a measurable physical quantity; it remains an open debate; the work models simulated glasses, not physical samples.",
      "ai_role": "Graph neural networks predicted particle mobility and probed spatial correlations by perturbing particles in shells around a marked particle.",
      "evidence_urls": [
        "https://deepmind.google/blog/towards-understanding-glasses-with-graph-neural-networks/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-03-metnet",
    "title": "MetNet: neural rainfall forecasting up to 8 hours ahead",
    "summary": "The model forecasts precipitation at 1 km resolution and 2-minute intervals up to 8 hours ahead from radar and satellite, without physical equations. According to the page, it outperforms NOAA’s operational HRRR model up to about 7–8 hours, in seconds rather than about an hour. US coverage only, author-only comparisons, a preprint, and no stated operational use.",
    "field": "climate",
    "ai_technology": "MetNet (neural network)",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/a-neural-weather-model-for-eight-hour-precipitation-forecasting/",
    "source_type": "blog",
    "publication_date": "2020-03-25",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2003.12140",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Evaluated on a precipitation-rate forecasting benchmark against NWP (HRRR) and an optical-flow baseline, scored by F1 at 1.0 mm/h; outperforms NWP at forecasts up to about 7 to 8 hours.",
      "unknown": "Demonstrated only for the continental US; the authors call it a small stepping stone; comparisons are author-run.",
      "ai_role": "Deep neural network with axial self-attention forecasting precipitation probabilities from radar and satellite inputs, without physical equations.",
      "evidence_urls": [
        "https://research.google/blog/a-neural-weather-model-for-eight-hour-precipitation-forecasting/",
        "https://arxiv.org/abs/2003.12140"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-03-deepscribe",
    "title": "DeepScribe: reading cuneiform signs on Persepolis tablets",
    "summary": "A machine-learning model recognizes signs on tablets from the Persepolis Fortification Archive with about 80% accuracy on tablets excluded from training (over 6,000 annotated images and over 100,000 signs in the database). About 20% of signs are wrong; testing is internal only, and the source is a university announcement.",
    "field": "humanities",
    "ai_technology": "DeepScribe (machine learning on images)",
    "organization": "University of Chicago",
    "source_url": "https://news.uchicago.edu/story/how-ai-could-help-translate-written-language-ancient-civilizations",
    "source_type": "press release",
    "publication_date": "2020-03-11",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Recognition of signs on tablets excluded from training reached about 80% accuracy; the source describes the testing as internal only.",
      "unknown": "About 20% of signs are still wrong; the size of the held-out set and an error breakdown are not given; other uses are described as aspirations.",
      "ai_role": "Machine-learning model trained on over 100,000 annotated signs from Persepolis Fortification Archive tablet images to recognize cuneiform signs.",
      "evidence_urls": [
        "https://news.uchicago.edu/story/how-ai-could-help-translate-written-language-ancient-civilizations"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-03-alphafold-covid",
    "title": "Predicted structures of SARS-CoV-2 proteins",
    "summary": "DeepMind openly shared AlphaFold predictions, without waiting for review, for six poorly studied viral proteins (membrane, ORF3a, Nsp2, Nsp4, Nsp6, and the C-terminal domain of papain-like protease). They were not experimentally verified; an ORF3a experimental structure deposited in the PDB in June 2020 matched the second predicted model well.",
    "field": "biology",
    "ai_technology": "AlphaFold",
    "organization": "Google DeepMind",
    "source_url": "https://deepmind.google/blog/computational-predictions-of-protein-structures-associated-with-covid-19/",
    "source_type": "blog",
    "publication_date": "2020-03-04",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Predictions compared with experimental structures: the spike prediction matched the PDB structure, and the second-ranked ORF3a model agreed well with later experimental work.",
      "unknown": "Predictions were not experimentally verified overall; the system was still in development and accuracy is not certain; released before peer review; the top ORF3a model misplaced some transmembrane helices.",
      "ai_role": "AlphaFold free-modelling predicted structures for six understudied SARS-CoV-2 proteins, with per-residue confidence scores.",
      "evidence_urls": [
        "https://deepmind.google/blog/computational-predictions-of-protein-structures-associated-with-covid-19/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2020-02-halicin",
    "title": "Halicin: deep learning identifies an antibiotic",
    "summary": "A model trained on about 2,500 molecules screened the roughly 6,000 compounds in the Broad Institute’s Drug Repurposing Hub and identified halicin, which killed many resistant bacteria in the lab and cleared A. baumannii infections in mice within 24 hours. Cell-culture and mouse results, not human studies; the mechanism remains preliminary.",
    "field": "medicine",
    "ai_technology": "Graph neural network for predicting antibacterial activity",
    "organization": "MIT",
    "source_url": "https://news.mit.edu/2020/artificial-intelligence-identifies-new-antibiotic-0220",
    "source_type": "press release",
    "publication_date": "2020-02-20",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2020.01.021",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Bactericidal activity against a wide range of pathogens, including M. tuberculosis and carbapenem-resistant Enterobacteriaceae; halicin treated C. difficile and pan-resistant A. baumannii infections in mice. Of 23 tested predictions from over 107 million ZINC15 molecules, eight were antibacterial.",
      "unknown": "Human safety and efficacy are not reported in the abstract.",
      "ai_role": "A deep neural network predicted antibacterial activity across chemical libraries; candidates were then tested experimentally.",
      "evidence_urls": [
        "https://doi.org/10.1016/j.cell.2020.01.021"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Original institutional announcement checked for the specific results, AI contribution and limitations cited in the antibiotic dossier or mathematical follow-up pilot. Full-paper reanalysis, external replication and an exhaustive follow-up search were not performed."
    },
    "follow_up": [
      {
        "id": "abaucin-program-extension",
        "kind": "same-team extension",
        "date": "2023-05-25",
        "date_precision": "day",
        "note": "MIT describes abaucin as a subsequent application of the antibiotic-screening program to A. baumannii. This is a different candidate and target-specific study, not replication of halicin.",
        "source_urls": [
          "https://news.mit.edu/2023/using-ai-scientists-combat-drug-resistant-infections-0525"
        ],
        "recorded_date": "2026-11-16"
      },
      {
        "id": "generative-design-program-extension",
        "kind": "same-team extension",
        "date": "2025-08-14",
        "date_precision": "day",
        "note": "MIT identifies halicin and abaucin as earlier screening results motivating generative design of new NG1/DN1 candidates. The later study does not establish clinical efficacy of these earlier candidates.",
        "source_urls": [
          "https://news.mit.edu/2025/using-generative-ai-researchers-design-compounds-kill-drug-resistant-bacteria-0814"
        ],
        "recorded_date": "2026-11-16"
      }
    ],
    "follow_up_review": {
      "date": "2026-11-16",
      "scope": "Bounded check of the original announcements cited by the events recorded here, including explicit links to later program work. This is not an exhaustive literature search or ongoing monitoring commitment."
    }
  },
  {
    "id": "2020-02-ricarica-rapida",
    "title": "Closed-loop optimization of fast-charging battery protocols",
    "summary": "A model estimating cycle life from the first 100 cycles, paired with Bayesian optimization, assessed 224 fast-charging protocols in 16 days; the authors estimate exhaustive search without early prediction would take over 500 days. It optimizes protocols, not new materials; only the abstract was reviewed.",
    "field": "materials",
    "ai_technology": "Bayesian optimization with early prediction",
    "organization": "Stanford University; MIT, and Toyota Research Institute",
    "source_url": "https://www.nature.com/articles/s41586-020-1994-5",
    "source_type": "paper",
    "publication_date": "2020-02-19",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "224 fast-charging protocols tested experimentally by measuring battery cycle life; high-cycle-life protocols identified in 16 days, versus an estimated 500+ days for exhaustive search without early prediction.",
      "unknown": "The abstract lists no explicit limitations; generalization to other problems is suggested but not discussed; only the abstract was read.",
      "ai_role": "An early-prediction model estimated final cycle life from the first cycles, and Bayesian optimization chose which protocols to test next.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-020-1994-5&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-01-dsp-1181",
    "title": "DSP-1181: an AI-created drug candidate enters Phase I",
    "summary": "Sumitomo Dainippon Pharma and Exscientia announced the start in Japan of a Phase I study of DSP-1181, a 5-HT1A serotonin receptor agonist created with the Centaur Chemist platform, with obsessive-compulsive disorder as the first indication. The companies said exploratory research took under 12 months versus a typical 4.5 years. The announcement gives no efficacy results. In the January 31, 2022 investor-call transcript, Sumitomo said development was discontinued because the candidate did not meet the expected Phase I criterion.",
    "field": "medicine",
    "ai_technology": "Centaur Chemist (molecule-design platform)",
    "organization": "Sumitomo Dainippon Pharma; and Exscientia",
    "source_url": "https://www.sumitomo-pharma.com/news/20200130.html",
    "source_type": "press release",
    "publication_date": "2020-01-30",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 1,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Not reported",
      "unknown": "Not reported",
      "ai_role": "Candidate created in joint research with Exscientia using its Centaur Chemist AI platform; the page does not describe what the AI did beyond design.",
      "evidence_urls": [
        "https://www.sumitomo-pharma.com/news/20200130.html"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2020-01-screening-seno",
    "title": "An AI breast-cancer screening system evaluated in two countries",
    "summary": "Among 25,856 women in the UK and 3,097 in the US, the system reduced false positives by 5.7% (US) and 1.2% (UK), and false negatives by 9.4% and 2.7% in absolute terms compared with radiologists. Trained only on UK data, it also outperformed specialists on US data. Retrospective evaluation (Nature): prospective studies are needed, and clinical deployment is unclear.",
    "field": "medicine",
    "ai_technology": "Neural network for mammograms",
    "organization": "Google DeepMind; Google Health, Cancer Research UK Imperial Centre, and Northwestern University",
    "source_url": "https://deepmind.google/blog/international-evaluation-of-an-ai-system-for-breast-cancer-screening/",
    "source_type": "blog",
    "publication_date": "2020-01-01",
    "added_date": "2026-11-11",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-019-1799-6",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Retrospective evaluation on UK (25,856 women) and US (3,097 women) screening mammograms, compared with individual radiologists' decisions; a reader study and a simulated double-reading workflow were also run.",
      "unknown": "The authors state prospective studies are required and that how to deploy the system clinically is not yet determined; the US data were enriched; the second-reader scenario was only simulated.",
      "ai_role": "Predicted from screening mammograms alone whether a biopsy would later show cancer.",
      "evidence_urls": [
        "https://deepmind.google/blog/international-evaluation-of-an-ai-system-for-breast-cancer-screening/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-019-1799-6&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-09-delphi-2m",
    "title": "Delphi-2M forecasts the risk of more than 1,000 diseases",
    "summary": "The September 2025 Nature version reports a GPT-style transformer trained on about 0.4 million UK Biobank participants and evaluated without retraining on 1.9 million Danish individuals. It models rates for more than 1,000 diseases and generates potential health trajectories up to 20 years ahead. This is retrospective risk modelling, with training-population biases; it does not establish a clinical benefit. The work also has a medRxiv preprint first indexed in June 2024.",
    "field": "medicine",
    "ai_technology": "Delphi-2M (GPT-style generative transformer)",
    "organization": "German Cancer Research Centre DKFZ; EMBL-EBI; University of Copenhagen",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12589094/",
    "source_type": "paper",
    "publication_date": "2025-09-17",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09529-3",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Existing-record predictions in UK Biobank and Danish registries.",
      "unknown": "Not assessed: supplementary analyses and clinical deployment; population biases are discussed in the paper.",
      "ai_role": "A modified generative transformer models disease events and timing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12589094/",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12657216/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-04-serine-hydrolases",
    "title": "Serine hydrolases designed from scratch",
    "summary": "The April 2025 Science version describes RFdiffusion enzyme backbones and PLACER assessment of catalytic-state ensembles. Laboratory characterization found catalytic efficiency up to 2.2 × 10^5 M^-1 s^-1, crystal structures with C-alpha RMSD below 1 angstrom, and catalysts in five folds different from natural serine hydrolases. This summary follows the abstract and figure descriptions; an earlier bioRxiv version was posted in August 2024.",
    "field": "biology",
    "ai_technology": "RFdiffusion and PLACER",
    "organization": "University of Washington; University of California, Los Angeles; Howard Hughes Medical Institute",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39946508/",
    "source_type": "paper",
    "publication_date": "2025-04-18",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adu2454",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Enzyme activity assays and crystallographic comparison of designed catalysts.",
      "unknown": "Not assessed: complete design success rates and supplementary data.",
      "ai_role": "RFdiffusion generated backbones; PLACER assessed catalytic-state geometry.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39946508/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-bindcraft",
    "title": "BindCraft designs protein binders in one shot",
    "summary": "An open AlphaFold2-based hallucination pipeline designs binders with reported experimental success rates of 10-100% across 12 targets, with affinities predominantly in the nanomolar range. The authors also demonstrated effects in cell assays and patient-derived samples; these are experimental results, not established clinical treatments. This entry uses the Nature publication date (27 August 2025); the preprint appeared in 2024.",
    "field": "biology",
    "ai_technology": "BindCraft (AlphaFold2-based design)",
    "organization": "EPFL; Swiss Institute of Bioinformatics",
    "source_url": "https://www.nature.com/articles/s41586-025-09429-6",
    "source_type": "paper",
    "publication_date": "2025-08-27",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09429-6",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Binders tested experimentally across 12 targets (receptors, allergens, de novo proteins, nucleases) with reported success rates of 10-100%, plus functional assays in patient-derived samples and cell assays.",
      "unknown": "The abstract gives no number of designs per target or per-target success rates, and states no limitations.",
      "ai_role": "Open-source automated pipeline using AlphaFold2 weights to design binders in one shot, without high-throughput screening.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-09429-6&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-drugclip",
    "title": "DrugCLIP screens about 10,000 human proteins against 500 million compounds",
    "summary": "The January 2026 Science abstract reports DrugCLIP screening roughly 10,000 human proteins against 500 million compounds. Experimental hits included 15% for the norepinephrine transporter, with structures of two inhibitor complexes, and 17.5% for TRIP12 using AlphaFold2 structures. Most database results remain computational. The work also has a bioRxiv preprint first indexed in September 2024.",
    "field": "chemistry",
    "ai_technology": "DrugCLIP (deep contrastive learning)",
    "organization": "Tsinghua University; Peking University; Beijing Academy of Artificial Intelligence",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41505557/",
    "source_type": "paper",
    "publication_date": "2026-01-08",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.ads9530",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Biochemical hit testing and structures of two NET inhibitor complexes.",
      "unknown": "Not assessed: full assay denominators, uncertainty and genome-wide experimental coverage.",
      "ai_role": "Contrastive learning ranked protein-compound pairs.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41505557/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-smrtnet",
    "title": "SMRTnet predicts small molecules that bind RNA",
    "summary": "The January 2026 Nature Biotechnology abstract reports 40 small-molecule hits across ten disease-associated RNA targets, with nanomolar-to-micromolar binding. One predicted compound reduced MYC expression, inhibited proliferation and promoted apoptosis in three cancer cell lines. SMRTnet combines language models with convolutional and graph attention networks, using RNA secondary structure. These are laboratory results.",
    "field": "biology",
    "ai_technology": "SMRTnet (language-model-based)",
    "organization": "Tsinghua University; Peking University; Southwest United Graduate School",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41482542/",
    "source_type": "paper",
    "publication_date": "2026-01-02",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-025-02942-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "RNA-binding assays and cancer-cell experiments.",
      "unknown": "Not assessed: full-paper validation rates and therapeutic performance.",
      "ai_role": "Multimodal deep learning predicted small molecule-RNA interactions.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41482542/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-04-urania-detectors",
    "title": "Urania designs new gravitational-wave detectors",
    "summary": "The April 2025 Physical Review X article describes Urania, a computational search for gravitational-wave interferometer topologies. Its simulated solutions increase potentially observable volume by up to 50-fold under the study constraints; 50 designs are shared in a detector zoo. These are proposed designs rather than constructed detectors. The work first appeared as an arXiv preprint in December 2023.",
    "field": "physics",
    "ai_technology": "Urania (continuous optimization with ML-inspired search)",
    "organization": "Max Planck Institute for the Science of Light; California Institute of Technology",
    "source_url": "https://journals.aps.org/prx/abstract/10.1103/PhysRevX.15.021012",
    "source_type": "paper",
    "publication_date": "2025-04-11",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevX.15.021012",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Simulated performance checks of newly designed interferometer topologies under experimental constraints.",
      "unknown": "Not assessed: detailed noise and hardware feasibility assumptions in the full article.",
      "ai_role": "Machine-learning-inspired continuous optimization searched detector configurations.",
      "evidence_urls": [
        "https://journals.aps.org/prx/abstract/10.1103/PhysRevX.15.021012",
        "https://arxiv.org/abs/2312.04258"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-erdos-728",
    "title": "Erdős Problem #728 solved with GPT-5.2 Pro and Aristotle",
    "summary": "GPT-5.2 Pro produced the argument and Harmonic's Aristotle a Lean proof of infinitely many triples with a logarithmic gap in a factorial divisibility condition. The arXiv writeup calls it resolved autonomously, which is the author's claim; no independent review is stated.",
    "field": "mathematics",
    "ai_technology": "GPT-5.2 Pro and Aristotle (Lean)",
    "organization": "OpenAI; and Harmonic",
    "source_url": "https://arxiv.org/abs/2601.07421",
    "source_type": "paper",
    "publication_date": "2026-01-12",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The argument produced by GPT-5.2 Pro was formalized as a Lean proof by Harmonic's Aristotle; the proof is machine-checked, and the abstract describes no independent human review.",
      "unknown": "The abstract states no limitations and describes only the formal Lean proof as verification.",
      "ai_role": "GPT-5.2 Pro produced the argument; Aristotle produced the Lean proof, which the author translated into informal mathematics.",
      "evidence_urls": [
        "https://arxiv.org/abs/2601.07421"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-03-alphaevolve-ramsey",
    "title": "AlphaEvolve raises nine Ramsey lower bounds",
    "summary": "AlphaEvolve-generated search algorithms improved bounds such as R(3,13) from 60 to 61, R(4,16) from 170 to 174 and R(4,19) from 213 to 219. The paper does not describe independent verification; search code is public. arXiv preprint, latest version read.",
    "field": "mathematics",
    "ai_technology": "AlphaEvolve",
    "organization": "Google DeepMind; and academic co-authors",
    "source_url": "https://arxiv.org/abs/2603.09172",
    "source_type": "paper",
    "publication_date": "2026-03-10",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Search algorithms evolved by AlphaEvolve improved nine Ramsey lower bounds; the authors report recovering bounds for all known exact Ramsey numbers and matching best known bounds in many other cases. No independent verification is described.",
      "unknown": "Not reported",
      "ai_role": "LLM-based code mutation agent that produced the search algorithms behind all results.",
      "evidence_urls": [
        "https://arxiv.org/abs/2603.09172"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-06-herculaneum-pherc-1667",
    "title": "A Herculaneum scroll read end to end without opening",
    "summary": "Machine-learning ink detection on synchrotron scans let papyrologists read about 22 columns of a likely Stoic ethics text from scroll PHerc. 1667. The text is fragmentary because only the inner core survives, and humans transcribed and reviewed every reading.",
    "field": "humanities",
    "ai_technology": "Machine-learning ink detection with virtual unwrapping",
    "organization": "Vesuvius Challenge; EduceLab and ESRF",
    "source_url": "https://scrollprize.org/firstscroll",
    "source_type": "other",
    "publication_date": "2026-06-25",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2606.29085",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Readings checked against explicit coverage and papyrological-review criteria; ink recovery validated independently on another scroll (PHerc. Paris 4).",
      "unknown": "The text is fragmentary because only the inner core survives; systematic recovery across the library is described as future work.",
      "ai_role": "Machine learning used for ink detection on synchrotron scans, alongside improved computational unrolling.",
      "evidence_urls": [
        "https://arxiv.org/abs/2606.29085"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2024-11-evolvepro",
    "title": "EVOLVEpro improves proteins up to 100-fold with few experiments",
    "summary": "For the January 2025 journal version indexed by PubMed and Europe PMC, the Science abstract reports that EVOLVEpro combines protein language models, regression and experimental feedback to improve six proteins used in RNA production, genome editing and antibody binding, with gains up to 100-fold in desired properties. A July 2024 preprint described five proteins. The date identifies the indexed journal version, not the first public announcement.",
    "field": "biology",
    "ai_technology": "EVOLVEpro (protein language model with active learning)",
    "organization": "Mass General Brigham; Beth Israel Deaconess Medical Center; Harvard Medical School; Massachusetts Institute of Technology",
    "source_url": "https://doi.org/10.1126/science.adr6006",
    "source_type": "paper",
    "publication_date": "2025-01-24",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/39571002/",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimental engineering of proteins for RNA production, genome editing and antibody binding.",
      "unknown": "Not assessed: full-paper per-protein data and generalization beyond the studied proteins.",
      "ai_role": "Active learning used protein-language-model representations and regression to select variants.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39571002/",
        "https://news.massgeneralbrigham.org/en/mgb-bidmc-unveil-ai-protein-engineer"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-autonomous-enzyme-engineering",
    "title": "A closed-loop platform engineers enzymes with minimal human input",
    "summary": "The July 2025 Nature Communications paper reports AI-guided biofoundry engineering of two enzymes in four rounds over four weeks, testing fewer than 500 variants per enzyme. Different AtHMT variants achieved about 90-fold greater ethyl-iodide substrate preference and 16-fold ethyltransferase activity; YmPhytase activity improved 26-fold at neutral pH. The system combines ML variant selection, language-model access and automated experiments; the authors describe conditional-to-high automation.",
    "field": "biology",
    "ai_technology": "Machine learning and LLM-guided variant design with biofoundry automation",
    "organization": "University of Illinois Urbana-Champaign",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12215622/",
    "source_type": "paper",
    "publication_date": "2025-07-01",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-61209-y",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Enzyme activity and purified-variant kinetic measurements in two campaigns.",
      "unknown": "Not assessed: supplements; broad applicability beyond two enzymes. Authors classify automation between levels 3 and 4.",
      "ai_role": "Protein models and regression selected variants; the biofoundry built and tested them.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12215622/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-llm-rdf",
    "title": "GPT-4 agents guide chemical synthesis development and purification",
    "summary": "The November 2024 Nature Communications paper describes six GPT-4 agents guiding copper/TEMPO alcohol oxidation, from literature search through optimization and purification. An agent-assisted scale-up produced 915 mg of product at 86% isolated yield and above 98% purity. Additional reaction tasks were demonstrated. The workflow involved chemists: thin-layer chromatography was performed manually, so this was not fully autonomous synthesis.",
    "field": "chemistry",
    "ai_technology": "LLM-RDF (six GPT-4 agents)",
    "organization": "Zhejiang University; Hong Kong University of Science and Technology; Suzhou Novartis Technical Development Co. Ltd.; Rezubio Pharmaceuticals Co. Ltd.",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11585555/",
    "source_type": "paper",
    "publication_date": "2024-11-23",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-54457-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Physical synthesis, optimization, scale-up and product purification.",
      "unknown": "Not assessed: supplements; manual TLC and chemist participation are explicit in the paper.",
      "ai_role": "Six GPT-4 agents supported literature, design, execution, analysis, separation and interpretation.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11585555/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-gauss-strong-pnt",
    "title": "Gauss formalizes the Strong Prime Number Theorem in Lean",
    "summary": "Math Inc reports that Gauss produced most statements and proofs for a Lean formalization of the strong Prime Number Theorem, with about 25,000 lines and 1,100 theorems/definitions in three weeks. Humans supplied mathematical scaffolding and reviewed strategies, and some files adapt prior work. The date follows the original project blueprint dated 11 September 2025, rather than an inferred announcement timestamp. This verifies a formalization of an existing theorem, not a new theorem.",
    "field": "mathematics",
    "ai_technology": "Gauss (Lean autoformalization agent)",
    "organization": "Math Inc",
    "source_url": "https://math-inc.github.io/strongpnt/",
    "source_type": "other",
    "publication_date": "2025-09-11",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "A published Lean formalization of the encoded theorem; this review did not compile it.",
      "unknown": "The original dated blueprint and published project/verification context were checked; the Lean project was not rebuilt in this review.",
      "ai_role": "Gauss generated most statements and proofs with human scaffolding and reuse of prior formalization.",
      "evidence_urls": [
        "https://math-inc.github.io/strongpnt/",
        "https://github.com/math-inc/strongpnt",
        "https://www.math.inc/gauss",
        "https://math-inc.github.io/strongpnt/blueprint.pdf"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-05-polaris-perovskite-nanoplatelets",
    "title": "PoLARIS: a self-driving microfluidic lab finds brighter lead-free perovskite nanoplatelets",
    "summary": "A closed-loop microfluidic platform ran 120 experiments in one 12-hour campaign on double perovskite nanoplatelets with up to six elements and identified the brightest lead-free composition found in that campaign. The press release gives no numerical photoluminescence gain, so the improvement is not quantified in the sources checked. Validation is the authors' own experimental measurements, and the paper should be checked for the brightness figures.",
    "field": "materials",
    "ai_technology": "PoLARIS (microfluidic self-driving lab with closed-loop experiment selection)",
    "organization": "North Carolina State University; Brown University",
    "source_url": "https://www.eurekalert.org/news-releases/1126928",
    "source_type": "press release",
    "publication_date": "2026-05-04",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-72765-2",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Lab experiments in one 12-hour campaign of 120 runs identified the brightest lead-free composition found in that campaign; the abstract does not report photoluminescence measurements.",
      "unknown": "The press release gives no numerical brightness gain; the abstract does not state limitations.",
      "ai_role": "Machine learning guided closed-loop experiment selection on a modular microfluidic reactor.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-026-72765-2&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-cof-2000-lfast",
    "title": "LFAST: an LLM agent drives robotic synthesis of the new COF-2000 framework",
    "summary": "A GPT-4o-based agent (LFAST) screened reaction conditions over a 96-condition matrix and refined them after each round, raising the crystallinity index of a reference covalent organic framework by 350% and yielding COF-2000, a previously unreported material with higher structural order. The reported validation is crystallinity measurement of the synthesized products; the authors do not pursue applications. Publication date is taken from secondary reports of the JACS paper and should be checked against the journal page.",
    "field": "chemistry",
    "ai_technology": "LFAST (LLM For Accelerated Synthesis Technique), GPT-4o agent with robotic experimentation",
    "organization": "University of California, Berkeley (Yaghi group)",
    "source_url": "https://bidmap.berkeley.edu/news/new-ai-agent-paradigm-shift-cof-synthesis",
    "source_type": "press release",
    "publication_date": "2026-02-23",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/jacs.5c23233",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Products of LFAST-guided conditions characterized by powder X-ray diffraction: crystallinity index raised by 350% on the benchmark COF TpPa-SO3H; COF-2000 characterized by PXRD and crystallinity index only.",
      "unknown": "Evidence rests on one benchmark system and one new COF; the full text was not read, so limitations discussed there are unknown.",
      "ai_role": "LLM deep research agent mined the literature for synthesis conditions; an automated platform ran them and PXRD analyzed the products.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1021/jacs.5c23233&resultType=core&format=json",
        "https://bidmap.berkeley.edu/news/new-ai-agent-paradigm-shift-cof-synthesis"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-qiushi-optical-bilinear-interaction",
    "title": "Qiushi Discovery Engine proposes and validates an optical bilinear interaction",
    "summary": "An LLM agent (Qiushi Discovery Engine) given only the topic optical computing for AI ran an open-ended study of 145.9 million tokens, 3,242 LLM calls and 1,242 tool calls, and proposed an optical bilinear interaction analogous to the attention operation in Transformers, which the authors say they experimentally validated. The abstract gives no accuracy figure or number of experimental runs. This is an arXiv preprint and the validation is reported by the authors themselves.",
    "field": "physics",
    "ai_technology": "Qiushi Discovery Engine (dual-layer LLM agent with Meta-Trace memory)",
    "organization": "Zhejiang University",
    "source_url": "https://arxiv.org/abs/2604.27092",
    "source_type": "paper",
    "publication_date": "2026-04-29",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The abstract says the proposed optical bilinear interaction was experimentally validated on a real optical platform; it gives no measurement details, trial counts or statistics.",
      "unknown": "The abstract gives no validation figures and hedges novelty with 'to our knowledge'; hardware use is framed only as a possible route.",
      "ai_role": "An LLM-based agentic system (Qiushi Discovery Engine) ran an open-ended study (3,242 LLM calls, 1,242 tool calls, 163 research notes, 44 scripts) and proposed the mechanism.",
      "evidence_urls": [
        "https://arxiv.org/abs/2604.27092"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-spinel-conductors-agentic-lab",
    "title": "An LLM agent in an air-free robotic lab maps lithium halide spinel conductors",
    "summary": "An April 2026 arXiv preprint reports 352 lithium-halide spinel samples, covering72% of 171 metal pairs drawn from 19 metals. The proportion meeting conductivity above 0.05 mS/cm and high spinel purity rose from 1.33% in the first 75 agent-proposed samples to 5.33% in the final 75. LLM agents selected experiments; characterization used X-ray diffraction and impedance spectroscopy with manual handling steps. These are within-study laboratory results.",
    "field": "materials",
    "ai_technology": "A-Lab GPSS (LLM agent integrated into a robotic synthesis platform)",
    "organization": "Lawrence Berkeley National Laboratory; University of California, Berkeley; Argonne National Laboratory; Korea Advanced Institute of Science and Technology; Florida State University",
    "source_url": "https://arxiv.org/abs/2604.11957",
    "source_type": "paper",
    "publication_date": "2026-04-13",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Robotic synthesis with semi-automated XRD and impedance measurements.",
      "unknown": "Not assessed: supplements and scalability outside this campaign. Manual sample handling remains in the characterization workflow.",
      "ai_role": "LLM agents used abductive and inductive reasoning for experimental design.",
      "evidence_urls": [
        "https://arxiv.org/abs/2604.11957",
        "https://arxiv.org/html/2604.11957v1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-collaborative-agents-quantum-spins",
    "title": "Collaborating LLM agents propose and test a calibration rule for diamond quantum spins",
    "summary": "Mutually reviewing LLM agents selected a question about the minimum number of calibration settings needed to predict responses to microwave pulse sequences in diamond spin ensembles, derived the theory and tested it experimentally. The first measurement was inconclusive because of signal drift, and after the protocol was revised the result matched the predictions under unchanged acceptance criteria. The abstract reports no numerical values, and this is an unrefereed arXiv preprint with an authors-reported validation.",
    "field": "physics",
    "ai_technology": "Collaborative LLM agents that review each other's work",
    "organization": "Institute of Science Tokyo",
    "source_url": "https://arxiv.org/abs/2610.40589",
    "source_type": "paper",
    "publication_date": "2026-10-29",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The agents derived the calibration rule computationally and ran diamond spin-ensemble experiments designed to distinguish predictions; the revised measurement matched the predictions under unchanged acceptance criteria.",
      "unknown": "The abstract states no explicit limitations; the main caveat reported is signal drift in the first, inconclusive experiment.",
      "ai_role": "Mutually reviewing LLM agents generated and rejected research questions, derived theory, and designed and revised experiments.",
      "evidence_urls": [
        "https://arxiv.org/abs/2610.40589"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-10-eve-variant-effect",
    "title": "EVE: evolutionary model predicts disease-variant pathogenicity without labels",
    "summary": "EVE, an unsupervised deep generative model trained on evolutionary sequence variation across about 140,000 species, predicted pathogenicity for more than 36 million variants in 3,219 disease genes and classified over 256,000 variants of unknown significance. The authors report it outperforms methods trained on labelled disease data and performs on par with high-throughput experimental predictions. Caveat: the authors state EVE is not a diagnostic test.",
    "field": "biology",
    "ai_technology": "EVE (deep generative evolutionary model, variational autoencoder)",
    "organization": "University of Oxford; Harvard Medical School",
    "source_url": "https://www.nature.com/articles/s41586-021-04043-8",
    "source_type": "paper",
    "publication_date": "2021-10-27",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-021-04043-8",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "EVE's predictions were compared with methods trained on labelled disease data and with high-throughput experimental predictions.",
      "unknown": "The authors state EVE is not a diagnostic test.",
      "ai_role": "An unsupervised deep generative model trained on evolutionary sequence variation across about 140,000 species, predicting pathogenicity for over 36 million variants.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-021-04043-8&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-10-akita-3d-genome",
    "title": "Akita: predicting 3D genome folding from DNA sequence",
    "summary": "Akita, a convolutional network, predicts chromatin contact maps for about 1 Mb of DNA sequence. On held-out data it reaches a Pearson R of 0.61 and Spearman R of 0.56, and it learns an orientation-specific grammar for CTCF binding sites. Caveat: the predictions are computational; the paper applies them to in silico mutagenesis, eQTL interpretation and structural-variant prediction.",
    "field": "biology",
    "ai_technology": "Akita (convolutional neural network, Basenji-based trunk)",
    "organization": "UC San Francisco; Gladstone Institutes",
    "source_url": "https://www.nature.com/articles/s41592-020-0958-x",
    "source_type": "paper",
    "publication_date": "2020-10-12",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-020-0958-x",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Akita's contact-map predictions were evaluated on held-out data and applied in silico to mutagenesis, eQTL interpretation and structural-variant prediction.",
      "unknown": "Not reported in the abstract read; the predictions are computational and no experimental validation is described there.",
      "ai_role": "A convolutional network predicting chromatin contact maps from about 1 Mb of DNA sequence, which learned an orientation-specific CTCF binding grammar.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41592-020-0958-x&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-05-colabfold",
    "title": "ColabFold: free protein folding with AlphaFold2 or RoseTTAFold",
    "summary": "ColabFold pairs AlphaFold2 or RoseTTAFold with fast MMseqs2 homology search, running free on Google Colab. Beyond benchmark comparisons, the paper reports predicting structures for 1,762 Methanocaldococcus jannaschii proteins in 48 hours on one GPU and using predicted complexes to help determine the 120 MDa human nuclear-pore structure with cryo-electron microscopy. The proteome structures are computational predictions, not experimentally validated structures. The nuclear-pore application overlaps the separate nuclear-pore entry; this is a methodological contribution, not an additional independent biological finding.",
    "field": "biology",
    "ai_technology": "ColabFold (AlphaFold2, RoseTTAFold, MMseqs2)",
    "organization": "Max Planck Institute for Multidisciplinary Sciences; Seoul National University",
    "source_url": "https://www.nature.com/articles/s41592-022-01488-1",
    "source_type": "paper",
    "publication_date": "2022-05-30",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-022-01488-1",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Throughput was measured (close to 1,000 structures per day on one GPU) and predicted complexes were used alongside cryo-EM to help determine the human nuclear-pore structure.",
      "unknown": "Not reported in the abstract read.",
      "ai_role": "AlphaFold2 or RoseTTAFold, paired with MMseqs2 homology search, run free on Google Colaboratory.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41592-022-01488-1&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-07-protgpt2",
    "title": "ProtGPT2: a language model generating novel protein sequences",
    "summary": "ProtGPT2, a 738-million-parameter autoregressive transformer trained on UniRef50, generates de novo protein sequences in seconds. The authors report the generated sequences are distantly related to natural ones, sample unexplored regions of protein space, and match natural sequences in amino acid and disorder propensities. Caveat: the evaluation is computational (sequence searches and similarity networks), with no experimental validation in the sources checked.",
    "field": "biology",
    "ai_technology": "ProtGPT2 (transformer protein language model)",
    "organization": "Center for Genomic Regulation, Barcelona; Universitat Pompeu Fabra",
    "source_url": "https://www.nature.com/articles/s41467-022-32007-7",
    "source_type": "paper",
    "publication_date": "2022-07-27",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-022-32007-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Generated sequences were assessed computationally: composition and disorder propensities, sequence searches, and AlphaFold-predicted folds; no experiments are described in the abstract read.",
      "unknown": "Not reported in the abstract read; no experimental validation is described.",
      "ai_role": "ProtGPT2, a 738-million-parameter autoregressive transformer trained on UniRef50, generated de novo protein sequences.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-022-32007-7&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-08-wittmann-directed-evolution",
    "title": "Machine learning-assisted directed evolution with in silico combinatorial screening",
    "summary": "A path-independent machine-learning protocol screens full combinatorial protein libraries in silico, with training-set design that avoids near-zero-fitness variants. On a four-site epistatic fitness landscape the optimized protocol reached the global fitness maximum up to 81-fold more often than single-step greedy optimization. Caveat: validation is on a combinatorial fitness landscape dataset, not a new wet-lab evolution campaign.",
    "field": "biology",
    "ai_technology": "Machine-learning-assisted directed evolution (in silico combinatorial screening)",
    "organization": "California Institute of Technology",
    "source_url": "https://doi.org/10.1016/j.cels.2021.07.008",
    "source_type": "paper",
    "publication_date": "2021-08-19",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The ML-assisted directed evolution protocol was compared with single-step greedy optimization on a four-site epistatic fitness landscape.",
      "unknown": "Not reported in the abstract read.",
      "ai_role": "A path-independent machine-learning protocol that screens full combinatorial libraries in silico, with training-set design avoiding near-zero-fitness variants.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.cels.2021.07.008&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-01-cell2location",
    "title": "Cell2location: mapping fine-grained cell types in spatial transcriptomics",
    "summary": "Cell2location is a Bayesian model that resolves fine-grained cell types in spatial transcriptomics data by accounting for technical variation and sharing information across locations. Applied to three tissues, it revealed regional astrocyte subtypes in mouse brain, a rare pre-germinal center B cell population in human lymph node, and fine immune populations in human gut lymphoid follicles. Caveat: the reported results are biological findings from the method's application, with no single quantitative accuracy figure confirmed in the abstract.",
    "field": "biology",
    "ai_technology": "Cell2location (Bayesian deep probabilistic model)",
    "organization": "Helmholtz Munich; Wellcome Sanger Institute",
    "source_url": "https://www.nature.com/articles/s41587-021-01139-4",
    "source_type": "paper",
    "publication_date": "2022-01-13",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-021-01139-4",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Cell2location was assessed in three tissues (mouse brain, human lymph node, human gut) and reported to give higher sensitivity and resolution than existing tools.",
      "unknown": "The abstract names no conditions under which the claimed higher sensitivity holds and no limitations.",
      "ai_role": "A Bayesian model that resolves fine-grained cell types in spatial transcriptomics data.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41587-021-01139-4&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-08-archaeasins-antibiotics",
    "title": "Archaeasins: deep learning finds antibiotic candidates in the archaeal proteome",
    "summary": "The APEX 1.1 deep learning platform scanned 233 archaeal proteomes and flagged 12,623 candidate antimicrobial molecules. Of 80 synthesized candidates, 93% showed in vitro activity against key pathogens, and three selected compounds reduced a drug-resistant bacterium in animal models after a single dose. Candidate and synthesis counts differ between press coverage and secondary reports, and the mouse details come from press coverage rather than the full text, so the paper's own figures should be checked.",
    "field": "biology",
    "ai_technology": "APEX 1.1 deep learning antimicrobial peptide predictor",
    "organization": "University of Pennsylvania (de la Fuente lab)",
    "source_url": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12408343/",
    "source_type": "paper",
    "publication_date": "2025-08-12",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41564-025-02061-0",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "80 top-ranked APEX-predicted peptides were synthesized and tested by broth microdilution MIC assays, then in mouse skin-abscess and thigh-infection models (six mice per group).",
      "unknown": "Authors state the model is sequence-based without structural information, pseudo-MIC labels may add noise, screening used only reviewed UniProt sequences, the thigh model did not prevent infection, efficacy declined between days 2 and 4 in the skin model, and only female mice were used.",
      "ai_role": "APEX 1.1, a bacterial-strain-specific antimicrobial activity predictor, scored 193,331,608 peptides from the archaeome and selected candidates for synthesis.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12408343/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-modelangelo-cryoem-model-building",
    "title": "ModelAngelo: automated atomic model building in cryo-EM maps",
    "summary": "ModelAngelo combines cryo-EM density, protein sequence and local geometry to build atomic models. Beyond method benchmarks, the paper identifies six previously unassigned proteins in a photosynthetic supercomplex and four additional radial-spoke proteins plus two central-apparatus proteins in Chlamydomonas reinhardtii. These assignments use existing experimental cryo-EM maps and sequence searches; the paper leaves the newly assigned ciliary proteins' functional roles for further investigation.",
    "field": "biology",
    "ai_technology": "ModelAngelo graph neural network with hidden Markov model sequence search",
    "organization": "MRC Laboratory of Molecular Biology (Scheres group)",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38408488/",
    "source_type": "paper",
    "publication_date": "2024-02-26",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07215-4",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Models built by ModelAngelo were compared with those of human experts, and its protein identification was compared with human experts on proteins of unknown sequence.",
      "unknown": "Not reported in the abstract read; the nucleotide result is limited to backbone accuracy.",
      "ai_role": "A graph neural network combining cryo-EM density, protein sequence and local geometry to build atomic models and identify proteins.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-024-07215-4&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-12-rfdiffusion2-enzyme-active-sites",
    "title": "RFdiffusion2 scaffolds enzyme active sites at atomic level",
    "summary": "RFdiffusion2 scaffolded all 41 active sites in a diverse benchmark, compared with 16 for previous methods. Designs were tested experimentally, with fewer than 96 sequences needed per case to find active enzymes across three catalytic mechanisms. The journal abstract's three-mechanism count conflicts with a five-reaction claim in the Institute for Protein Design announcement, so the abstract should be treated as authoritative.",
    "field": "biology",
    "ai_technology": "RFdiffusion2",
    "organization": "University of Washington Institute for Protein Design (Baker lab)",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12791007/",
    "source_type": "paper",
    "publication_date": "2025-12-03",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-025-02975-x",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Designed enzymes were expressed and screened experimentally for retroaldolase, cysteine hydrolase and zinc hydrolase activity, with kinetic constants measured for the best variants.",
      "unknown": "Authors state the designs are not as active as native enzymes, the theozymes may miss interactions needed for high activity, and scaffolding catalytic atoms does not guarantee proficient enzymes.",
      "ai_role": "RFdiffusion2 generated scaffolds from atom-level active-site descriptions, LigandMPNN assigned sequences, and structure predictors filtered designs before testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12791007/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-computational-metallohydrolases",
    "title": "Computational design of zinc metallohydrolases with RFdiffusion2",
    "summary": "RFdiffusion2 designed zinc metallohydrolases from quantum-chemistry-derived active-site geometries. The first round of 96 designs produced an enzyme with catalytic efficiency of 16,000 M-1 s-1, and a second round of 96 yielded three more highly active enzymes, with k/K up to 53,000 M-1 s-1. Crystal structures (PDB 9PYJ, 9PYL) were determined, but the evidence covers one designed enzyme family.",
    "field": "biology",
    "ai_technology": "RFdiffusion2 with quantum-chemistry active-site models",
    "organization": "University of Washington Institute for Protein Design (Baker lab)",
    "source_url": "https://www.rcsb.org/structure/9PYJ",
    "source_type": "other",
    "publication_date": "2025-12-03",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09746-w",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "From 96 designs, the best had kcat/KM of 16,000 M^-1 s^-1; a second round of 96 gave three more active enzymes, up to 53,000 M^-1 s^-1 and kcat up to 1.5 s^-1. The crystal structure of the most active design is very close to the model.",
      "unknown": "Use in practical applications is not reported in the abstract.",
      "ai_role": "RFdiffusion2 designed zinc metallohydrolases from quantum-chemistry active sites; PLACER and Chai-1 analysed the designs.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41586-025-09746-w"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-riff-diff-enzyme-design",
    "title": "Riff-Diff: computational enzyme design by catalytic motif scaffolding",
    "summary": "The December 2025 Nature version reports Riff-Diff designs for retro-aldol and Morita-Baylis-Hillman reactions, with activities comparable to enzymes optimized by experimental evolution. Six high-resolution structures support near-atomic active-site placement. The abstract describes a hybrid of machine learning and atomistic modelling. The work also has a bioRxiv preprint first indexed in August 2024; the January 2026 journal issue is later than online publication.",
    "field": "biology",
    "ai_technology": "Riff-Diff (RFdiffusion with rotamer-based motif placement)",
    "organization": "Graz University of Technology; University of Graz; Weizmann Institute of Science; BioTechMed Graz",
    "source_url": "https://doi.org/10.1038/s41586-025-09747-9",
    "source_type": "paper",
    "publication_date": "2025-12-03",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Designed-enzyme activity and six experimental high-resolution structures.",
      "unknown": "Not assessed: full-paper success rates, thermal assays and supplementary data.",
      "ai_role": "Riff-Diff combines learned protein generation and atomistic catalytic-motif scaffolding.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41586-025-09747-9%22&resultType=core&format=json",
        "https://www.tugraz.at/en/news/article/enzyme-design-ai-riff-diff"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-designed-generalist-binder-catalysis",
    "title": "Fragment-guided AI protein redesign yields a Kemp eliminase and fluorophore binder",
    "summary": "Crystallographic fragment screening of a designed binding protein guided redesign into a fluorophore binder and a Kemp eliminase. LigandMPNN and structure-prediction models supported the design workflow; crystal structures and enzyme assays checked the resulting functions. The paper reports catalytic efficiency of 3.2 million M⁻¹ s⁻¹ for the eliminase. This entry tracks the 4 May 2026 journal version; a preprint preceded it.",
    "field": "biology",
    "ai_technology": "LigandMPNN with Rosetta relaxation for sequence design; RaptorX and later AlphaFold 3 for structure checking",
    "organization": "University of California, San Francisco (Fraser and DeGrado groups)",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13245809/",
    "source_type": "paper",
    "publication_date": "2026-05-04",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41557-026-02125-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Fragment screening, crystallography and enzyme assays of redesigned proteins; reported Kemp eliminase catalytic efficiency 3,200,000 M⁻¹ s⁻¹.",
      "unknown": "The reported kinetics and structural tests are within the study; independent replication was not assessed. The journal version follows an earlier preprint.",
      "ai_role": "LigandMPNN with Rosetta relaxation designed enzyme sequences; RaptorX and later AlphaFold 3 predictions checked candidate structures.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13245809/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41557-026-02125-6%22&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-05-self-driving-thin-film",
    "title": "Self-driving laboratory optimizes organic hole-transport thin films",
    "summary": "A robotic laboratory used model-based optimization to select thin-film compositions and processing conditions, execute experiments and learn from their measured optical and electronic properties. The reported demonstration optimized hole mobility in organic hole-transport materials used in perovskite solar cells and electronics. This description is limited to the paper abstract; it does not assert a numerical mobility gain.",
    "field": "materials",
    "ai_technology": "Closed-loop self-driving laboratory with autonomous optimization",
    "organization": "University of British Columbia; Harvard University; University of Toronto",
    "source_url": "https://doi.org/10.1126/sciadv.aaz8867",
    "source_type": "paper",
    "publication_date": "2020-05-13",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Closed-loop composition and processing experiments on organic hole-transport thin films, with measured hole mobility as an optimization target.",
      "unknown": "Abstract and publication metadata reviewed; detailed optimization history, baseline comparisons and durability were not assessed.",
      "ai_role": "A model-based optimizer selected subsequent experiments using measurements from the robotic platform.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1126%2Fsciadv.aaz8867%22&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-06-generative-ai-onchip-lxr-agonists",
    "title": "Generative AI and on-chip synthesis produce new LXR agonists",
    "summary": "A recurrent neural network generated de novo LXR agonist designs, and 25 were synthesized in a microfluidic flow system. In vitro screening found 17 hits (68%), and 12 were confirmed as potent LXR agonists after batch resynthesis. Validation is in vitro only, with no in vivo testing, and the sources disagree on the number of designs retained for synthesis.",
    "field": "chemistry",
    "ai_technology": "LSTM recurrent neural network generator with transfer learning, coupled to on-chip flow synthesis",
    "organization": "ETH Zurich",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8195470/",
    "source_type": "paper",
    "publication_date": "2021-06-11",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.abg3338",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "25 synthesized designs were screened in reporter gene assays (17 hits); 12 were confirmed as LXR agonists after batch resynthesis, with EC50 0.18 to 4.5 micromolar for LXR-alpha. No in vivo testing was reported.",
      "unknown": "Chemical space was limited to one-step reactions, the primary screen used crude mixtures, two compounds were not confirmed, and the authors describe the work as proof-of-concept.",
      "ai_role": "A generative LSTM pretrained on 656,070 compounds and fine-tuned on 40 known LXR-alpha agonists, with a virtual reaction filter of 17 reaction schemes.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8195470/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-08-polymer-dielectrics-experimental-validation",
    "title": "Machine learning and experiments yield tolerant polymer dielectrics",
    "summary": "Machine-learning property models and phenomenological screening rules narrowed a list of more than 13,000 previously synthesized polymers to candidates for high-temperature dielectric applications. The authors report experimental testing of selected candidates and derive design rules for electric-field and temperature tolerance. Review is limited to the abstract and publication metadata.",
    "field": "materials",
    "ai_technology": "Machine learning and informatics models for polymer dielectric screening",
    "organization": "University of Connecticut; Georgia Institute of Technology",
    "source_url": "https://doi.org/10.1021/acsami.1c11885",
    "source_type": "paper",
    "publication_date": "2021-08-26",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Property-based screening of over 13,000 known polymers, followed by experimental validation of some selected dielectric candidates.",
      "unknown": "Review scope: abstract, online date and affiliations. Candidate-by-candidate property values and the full experimental sample count were not assessed.",
      "ai_role": "Machine-learning models predicted screening properties such as band gap and cohesive energy density.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1021%2Facsami.1c11885%22&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-09-element-selection-li-solid-electrolyte",
    "title": "Unsupervised machine learning guides discovery of a lithium solid electrolyte",
    "summary": "A variational autoencoder ranked unexplored element combinations for lithium solid electrolytes. Researchers combined those rankings with crystal-structure prediction and synthesis to discover Li₃.₃SnS₃.₃Cl₀.₇, then characterized its structure and lithium-ion transport. Impedance, NMR and lithium-cell measurements supported a new transport pathway and compatibility with lithium metal.",
    "field": "materials",
    "ai_technology": "Unsupervised machine learning for element selection with crystal structure prediction",
    "organization": "University of Liverpool",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8455628/",
    "source_type": "paper",
    "publication_date": "2021-09-21",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-021-25343-7",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Synthesis, diffraction, impedance and NMR characterization of the selected lithium solid electrolyte, with lithium plating/stripping measurements.",
      "unknown": "This is a laboratory material study; full-battery performance and independent replication were not assessed.",
      "ai_role": "An unsupervised variational autoencoder ranked chemical element combinations; human selection and crystal-structure calculations directed synthesis.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8455628/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41467-021-25343-7%22&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-10-high-entropy-invar-alloys",
    "title": "Machine learning and DFT discover high-entropy Invar alloys",
    "summary": "A closed-loop active-learning framework combining machine learning, density-functional theory and thermodynamic calculations screened millions of compositions. Of 17 new alloys processed and characterized, two high-entropy Invar alloys showed very low thermal expansion, about 2 x 10^-6 per kelvin at 300 K. Only the top-ranked candidates were synthesized, so the hit rate is small.",
    "field": "materials",
    "ai_technology": "Closed-loop active learning combining machine learning, density-functional theory and thermodynamic calculations",
    "organization": "Max Planck Institute for Iron Research; Delft University of Technology",
    "source_url": "https://doi.org/10.1126/science.abo4940",
    "source_type": "paper",
    "publication_date": "2022-10-06",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "17 new alloys were processed and characterized; two high-entropy Invar alloys showed very low thermal expansion of about 2 x 10^-6 per kelvin at 300 K.",
      "unknown": "The method works on very sparse data, and the abstract notes that thermodynamic design rules alone often fail in high-dimensional spaces; no further limits are stated.",
      "ai_role": "A closed-loop active-learning framework integrating machine learning with density-functional theory and thermodynamic calculations.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.abo4940&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-tohoku-cu-saa-agent",
    "title": "Catalysis AI Agent derives design rules for copper single-atom alloys for CO2 electroreduction",
    "summary": "An LLM-based Catalysis AI Agent identified an energy descriptor for classifying copper-based single-atom alloys and a structural descriptor for predicting activation energy. The team reports experimental testing and that the approach extends to other metal dopants. Evidence caveat: the press release gives no quantitative activity or selectivity values and does not name the experimental group, so validation is limited.",
    "field": "chemistry",
    "ai_technology": "Catalysis AI Agent (LLM-based agent)",
    "organization": "Tohoku University, Advanced Institute for Materials Research (AIMR)",
    "source_url": "https://www.wpi-aimr.tohoku.ac.jp/en/achievements/press/2026/20260326_002166.html",
    "source_type": "press release",
    "publication_date": "2026-03-26",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/anie.202524612",
    "relevance": {
      "impact": 3,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The press release says the approach was tested experimentally and extends to other metal dopants, but the experiments are not described.",
      "unknown": "Not reported; the release gives no quantitative activity or selectivity values and names no experimental group.",
      "ai_role": "An LLM-based Catalysis AI Agent trained on the DigCat database mined trends and identified an energy descriptor and a structural descriptor.",
      "evidence_urls": [
        "https://www.wpi-aimr.tohoku.ac.jp/en/achievements/press/2026/20260326_002166.html"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2026-03-chathea-hea-fuel-cell",
    "title": "ChatHEA assistant guides high-entropy alloy catalysts for fuel cells",
    "summary": "ChatHEA, a domain-specific LLM assistant for high-entropy alloy electrocatalysis, guided synthesis and screening of 100 five-element alloy catalysts. The FeCoCuPtIr catalyst reached a peak power density of 0.789 W cm-2 in a fuel cell and outperformed commercial Pt/C in electrochemical tests. These figures are reported by the institutional press release; its numeric activity values and DOE target are not given. The catalog date is the paper's 14 March 2026 publication date, stated in the 8 July press release; the full paper was not reviewed.",
    "field": "chemistry",
    "ai_technology": "ChatHEA (LLM-based domain assistant)",
    "organization": "Tohoku University, Advanced Institute for Materials Research (AIMR)",
    "source_url": "https://www.wpi-aimr.tohoku.ac.jp/en/achievements/press/2026/20260708_002238.html",
    "source_type": "press release",
    "publication_date": "2026-03-14",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/nsr/nwag161",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "100 five-element alloy catalysts were screened by high-throughput experimentation; FeCoCuPtIr outperformed commercial Pt/C in electrochemical tests and reached 0.789 W cm-2 peak power density in a fuel cell.",
      "unknown": "Not reported; the page gives no detailed durability data and lists no limitations.",
      "ai_role": "ChatHEA, a domain-specific LLM assistant, supported literature review, element design, experiment planning, data processing and activity analysis.",
      "evidence_urls": [
        "https://www.wpi-aimr.tohoku.ac.jp/en/achievements/press/2026/20260708_002238.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-cbnn-multimetallic-sac",
    "title": "Deep learning finds multimetallic single-atom catalysts across material families",
    "summary": "The Crossbreeding Neural Network (CBNN), trained jointly on carbon-supported single-atom catalysts and perovskite oxides, screened 8,008 candidates. Its top W-Mo-Ru-Rh design on a perovskite support outperformed the comparison catalysts in electrochemical tests, and the activity ranking of 12 tested catalysts matched the predictions. These counts come from the IBS institutional release; the paper abstract confirms the experimental result but does not quantify the activity gain. The catalog date is the Nature Materials publication date, 28 May 2026.",
    "field": "chemistry",
    "ai_technology": "CBNN (Crossbreeding Neural Network)",
    "organization": "Institute for Basic Science (IBS) Center for Nanoparticle Research; Seoul National University",
    "source_url": "https://www.ibs.re.kr/cop/bbs/BBSMSTR_000000000738/selectBoardArticle.do?nttId=26725&pageIndex=1&searchCnd=&searchWrd=",
    "source_type": "press release",
    "publication_date": "2026-05-28",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41563-026-02622-6",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The AI-predicted catalysts were synthesized and tested for alkaline oxygen evolution; the activity ranking of 12 tested catalysts matched the predictions.",
      "unknown": "The comparison is limited to catalysts synthesized in this study; use in batteries, energy storage and drug discovery is not demonstrated.",
      "ai_role": "A Crossbreeding Neural Network trained on carbon-supported single-atom catalysts and perovskite oxides screened 8,008 multimetallic candidates.",
      "evidence_urls": [
        "https://www.ibs.re.kr/cop/bbs/BBSMSTR_000000000738/selectBoardArticle.do?nttId=26725&pageIndex=1&searchCnd=&searchWrd="
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-aalto-kagome-superconductors",
    "title": "Machine learning pre-screening finds kagome superconductors YRu3B2 and LuRu3B2",
    "summary": "A machine-learning pre-screen of candidate compounds, followed by first-principles calculations, selected YRu3B2 and LuRu3B2, which were measured as bulk superconductors with Tc of 0.81 K and 0.95 K and near 100% superconducting volume fraction. Evidence caveat: these are low-temperature superconductors and the work is a proof-of-concept screening workflow, not a room-temperature result; LuRu3B2 was initially overlooked because of a phonon instability.",
    "field": "materials",
    "ai_technology": "Machine-learning pre-screening combined with first-principles calculations",
    "organization": "Aalto University; Rice University; Princeton University",
    "source_url": "https://www.aalto.fi/en/news/researchers-identify-new-superconductors-unlocking-process-that-could-yield-thousands-more",
    "source_type": "press release",
    "publication_date": "2026-06-17",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/lpqj-7hyg",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The Rice team synthesized YRu3B2 and LuRu3B2 and tested them to confirm superconductivity; the page gives no transition temperatures.",
      "unknown": "A proof-of-concept study; the page does not report operating temperature, and most viable-looking combinations are described as unusable.",
      "ai_role": "Machine learning narrowed elemental combinations, then a pre-screening algorithm and targeted calculations selected candidates.",
      "evidence_urls": [
        "https://www.aalto.fi/en/news/researchers-identify-new-superconductors-unlocking-process-that-could-yield-thousands-more"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-04-mofgen-mof-agentici",
    "title": "MOFGen: agentic AI generates MOF linkers and five AI-designed MOFs are synthesized",
    "summary": "MOFGen, a multi-agent system combining language models, diffusion models and quantum-mechanical checks, generated hundreds of thousands of novel MOF structures and synthesizable linkers. The authors report that five AI-designed MOFs were synthesized through high-throughput experiments. Evidence caveat: the abstract describes the validation only generally, the structure counts are computational, and this is an arXiv preprint not confirmed as peer-reviewed.",
    "field": "chemistry",
    "ai_technology": "MOFGen (multi-agent LLM system with diffusion model)",
    "organization": "UC Berkeley; Lawrence Berkeley National Laboratory; Google DeepMind",
    "source_url": "https://arxiv.org/abs/2504.14110",
    "source_type": "paper",
    "publication_date": "2025-04-18",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The abstract says five AI-designed MOFs were synthesized through high-throughput experiments; the validation is described only generally.",
      "unknown": "Generative models face challenges ensuring synthesizability; the abstract lists no other limitations.",
      "ai_role": "A multi-agent system: a language model proposed compositions, a diffusion model generated structures, quantum-mechanical agents filtered candidates, and feasibility agents applied rules and machine learning.",
      "evidence_urls": [
        "https://arxiv.org/abs/2504.14110"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-03-canopy-traits-tropical-forests",
    "title": "Canopy traits across tropical forests mapped with deep learning and satellite data",
    "summary": "Field data from more than 1,800 vegetation plots, combined with satellite, terrain, climate and soil data and convolutional neural networks, predicted variation in 13 tree functional traits. Tropical American forests show about 40% greater functional richness than African and Asian forests, and African forests show 32% higher functional divergence than American forests. The authors caution that accuracy depends on data quality and coverage, and that AI should support rather than replace field sampling.",
    "field": "biology",
    "ai_technology": "Convolutional neural networks combined with satellite spectral data",
    "organization": "University of Oxford; 119-author international team",
    "source_url": "https://www.nature.com/articles/s41586-025-08663-2",
    "source_type": "paper",
    "publication_date": "2025-03-05",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-08663-2",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Trait predictions from field data on more than 1,800 vegetation plots, combined with satellite, terrain, climate and soil data, were checked with an uncertainty analysis.",
      "unknown": "Accuracy depends on data quality and coverage, and further data collection would refine the maps.",
      "ai_role": "Not stated in the page read; the page does not mention artificial intelligence or machine learning.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-08663-2&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-03-wenhai-eddying-ocean",
    "title": "WenHai: a deep neural network forecasts the eddying ocean",
    "summary": "The WenHai data-driven global ocean forecast system outperforms a state-of-the-art eddy-resolving numerical model and an AI-based ocean forecast system for temperature and salinity profiles, sea surface temperature, sea level anomaly and near-surface currents at lead times from 1 day to at least 10 days. Skill beyond 10 days is not reported in the abstract, and the computational speed-up claimed in secondary coverage was not confirmed in the paper abstract.",
    "field": "climate",
    "ai_technology": "WenHai deep neural network ocean forecasting model",
    "organization": "Laoshan Laboratory; Ocean University of China",
    "source_url": "https://www.nature.com/articles/s41467-025-57389-2",
    "source_type": "paper",
    "publication_date": "2025-03-06",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-57389-2",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "WenHai forecasts were compared with a state-of-the-art eddy-resolving numerical system and an AI-based system for temperature, salinity, sea surface temperature, sea level anomaly and currents at 1 to at least 10 day leads.",
      "unknown": "Skill beyond 10 days is not reported in the page read, and the authors describe ocean AI forecasting as remaining challenging.",
      "ai_role": "A data-driven deep neural network with bulk flux formulae, with ocean dynamics informing the architecture.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-025-57389-2&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-05-samudra-ocean-emulator",
    "title": "Samudra: an AI emulator of the global ocean for climate simulations",
    "summary": "Samudra, a UNet-style emulator trained on the ocean component of a state-of-the-art climate model (OM4 within CM4), reproduces sea surface height, velocities, temperature and salinity across the full ocean depth. It runs about 150 times faster than the original ocean model and remains stable for centuries. The authors note it struggles to reproduce the correct magnitude of forcing trends while staying stable, so it needs further work.",
    "field": "climate",
    "ai_technology": "Samudra UNet-style ocean emulator",
    "organization": "Princeton University; and NOAA Geophysical Fluid Dynamics Laboratory",
    "source_url": "https://doi.org/10.1029/2024GL114318",
    "source_type": "paper",
    "publication_date": "2025-05-24",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2412.03795",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Samudra was compared with the original climate model's ocean component: no drift relative to the truth, reproduction of depth structure and interannual variability, stability for centuries, and about 150 times faster.",
      "unknown": "The authors state the emulator struggles to reproduce the correct magnitude of forcing trends while staying stable, and this needs further work.",
      "ai_role": "A ConvNeXt-UNet-style emulator trained on the ocean component of a climate model to predict sea surface height, velocities, temperature and salinity.",
      "evidence_urls": [
        "https://arxiv.org/abs/2412.03795"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-11-aftershock-rate-deep-learning",
    "title": "Deep learning forecasts aftershock rates in seconds, comparable to ETAS",
    "summary": "Neural networks trained on earthquake catalogues from California, New Zealand, Italy, Japan and Greece forecast the number of aftershocks in the following 24 hours after magnitude 4+ earthquakes. Their quality is reported as comparable to the operational ETAS model, while producing results in seconds rather than hours or days. The abstract reports no numeric skill scores, so the comparison rests on the full paper.",
    "field": "climate",
    "ai_technology": "Deep neural networks for spatiotemporal seismicity rate forecasting, benchmarked against ETAS",
    "organization": "University of Edinburgh; British Geological Survey and University of Padua",
    "source_url": "https://earth-planets-space.springeropen.com/articles/10.1186/s40623-025-02241-6",
    "source_type": "paper",
    "publication_date": "2025-11-25",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s40623-025-02241-6",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Deep-learning seismicity forecasts were compared with an epidemic-type aftershock sequence benchmark and a persistence null hypothesis on Southern California and Italy test datasets; they matched ETAS at much lower processing time.",
      "unknown": "The abstract says performance depends on training data that represent the properties of earthquake sequences.",
      "ai_role": "Deep neural networks learned short-term spatiotemporal seismicity rates from high-resolution catalogues.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s40623-025-02241-6"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-04-sentinel2-methane-super-emitters",
    "title": "Deep transfer learning detects methane super-emitters in oil and gas fields from Sentinel-2",
    "summary": "A two-part framework (adaptive artifact removal plus a deep subdomain adaptation network) was trained on 1,627 Sentinel-2 images from six known methane super-emitters. Applied to an unannotated oil field in Algeria, it flagged 34 anomalous emission events over one year, attributed to three super-emitters linked to production and transmission infrastructure. DSAN achieved an average macro F1 of 0.86 on transfer tasks, versus 0.78 for the EfficientNet-V2L comparator; independent field verification is not reported in the abstract.",
    "field": "climate",
    "ai_technology": "Deep subdomain adaptation network (DSAN) with low-reflectance artifact detection (LRAD)",
    "organization": "Westlake University; Zhejiang University; Harvard University",
    "source_url": "https://acp.copernicus.org/articles/25/4035/2025/",
    "source_type": "paper",
    "publication_date": "2025-04-09",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.5194/acp-25-4035-2025",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The model was tested on 36 cross-region transfer tasks (macro F1 0.86 for 1-to-1, 0.89 for 5-to-1) and 33 of 369 positive detections were manually verified as true methane plumes.",
      "unknown": "Real-world macro F1 was 0.56 because of 336 false positives, the dataset covers only three countries, label confidence is hard to quantify, the detection limit is roughly 5000 kg/h, and the abstract and body differ on the event count (34 vs 33).",
      "ai_role": "A ResNet-50 deep subdomain adaptation network classified plume presence in place of human inspection, after a band-threshold artifact-removal step.",
      "evidence_urls": [
        "https://acp.copernicus.org/articles/25/4035/2025/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-02-gpcr-exoframe-modulators",
    "title": "De novo designed proteins modulate GPCR signalling in the dopamine D1 receptor",
    "summary": "Researchers designed four protein exoframe modulators targeting the transmembrane region of the dopamine D1 receptor. Structural studies and functional assays showed agonist-positive, negative and biased allosteric modulation; an agonist-positive modulator restored activity in loss-of-function receptor mutants. The journal abstract supports these receptor-level experiments, rather than a clinical treatment claim.",
    "field": "biology",
    "ai_technology": "Computational de novo protein design (AI-based protein design pipeline)",
    "organization": "Zhejiang University; Soochow University; Hangzhou Medical College",
    "source_url": "https://doi.org/10.1038/s41586-025-09957-1",
    "source_type": "paper",
    "publication_date": "2026-02-16",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Structural and functional investigation of four designed modulators of the D1 receptor, including rescue of loss-of-function mutants.",
      "unknown": "Review limited to abstract, online date and affiliations. Therapeutic efficacy in patients is not established by these receptor experiments.",
      "ai_role": "A deep-learning, hallucination-like protein-design workflow used structural prompts to generate transmembrane binding modes.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41586-025-09957-1%22&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-designed-protein-small-molecule-sensing",
    "title": "AI-designed protein family binds and senses small molecules, including cortisol",
    "summary": "The study combines trRosetta-based scaffold hallucination and Rosetta design with ProteinMPNN/LigandMPNN sequence design and AlphaFold structural filtering to create NTF2-like binders for six small-molecule targets. An optimized cortisol binder had a measured dissociation constant of 68 nM. A purified-protein, split-luciferase cortisol sensor showed an estimated response EC50 of 72 nM; this is a sensor response measurement, not the binding-affinity value or a clinical detection limit. The sensor also responded to cortisone and 17-alpha-hydroxyprogesterone, so specificity needs improvement. The March 2026 peer-reviewed paper is the source of record.",
    "field": "biology",
    "ai_technology": "trRosetta-based hallucination; Rosetta; ProteinMPNN/LigandMPNN sequence design; AlphaFold structural filtering",
    "organization": "KAIST; University of Washington Institute for Protein Design",
    "source_url": "https://doi.org/10.1038/s41467-026-70953-8",
    "source_type": "paper",
    "publication_date": "2026-03-28",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Designed binders for six small-molecule targets were measured for binding affinity (nanomolar to low micromolar), and a cortisol binder was used to build a biosensor.",
      "unknown": "The abstract states no specific limitations.",
      "ai_role": "Deep learning combined with physics-based methods generated a protein family with varied designable pocket geometries.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-026-70953-8&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-adaptive-experiment-planning-catalysts",
    "title": "Adaptive AI experiment planning finds multi-promoter catalysts in under 50 experiments",
    "summary": "Fritz Haber Institute, with BASF and BasCat, used adaptive experiment planning in a self-driving lab to search over 10^13 promoter combinations with fewer than 50 experiments. The method was validated on propane-to-propylene conversion, and the authors report synergistic promoter effects that explain the catalyst's performance. The claim of outperforming the industry reference is not quantified in the release, so check the paper for numbers.",
    "field": "chemistry",
    "ai_technology": "Adaptive experiment planning (Bayesian-style, gray-box) in a self-driving lab",
    "organization": "Fritz Haber Institute of the Max Planck Society; BASF; BasCat (UniCat BASF JointLab)",
    "source_url": "https://www.fhi.mpg.de/2217409/2026-03-24_ACS_Catal_SDLs",
    "source_type": "press release",
    "publication_date": "2026-03-19",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acscatal.6c00286",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "The method was validated on propane-to-propylene conversion, and the release says it identified a catalyst superior to the industry reference, without figures.",
      "unknown": "No quantitative results are given and the release states no explicit limitations.",
      "ai_role": "Adaptive AI planned experiments in a self-driving laboratory, using a gray-box strategy that also explained the effects of individual promoters.",
      "evidence_urls": [
        "https://www.fhi.mpg.de/2217409/2026-03-24_ACS_Catal_SDLs"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-hard-metallic-glasses-vibann",
    "title": "Machine learning designs bulk metallic glasses with peak hardness near 2,450 HV",
    "summary": "VIBANN used a dataset of 673 bulk metallic-glass compositions to guide the design of five alloys, synthesized as amorphous 2 mm rods. The paper's experimental results identify alloy A1, B68Nb24Fe4W4, as the hardest: 2,447 +/- 44 HV at a 0.5 N indentation load, with six replicate indentations. Hardness decreases with increasing load; the peak is not a value for every alloy or test condition. X-ray scattering and molecular dynamics support a structural interpretation within the studied composition range. The date refers to the May 2026 paper, rather than the institute's later August announcement.",
    "field": "materials",
    "ai_technology": "VIBANN (variational information-bottleneck attention network)",
    "organization": "Max Planck Institute for Sustainable Materials",
    "source_url": "https://www.nature.com/articles/s41467-026-73008-0",
    "source_type": "paper",
    "publication_date": "2026-05-12",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-73008-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Designed bulk metallic glasses reached a Vickers hardness of about 2450 HV, compared with other alloys under comparable conditions.",
      "unknown": "Discovery is constrained by the vastness of composition space and scarce validated data, and results apply under limited-data conditions.",
      "ai_role": "An uncertainty-aware latent-representation framework searched compositions, screened by chemical plausibility, novelty and predictive uncertainty.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-026-73008-0&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-patternboost-constructions",
    "title": "PatternBoost: constructions in mathematics with a little help from AI",
    "summary": "PatternBoost alternates a local search that generates candidate constructions with a transformer trained on the best ones. The authors report the best known solutions to several extremal combinatorics problems, including a counterexample to a conjecture that had been open for 30 years. Candidates are scored mechanically, so each construction is checked by computer; the abstract does not name the conjecture and the paper was not peer-reviewed at the time of posting.",
    "field": "mathematics",
    "ai_technology": "PatternBoost (transformer trained on search outputs, combined with local search)",
    "organization": "Meta AI; University of Wisconsin-Madison; Google DeepMind; University of Sydney",
    "source_url": "https://arxiv.org/abs/2411.00566",
    "source_type": "paper",
    "publication_date": "2024-11-01",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/pdf/2411.00566v1",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The abstract reports best known solutions to several extremal combinatorics problems, including a counterexample to a 30-year-old conjecture, but describes no verification procedure.",
      "unknown": "The abstract states that PatternBoost's performance varies across problems.",
      "ai_role": "A local search generated candidate constructions, and a transformer trained on the best candidates seeded the next search round.",
      "evidence_urls": [
        "https://arxiv.org/abs/2411.00566"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-10-adrs-systems-research-openevolve",
    "title": "Barbarians at the Gate: AI-driven research for systems (ADRS)",
    "summary": "An LLM-driven generate-evaluate-refine loop built on OpenEvolve produced systems algorithms in case studies, with the abstract reporting up to 5.0x runtime improvements or 50% cost reductions over baselines. Solutions are checked with simulators or real systems, which is the validation basis. The results are on specific case studies rather than a general claim, and the work is a preprint.",
    "field": "computer science",
    "ai_technology": "OpenEvolve (LLM-driven evolutionary code search, ADRS framework)",
    "organization": "UC Berkeley",
    "source_url": "https://arxiv.org/abs/2510.06189",
    "source_type": "paper",
    "publication_date": "2025-10-07",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/pdf/2510.06189v1",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Generated systems algorithms were run against predefined workloads in real systems or simulators and compared with state-of-the-art human designs across case studies (cloud load balancing, MoE inference, LLM SQL queries, transaction scheduling). The abstract reports up to 5.0x runtime improvement or 50% cost reduction.",
      "unknown": "The method assumes a reliable verifier, which the authors present as a precondition; the page has no dedicated limitations section and the work is a preprint.",
      "ai_role": "An LLM-driven generate-evaluate-refine loop (penEvolve, an open-source ADRS instance) proposed the algorithms.",
      "evidence_urls": [
        "https://arxiv.org/abs/2510.06189"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-04-gauss-sphere-packing-dim8",
    "title": "Gauss completes the Lean formalization of the dimension-8 sphere packing proof",
    "summary": "Math Inc's autoformalization model Gauss completed the final stages of the Lean verification of Viazovska's proof that the E8 lattice is the optimal sphere packing in dimension 8, working with the human formalization team. The paper describes the human-AI collaboration and the verification in the Lean kernel, with the milestone dated February 2026 in the abstract. Formalization of an existing 2016 result, not a new theorem, and the paper is a progress report.",
    "field": "mathematics",
    "ai_technology": "Gauss (Math Inc autoformalization model) with Lean",
    "organization": "Math Inc; Hariharan, Viazovska and collaborators",
    "source_url": "https://arxiv.org/abs/2604.23468",
    "source_type": "paper",
    "publication_date": "2026-04-25",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The completed Lean formalization of Viazovska's dimension-8 sphere packing proof was verified in Lean, with the milestone dated February 2026 in the abstract.",
      "unknown": "The paper discusses project objectives that remain, which the abstract page does not list; no other limitations are stated.",
      "ai_role": "Math Inc.'s autoformalization model Gauss performed the final stages of the Lean verification, working with the human formalization team.",
      "evidence_urls": [
        "https://arxiv.org/abs/2604.23468"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-05-deepmind-erdos-oeis-formal-search",
    "title": "AI-driven formal proof search resolves Erdős problems and OEIS conjectures",
    "summary": "An agent that alternates LLM proof generation with Lean verification autonomously resolved 9 of 353 open Erdős problems and proved 44 of 492 OEIS conjectures, at a cost of a few hundred dollars per problem according to the abstract. Proofs are Lean-verified, so the checking is mechanical; the abstract does not name the system and the results are limited to the problem sets tested.",
    "field": "mathematics",
    "ai_technology": "LLM-based agent with Lean-based proof verification (system unnamed in abstract)",
    "organization": "Google DeepMind",
    "source_url": "https://arxiv.org/abs/2605.22763",
    "source_type": "paper",
    "publication_date": "2026-05-21",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/pdf/2605.22763",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Proofs generated by the agent were checked in Lean; the agent resolved 9 of 353 open Erdos problems and proved 44 of 492 OEIS conjectures, at a stated cost of a few hundred dollars per problem.",
      "unknown": "The abstract does not describe how the problem sets were selected, and notes the simpler agent cost more on the hardest problems.",
      "ai_role": "An agent alternates LLM proof generation with Lean verification to prove open problems autonomously.",
      "evidence_urls": [
        "https://arxiv.org/abs/2605.22763"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2025-04-amie-diagnostic-dialogue",
    "title": "AMIE: conversational diagnostic AI rated above primary care physicians in simulated consultations",
    "summary": "In a blinded remote OSCE with 149 case scenarios, AMIE and primary care physicians held text consultations with simulated patients. Specialist physicians rated AMIE superior on 30 of 32 axes and patient actors on 25 of 26, and non-inferior on the rest. The evidence comes from simulated scenarios, not real patients, and the authors state that further research is needed before real-world use.",
    "field": "medicine",
    "ai_technology": "AMIE (Articulate Medical Intelligence Explorer), a large language model agent for diagnostic dialogue",
    "organization": "Google Research; Google DeepMind",
    "source_url": "https://www.nature.com/articles/s41586-025-08866-7",
    "source_type": "paper",
    "publication_date": "2025-04-09",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-08866-7",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A randomized, double-blind crossover study of text consultations with patient-actors across 159 case scenarios compared AMIE with 20 primary care physicians, rated by specialist physicians and patient actors on several axes.",
      "unknown": "The authors note the unfamiliar synchronous text-chat format and state that further research is needed before real-world use.",
      "ai_role": "AMIE, an LLM-based diagnostic dialogue system trained by self-play with automated feedback, conducted the consultations.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-08866-7&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-01-composer-sepsis-ucsd",
    "title": "COMPOSER deep learning sepsis model linked to 17% lower in-hospital sepsis mortality",
    "summary": "Across two UC San Diego Health emergency departments, 6,217 adult septic patients were studied before and after nurse-facing alerts driven by the COMPOSER model. Deployment was associated with a 1.9-point absolute (17% relative) drop in in-hospital sepsis mortality (95% CI 0.3-3.5%) and a 5-point rise in bundle compliance. The design is quasi-experimental, not randomized, and several authors co-founded a company that commercializes the tools.",
    "field": "medicine",
    "ai_technology": "COMPOSER (deep learning sepsis prediction model)",
    "organization": "UC San Diego Health; University of California San Diego",
    "source_url": "https://www.nature.com/articles/s41746-023-00986-6",
    "source_type": "paper",
    "publication_date": "2024-01-23",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41746-023-00986-6",
    "relevance": {
      "impact": 5,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A quasi-experimental before-and-after comparison across two UC San Diego emergency departments (6,217 septic patients) measured in-hospital sepsis mortality and bundle compliance before and after COMPOSER-driven nurse alerts.",
      "unknown": "The design is quasi-experimental, not randomized; other limitations were not reported in the fetched abstract.",
      "ai_role": "The COMPOSER deep-learning sepsis model triggered nurse-facing alerts.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41746-023-00986-6&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-09-retfound-retina-foundation-model",
    "title": "RETFound: a foundation model for disease detection from retinal images",
    "summary": "RETFound is self-pretrained on unlabelled retinal images, reportedly about 1.6 million from Moorfields Eye Hospital, and then adapted to disease tasks. The authors report better diagnosis of sight-threatening eye diseases than comparison models, better prediction of systemic conditions such as heart failure and myocardial infarction, and strong performance with fewer labels. The evaluation is retrospective, with no clinical deployment study, and the model was released openly.",
    "field": "medicine",
    "ai_technology": "RETFound (masked-autoencoder foundation model for retinal images)",
    "organization": "Moorfields Eye Hospital; University College London",
    "source_url": "https://www.nature.com/articles/s41586-023-06555-x",
    "source_type": "paper",
    "publication_date": "2023-09-13",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06555-x",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "RETFound, self-pretrained on unlabelled retinal images, was adapted to disease tasks and compared with other models on diagnosis of sight-threatening eye diseases and on prediction of heart failure and myocardial infarction.",
      "unknown": "The abstract states no limitations of this work; the evaluation is retrospective with no clinical deployment study.",
      "ai_role": "A self-supervised foundation model pretrained on retinal images and fine-tuned for disease tasks.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-023-06555-x&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-11-panda-pancreatic-cancer-ct",
    "title": "PANDA: pancreatic lesion detection on non-contrast CT with deep learning",
    "summary": "PANDA was trained on non-contrast abdominal CT scans from 3,208 patients at a high-volume pancreatic cancer centre in Shanghai. In a retrospective evaluation of 20,530 consecutive patients across four clinical settings, the paper reports 92.9% sensitivity and 99.9% specificity for pancreatic lesion detection. The study also reports five cancers and 26 other pancreatic lesions missed by the initial standard of care. These results do not establish population-screening benefit; the authors call for prospective risk-benefit and cost-effectiveness studies and broader international validation.",
    "field": "medicine",
    "ai_technology": "PANDA (three-stage deep learning cascade on non-contrast CT)",
    "organization": "DAMO Academy, Alibaba Group; Shanghai Institute of Pancreatic Diseases; First Affiliated Hospital of Zhejiang University",
    "source_url": "https://www.nature.com/articles/s41591-023-02640-w",
    "source_type": "paper",
    "publication_date": "2023-11-20",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-023-02640-w",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a retrospective evaluation of 20,530 consecutive patients across four clinical settings, PANDA's sensitivity and specificity for pancreatic lesion detection were reported, along with missed cancers relative to the initial standard of care.",
      "unknown": "The results do not establish population-screening benefit; the authors call for prospective risk-benefit and cost-effectiveness studies and broader international validation.",
      "ai_role": "PANDA, a deep-learning model trained on non-contrast abdominal CT, detected pancreatic lesions.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-023-02640-w&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-07-virchow-pathology-foundation-model",
    "title": "Virchow: pan-cancer detection using a pathology foundation model",
    "summary": "Using Virchow pathology-model embeddings, a pan-cancer detector achieved a specimen-level AUC of 0.95 across nine common and seven rare cancer types. The held-out evaluation included 6,142 specimens from Memorial Sloan Kettering Cancer Center and external institutions. Separate comparisons approached the performance of tissue-specific clinical products and exceeded them on some rare variants. This is diagnostic validation on existing specimens, not evidence that the Virchow detector itself was deployed or improved patient outcomes.",
    "field": "medicine",
    "ai_technology": "Virchow (vision transformer foundation model for pathology)",
    "organization": "Paige; Microsoft Research; Memorial Sloan Kettering Cancer Center",
    "source_url": "https://www.nature.com/articles/s41591-024-03141-0",
    "source_type": "paper",
    "publication_date": "2024-07-22",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-024-03141-0",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A pan-cancer detector built on Virchow embeddings was evaluated on a held-out set of 6,142 specimens from Memorial Sloan Kettering and external institutions, and compared with tissue-specific clinical products.",
      "unknown": "The abstract does not establish deployment or patient-outcome benefit; no explicit limitations are stated.",
      "ai_role": "The Virchow pathology foundation model provided embeddings for the pan-cancer detector.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-024-03141-0&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-06-pi-cai-prostate-mri-ai",
    "title": "PI-CAI: AI matches or beats radiologists in prostate cancer detection on MRI",
    "summary": "In a reader study with 62 radiologists on 400 cases, the AI detected about 6.8% more clinically significant cancers at the same specificity and produced about 50% fewer false positives. On 1,000 routine-practice cases, however, non-inferiority was not confirmed (specificity 68.9% vs 69.0% at 96.1% sensitivity). The authors call for prospective validation, and the system is not available for clinical use.",
    "field": "medicine",
    "ai_technology": "PI-CAI challenge super-algorithm (ensemble of top five submitted deep learning models)",
    "organization": "PI-CAI consortium; Radboud University Medical Center",
    "source_url": "https://doi.org/10.1016/s1470-2045(24)00220-1",
    "source_type": "paper",
    "publication_date": "2024-06-11",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/38876123/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A multireader study with 62 radiologists on 400 paired cases compared AI and PI-RADS 2.1 reads, and 1,000 routine-practice cases were used to test non-inferiority against historical clinical readings.",
      "unknown": "Non-inferiority to routine practice was not confirmed (specificity 68.9% vs 69.0%); the authors state prospective validation is needed and the system is not available for clinical use.",
      "ai_role": "A deep-learning system detected clinically significant prostate cancer on MRI.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:38876123%20AND%20SRC:MED&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-echonext-structural-heart-disease",
    "title": "EchoNext: detecting structural heart disease from ECGs with AI",
    "summary": "EchoNext was trained on more than one million heart-rhythm and imaging records to detect structural heart disease. The journal abstract reports internal and external validation, a controlled comparison outperforming cardiologists, and prospective clinical-trial evaluation in patients without previous cardiac imaging that identified undiagnosed disease. This supports a screening study, not proof that deployment improves patient outcomes. The authors also released model weights and an annotated dataset.",
    "field": "medicine",
    "ai_technology": "EchoNext (deep learning model on 12-lead ECG)",
    "organization": "Columbia University; NewYork-Presbyterian Hospital",
    "source_url": "https://www.nature.com/articles/s41586-025-09227-0",
    "source_type": "paper",
    "publication_date": "2025-07-16",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09227-0",
    "relevance": {
      "impact": 5,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Internal/external diagnostic validation, a controlled cardiologist comparison and prospective clinical-trial screening of patients without previous cardiac imaging.",
      "unknown": "Abstract and publication metadata reviewed. No claim of mortality reduction, universal clinical generalization or independent replication is made.",
      "ai_role": "A deep-learning model estimated structural-heart-disease risk from electrocardiograms to guide imaging-based screening.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41586-025-09227-0%22&resultType=core&format=json"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2024-09-debunkbot-conspiracy",
    "title": "Conversations with an AI chatbot durably reduce conspiracy beliefs",
    "summary": "Across experiments with more than 2,000 participants, three dialogues with GPT-4 reduced belief in the targeted conspiracy theory by about 20%, and the effect persisted for two months and spilled over to unrelated conspiracies. A professional fact-checker rated 99.2% of 128 sampled chatbot claims as true. Caveats: lab setting, the chatbot did not reduce belief in true conspiracies, and most participants still held the hoax belief.",
    "field": "multidisciplinary",
    "ai_technology": "GPT-4 (DebunkBot dialogue system)",
    "organization": "MIT Sloan School of Management; Cornell University; American University",
    "source_url": "https://doi.org/10.1126/science.adq1814",
    "source_type": "paper",
    "publication_date": "2024-09-13",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/39264999/",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "More than 2,000 conspiracy believers held three personalized, evidence-based dialogues with GPT-4 Turbo, and belief change was measured, including persistence at two months and spillover to unrelated conspiracies.",
      "unknown": "The abstract does not describe a control group and states no limitations; the page metadata lists a 2026 expression of concern and erratum.",
      "ai_role": "GPT-4 Turbo delivered the personalized counterevidence in the dialogues.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.adq1814&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-06-instantaneous-voice-bci",
    "title": "Instantaneous voice-synthesis neuroprosthesis for a man with ALS",
    "summary": "A brain-computer interface decoding 256 microelectrodes in the ventral precentral gyrus synthesized the participant's intended voice causally, with a reported latency of about 10 ms per decoding step. Listeners understood almost 60% of synthesized words, compared with 4% when he was not using the device, and he could modulate intonation and sing short melodies. Caveat: a single participant with ALS, within the BrainGate2 trial, and the authors describe the technology as early stage.",
    "field": "medicine",
    "ai_technology": "Deep neural network speech decoder for intracortical recordings",
    "organization": "UC Davis; Stanford University; BrainGate consortium",
    "source_url": "https://doi.org/10.1038/s41586-025-09127-3",
    "source_type": "paper",
    "publication_date": "2025-06-12",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/40506548/",
    "relevance": {
      "impact": 5,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A single participant with ALS used a neural decoder on 256 microelectrodes with closed-loop audio feedback to produce synthesized voice in near real time, and could modulate intonation and sing short melodies.",
      "unknown": "A single participant within a trial; the abstract states no explicit limitations, and the authors describe the technology as early stage.",
      "ai_role": "A neural decoder translated attempted-speech activity into synthesized voice; the abstract does not say whether it uses machine learning.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-09127-3&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-centaur-human-cognition",
    "title": "Centaur: a foundation model that predicts human behaviour across psychology experiments",
    "summary": "Fine-tuning Llama 3.1 70B on Psych-101, a dataset of about 10 million human choices from 160 psychology experiments, produced a model that predicts and simulates human behaviour in experiments expressed in natural language, including tasks it had not seen. The authors report that its internal representations align better with human neural activity after fine-tuning. Caveat: a 2025 critique argues that its language comprehension is limited and that it fails to follow altered instructions.",
    "field": "multidisciplinary",
    "ai_technology": "Centaur (Llama 3.1 70B fine-tuned on Psych-101)",
    "organization": "Helmholtz Munich; Max Planck Institute for Biological Cybernetics; University of Tübingen",
    "source_url": "https://doi.org/10.1038/s41586-025-09215-4",
    "source_type": "paper",
    "publication_date": "2025-07-02",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/40604288/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Centaur, a language model fine-tuned on Psych-101, was evaluated on held-out participants and on generalization to unseen cover stories, modified task structures and new domains, and its representations were checked against human neural activity.",
      "unknown": "The abstract states no limitations; the authors present the case study as a demonstration.",
      "ai_role": "A large language model was the base, fine-tuned on about 10 million human choices from 160 experiments.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-09215-4&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-generative-agents-1000-people",
    "title": "Generative agents built from interview transcripts simulate 1,000 people",
    "summary": "LLM agents built from participants' full interview transcripts reproduced their General Social Survey answers at about 85% of the rate at which participants reproduced their own answers two weeks later, a normalized score. They also matched personality and economic-game results (normalized correlations of about 0.80 and 0.66) and replicated four of five behavioural experiments. Caveat: arXiv preprint, not peer reviewed, and the results are normalized rather than raw accuracy.",
    "field": "multidisciplinary",
    "ai_technology": "Generative agent architecture with LLM and full interview transcripts",
    "organization": "Stanford University; Google DeepMind; Northwestern University; University of Washington",
    "source_url": "https://arxiv.org/abs/2411.10109",
    "source_type": "paper",
    "publication_date": "2024-11-15",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Agents built from interview and survey data of 1,052 Americans were scored on held-out General Social Survey items against participants' own two-week test-retest consistency, and on personality, economic-game and experiment replications.",
      "unknown": "Preprint, not peer reviewed; there is no dedicated limitations section, and combining interviews and surveys gave only modest gains over either source.",
      "ai_role": "An LLM was the engine of agents that answer questions on behalf of individuals, grounded in their self-report data.",
      "evidence_urls": [
        "https://arxiv.org/abs/2411.10109"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-03-angkor-reservoirs-deep-learning",
    "title": "Deep learning maps Angkorian-period reservoirs beyond Greater Angkor",
    "summary": "A DeepLabV3+ segmentation model trained on about 3,600 high-quality samples from over 11,000 candidates mapped Angkor-period reservoirs in satellite imagery across the wider Angkorian realm. The best F1 score reached 45% in favourable regions, comparable to a trained human interpreter. The authors present the model as a second opinion for landscape archaeologists that requires human validation.",
    "field": "humanities",
    "ai_technology": "DeepLabV3+ semantic segmentation",
    "organization": "Leiden University; University of Toronto; University of Colorado Boulder",
    "source_url": "https://doi.org/10.1371/journal.pone.0320452",
    "source_type": "paper",
    "publication_date": "2025-03-26",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/40138322/",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A DeepLabV3+ model trained on about 3,600 samples mapped Angkor-period reservoirs in satellite imagery and was compared with a trained human interpreter; the best F1 score was 45% in favourable regions.",
      "unknown": "The abstract states no limitations; the authors present the model as a second opinion that requires human validation.",
      "ai_role": "DeepLabV3+ semantic segmentation detected Angkor-period reservoirs in satellite imagery.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1371/journal.pone.0320452&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-11-mind-captioning-fmri",
    "title": "Mind captioning: text descriptions of viewed videos decoded from fMRI",
    "summary": "In six participants, a language-model pipeline translated fMRI responses to viewed videos into descriptions of their semantic content. Video identification reached about 50% among 100 candidates, compared with 1% chance. Results also extended to instructed video recall and persisted when the canonical language network was excluded, supporting a shared semantic representation beyond that network. The experiment used extensive participant-specific training; it was not unrestricted reading of private thoughts or a clinical aphasia trial.",
    "field": "medicine",
    "ai_technology": "Deep language model semantic-feature decoder with text generation search",
    "organization": "NTT Communication Science Laboratories",
    "source_url": "https://doi.org/10.1126/sciadv.adw1464",
    "source_type": "paper",
    "publication_date": "2025-11-05",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Video viewing and instructed recall in six participants; 2,180 unique videos in training and comparisons with analyses excluding the canonical language network.",
      "unknown": "Small participant group and extensive per-person training. Clinical communication benefits and arbitrary private-thought decoding were not tested.",
      "ai_role": "Linear decoders mapped fMRI signals to deep-language-model semantic features; a language model guided text optimization.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12588295/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1126%2Fsciadv.adw1464%22&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-12-levers-political-persuasion-ai",
    "title": "The levers of political persuasion with conversational AI",
    "summary": "In three large experiments with 76,977 participants, 19 LLMs argued 707 British political positions in dialogue. Persuasiveness was driven mainly by information density, while tailoring to user traits added little, and post-training for persuasion raised effectiveness by up to 50%. The most persuasive models also produced the most inaccurate claims. Caveat: the persuasion effects are measured in survey experiments, and accuracy was tested on a subset of claims.",
    "field": "multidisciplinary",
    "ai_technology": "19 open and closed LLMs (including GPT-4.5, GPT-4o, Grok-3)",
    "organization": "University of Oxford; UK AI Security Institute; London School of Economics; Stanford University; MIT",
    "source_url": "https://doi.org/10.1126/science.aea3884",
    "source_type": "paper",
    "publication_date": "2025-12-04",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/41343633/",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Three large experiments with 76,977 participants tested 19 LLMs arguing 707 British political positions in dialogue, and the factual accuracy of claims was checked on a subset.",
      "unknown": "Persuasion effects are measured in survey experiments, and accuracy was tested only on a subset of claims.",
      "ai_role": "LLMs, some post-trained for persuasion, were the persuaders in dialogue.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.aea3884&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-raven-tess-planets",
    "title": "RAVEN pipeline validates 118 new TESS planets and vets over 2,000 candidates",
    "summary": "A pipeline combining box least squares detection with machine-learning classifiers, trained on realistic simulations, searched 2.2 million TESS main-sequence stars in full-frame images (sectors 1 to 55). It newly validates 118 planets, 31 of them newly detected, and vets over 2,000 candidates, about 1,000 of them new. The machine-learning classifier is a validation tool rather than the detection method, and the paper is a preprint accepted for MNRAS; the peer-reviewed publication date was not confirmed.",
    "field": "physics",
    "ai_technology": "RAVEN: box least squares detection with machine-learning vetting classifiers trained on simulations",
    "organization": "Warwick University; and collaborators",
    "source_url": "https://arxiv.org/abs/2603.22597",
    "source_type": "paper",
    "publication_date": "2026-03-23",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Candidates from 2.2 million TESS main-sequence stars were classified by machine-learning models trained on realistic simulations and then statistically validated, yielding 118 validated planets.",
      "unknown": "Over 2,000 candidates remain unvalidated and may be false positives; the peer-reviewed publication date was not confirmed.",
      "ai_role": "Machine-learning classifiers separated planets from false positives; they are a validation tool, not the detection method.",
      "evidence_urls": [
        "https://arxiv.org/abs/2603.22597"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-dusty-plasma-forces",
    "title": "Physics-tailored machine learning reveals unexpected non-reciprocal forces in dusty plasmas",
    "summary": "A neural network built with physical constraints was trained on 3D particle-tracking data from a laboratory dusty plasma to infer the interaction forces between particles. The method describes the non-reciprocal forces more precisely than before and shows that some common theoretical assumptions about them are inaccurate. The authors checked their inferred force laws against laboratory experiments, which strengthens the claim, though the result comes from a single experimental system.",
    "field": "physics",
    "ai_technology": "Physics-tailored neural network for inferring force laws from particle trajectories",
    "organization": "Emory University",
    "source_url": "https://news.emory.edu/features/2025/07/ai-reveals-unexpected-new-physics-dusty-plasma",
    "source_type": "press release",
    "publication_date": "2025-07-31",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2505725122",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A physics-constrained neural network fitted to 3D particle-tracking data from a laboratory dusty plasma inferred the interaction forces; particle masses were inferred two independent ways and the force laws were checked against laboratory experiments.",
      "unknown": "The abstract states no explicit limitations; the result comes from a single experimental system.",
      "ai_role": "A physics-informed neural network learned the non-reciprocal force laws from tracking data.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1073/pnas.2505725122&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-supernova-2023zkd-black-hole-merger",
    "title": "Machine-learning alert flags supernova 2023zkd as a black hole and star binary merger",
    "summary": "The LAISS machine-learning alert system flagged the transient SN 2023zkd months before its most unusual behaviour, and the team then analysed its four-year optical precursor. The authors conclude that a massive, partly stripped helium star merged with a black hole companion, with the second light peak explained by a collision with a dense disk-like cloud. The ML step only prioritised the event; the physical interpretation is model-based, and outlets give conflicting accounts of who first detected it.",
    "field": "physics",
    "ai_technology": "LAISS machine-learning anomaly alert for transient detection",
    "organization": "Center for Astrophysics | Harvard & Smithsonian; MIT; University of California Santa Cruz; Young Supernova Experiment",
    "source_url": "https://news.ucsc.edu/2025/08/black-hole-supernova/",
    "source_type": "press release",
    "publication_date": "2025-08-13",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-4357/adea38",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Multiband optical, UV and X-ray data, optical and near-IR spectra and shock-powered light-curve models of about four years of precursor emission were fitted to infer a helium-star and black-hole merger scenario.",
      "unknown": "The authors state the observations favour either super-Eddington accretion or multiple eruptions, so the merger interpretation is not uniquely established; the binary precursor is an inference, not a direct detection.",
      "ai_role": "The LAISS machine-learning anomaly tool flagged the event for follow-up; the physical interpretation is model-based and does not use the ML step.",
      "evidence_urls": [
        "https://iopscience.iop.org/article/10.3847/1538-4357/adea38"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-02-fractional-hall-neural-wavefunction",
    "title": "First-principles neural network finds crystallization of fractional quantum Hall liquids",
    "summary": "A self-attention neural-network wavefunction (MagNet), trained only by minimising the microscopic Hamiltonian of interacting electrons in a magnetic field, finds both topological liquid and electron-crystal ground states across a range of Landau-level mixing. The authors present it as a method that discovers correlated phases without assumptions. This is an arXiv preprint with computational results only, not yet checked experimentally or peer reviewed.",
    "field": "physics",
    "ai_technology": "MagNet: self-attention neural-network variational wavefunction",
    "organization": "MIT",
    "source_url": "https://arxiv.org/abs/2602.03927",
    "source_type": "paper",
    "publication_date": "2026-02-03",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A self-attention neural-network wavefunction trained only by minimising the microscopic Hamiltonian found topological liquid and electron-crystal ground states across Landau-level mixing; the abstract reports computational results only.",
      "unknown": "The page states no limitations and mentions no experimental checks; the arXiv page lists a PRX Intelligence journal reference, while the source record describes it as a preprint.",
      "ai_role": "The MagNet self-attention neural network served as the variational wavefunction, trained by energy minimisation without labelled data.",
      "evidence_urls": [
        "https://arxiv.org/abs/2602.03927"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-01-cadedb-colonoscopy-adenomas",
    "title": "Deep-learning CADe colonoscopy raises adenoma detection in a double-blind trial",
    "summary": "In a double-blind randomised trial of 962 patients in China (484 analysed with CADe, 478 with a sham system), the deep-learning system raised adenoma detection to 34% (165/484) from 28% (132/478), OR 1.36 (95% CI 1.03-1.79). The primary analysis was per protocol, the trial was single-centre, and the abstract reports no ITT analysis.",
    "field": "medicine",
    "ai_technology": "Deep-learning computer-aided detection (CADe) system for polyps during colonoscopy",
    "organization": "Sichuan Academy of Medical Sciences and Sichuan Provincial People's Hospital; Beth Israel Deaconess Medical Center",
    "source_url": "https://doi.org/10.1016/S2468-1253(19)30411-X",
    "source_type": "paper",
    "publication_date": "2020-01-22",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/31981517/",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A double-blind randomised trial of 962 patients compared colonoscopy with a deep-learning CADe system against a sham system; the primary outcome was adenoma detection rate, analysed per protocol (CADe 34% vs sham 28%).",
      "unknown": "The trial is single-centre, the primary analysis is per protocol with no ITT analysis reported, and the effect on interval colorectal cancer still needs investigation.",
      "ai_role": "A deep-learning CADe system gave real-time visual alarms for possible polyps during colonoscopy.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:31981517%20AND%20SRC:MED&resultType=core&format=json"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2021-09-composer-sepsis-indeterminate",
    "title": "COMPOSER sepsis model flags unfamiliar cases instead of guessing",
    "summary": "COMPOSER, a deep-learning early sepsis predictor, reached AUCs of 0.925-0.953 in ICU and 0.938-0.945 in ED across six cohorts (515,720 patients) from two US health systems. About 20% of non-septic and 8% of septic prediction windows were marked indeterminate, and the model gave a median early warning of 12.2 hours in ICU and 2.1 hours in ED before the first antibiotics order. The abstract reports no prospective deployment or patient-outcome results.",
    "field": "medicine",
    "ai_technology": "COMPOSER deep-learning model with uncertainty-based indeterminate output",
    "organization": "University of California San Diego",
    "source_url": "https://doi.org/10.1038/s41746-021-00504-6",
    "source_type": "paper",
    "publication_date": "2021-09-09",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8429719/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "COMPOSER's AUCs were measured across six cohorts (515,720 patients) from two US health systems in ICU and ED settings, with the share of predictions marked indeterminate and early-warning time before the first antibiotics order.",
      "unknown": "The abstract reports no prospective deployment or patient-outcome results; data come from only two US health systems.",
      "ai_role": "A deep-learning model predicted early sepsis and marked unfamiliar cases as indeterminate rather than guessing.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41746-021-00504-6&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-03-mammography-ai-challenge-radiologists",
    "title": "Combining AI algorithms with radiologists improves mammography specificity",
    "summary": "An international crowdsourced challenge (over 1,100 participants, 126 teams, 44 countries) tested AI on screening mammograms from 144,231 US and 166,578 Swedish examinations. The best single algorithm did not beat radiologists (US AUC 0.858), but an ensemble of top algorithms combined with US radiologists' recall decisions reached AUC 0.942 and 92.0% specificity at the same sensitivity. The study is retrospective, uses challenge data and does not report recall rates.",
    "field": "medicine",
    "ai_technology": "Ensemble of crowdsourced deep-learning mammography algorithms combined with radiologist recall",
    "organization": "Sage Bionetworks; and DREAM challenge collaborators, with US and Swedish screening programmes",
    "source_url": "https://doi.org/10.1001/jamanetworkopen.2020.0265",
    "source_type": "paper",
    "publication_date": "2020-03-02",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/32119094/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A crowdsourced challenge of 126 teams was tested on 144,231 US and 166,578 Swedish screening examinations, with an ensemble of top algorithms combined with US radiologists' recall decisions compared on AUC and specificity.",
      "unknown": "The study is retrospective, no single algorithm beat radiologists, the ensemble was evaluated only against US radiologists, and recall rates are not reported.",
      "ai_role": "Algorithms produced cancer risk scores; an ensemble of top algorithms was combined with radiologist recall decisions.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1001/jamanetworkopen.2020.0265&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-09-pancreatic-cancer-ct-nationwide",
    "title": "Deep learning detects pancreatic cancer on CT in a nationwide study",
    "summary": "A deep-learning tool for contrast-enhanced CT reached 89.9% sensitivity and 95.9% specificity in an internal test set (not significantly different from the original radiologist reports, 96.1%). On 1,473 real-world Taiwanese CT studies it reached 89.7% sensitivity and 92.8% specificity, but sensitivity for cancers under 2 cm was 74.7%. The study is retrospective, with no prospective reader comparison in the abstract.",
    "field": "medicine",
    "ai_technology": "Segmentation CNN with an ensemble of five CNN classifiers for pancreatic cancer on CT",
    "organization": "National Taiwan University Hospital; National Taiwan University",
    "source_url": "https://doi.org/10.1148/radiol.220152",
    "source_type": "paper",
    "publication_date": "2022-09-13",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/36098642/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A deep-learning CT tool was tested on an internal retrospective test set against original radiologist reports and on 1,473 real-world Taiwanese CT studies, reaching 89.7% sensitivity and 92.8% specificity in the latter.",
      "unknown": "Sensitivity for tumours under 2 cm was 74.7%; the study is retrospective with no prospective reader comparison in the abstract.",
      "ai_role": "An end-to-end computer-aided detection tool (segmentation CNN plus an ensemble of five CNNs) detected pancreatic cancer on contrast CT.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1148/radiol.220152&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-06-pancreatic-cancer-ct-cross-racial",
    "title": "CNN distinguishes pancreatic cancer on contrast CT with cross-racial validation",
    "summary": "A CNN trained on contrast-enhanced CT from a Taiwanese centre reached 98.3% sensitivity on combined local test sets versus 92.9% for radiologists (difference 5.4 points, p=0.014). On a US external test set, sensitivity fell to 79.0% with specificity 97.6% (AUC 0.920), and sensitivity for tumours under 2 cm was 63.1% in that set. The study is retrospective and the authors describe it as supplementary to radiologists.",
    "field": "medicine",
    "ai_technology": "Convolutional neural network applied to CT image patches",
    "organization": "National Taiwan University; National Taiwan University Hospital",
    "source_url": "https://doi.org/10.1016/s2589-7500(20)30078-9",
    "source_type": "paper",
    "publication_date": "2020-06-01",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/33328124/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A CNN trained on Taiwanese contrast CT was tested on local test sets against radiologists and on an external US test set, where it reached 79.0% sensitivity and 97.6% specificity.",
      "unknown": "The study is retrospective; sensitivity for tumours under 2 cm was 63.1% on the US set, and radiologist comparisons covered only the local sets.",
      "ai_role": "A CNN classified image patches as cancerous and labelled each patient by the proportion of cancerous patches; the authors describe it as supplementary to radiologists.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID:33328124%20AND%20SRC:MED&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-05-med-bert-ehr-prediction",
    "title": "Med-BERT pretrained on structured health records improves disease prediction",
    "summary": "Med-BERT, a transformer pretrained on structured EHR data from 28,490,650 patients, improved AUC by 1.21% to 6.14% across two disease-prediction tasks compared with prior baselines. The gains are reported on retrospective structured-data tasks, and the abstract gives no clinical deployment or outcome results.",
    "field": "medicine",
    "ai_technology": "Med-BERT, BERT-style transformer pretrained on structured electronic health records",
    "organization": "University of Texas Health Science Center at Houston",
    "source_url": "https://doi.org/10.1038/s41746-021-00455-y",
    "source_type": "paper",
    "publication_date": "2021-05-20",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/34017034/",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Med-BERT, pretrained on structured EHR data from 28,490,650 patients, was fine-tuned on two disease-prediction tasks and compared with baselines trained without pretraining, including small-data settings.",
      "unknown": "The abstract states no explicit limitations and no clinical deployment or outcome results.",
      "ai_role": "A BERT-style transformer adapted to structured EHR data produced embeddings fine-tuned for prediction.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41746-021-00455-y&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-05-covid-symptom-app-prediction",
    "title": "Self-reported symptom app predicts likely COVID-19 infection in real time",
    "summary": "A COVID Symptom Study app with 2,618,862 participants and 18,401 tested users found loss of smell or taste in 65.03% of positives versus 21.71% of negatives (OR 6.74). A machine-learning model applied to 805,753 users predicted 140,312 (17.42%) as likely infected. The inputs are self-reported and the test-verified subset is small relative to the user base, so the prediction is not a clinical diagnosis.",
    "field": "medicine",
    "ai_technology": "Machine-learning symptom model on self-reported app data",
    "organization": "King's College London; Massachusetts General Hospital",
    "source_url": "https://doi.org/10.1038/s41591-020-0916-2",
    "source_type": "paper",
    "publication_date": "2020-05-11",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/32393804/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "App-reported symptoms from 2,618,862 participants were compared between 18,401 test-verified users, and a model applied to 805,753 users predicted 140,312 as likely infected.",
      "unknown": "Inputs are self-reported, the test-verified subset is small relative to the user base, the prediction is not a clinical diagnosis, and the abstract does not describe model validation.",
      "ai_role": "A model combining self-reported symptoms predicted likely COVID-19; the abstract does not state the method.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-020-0916-2&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-03-covnet-covid-pneumonia-ct",
    "title": "COVNet deep learning separates COVID-19 from other pneumonia on chest CT",
    "summary": "The COVNet 3D deep-learning model, trained on 4,352 CT scans from 3,322 patients, reached per-scan sensitivity 90% and specificity 96% (AUC 0.96) for COVID-19, and AUC 0.95 for community-acquired pneumonia. The study is retrospective and evaluated on Chinese cohorts, so performance in other populations is not established.",
    "field": "medicine",
    "ai_technology": "COVNet, a 3D convolutional neural network on chest CT",
    "organization": "Wuhan Huangpi People's Hospital; Wuhan Pulmonary Hospital; Liaocheng People's Hospital",
    "source_url": "https://doi.org/10.1148/radiol.2020200905",
    "source_type": "paper",
    "publication_date": "2020-03-19",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/32191588/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The COVNet 3D model, trained on 4,352 CT scans, was evaluated on an independent test set for COVID-19 (sensitivity 90%, specificity 96%, AUC 0.96) and community-acquired pneumonia.",
      "unknown": "The abstract does not state limitations and does not confirm whether the test set came from outside the training hospitals; the study is retrospective.",
      "ai_role": "COVNet, a fully automatic 3D deep-learning framework, classified chest CT scans.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1148/radiol.2020200905&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-04-ai-augments-radiologists-covid-ct",
    "title": "AI augmentation lifts radiologists in telling COVID-19 from other pneumonia",
    "summary": "A CT model trained on 1,186 patients (521 COVID-19, 665 other pneumonia) reached 95% sensitivity and 96% specificity internally, and AUC 0.90 on an independent test set. Radiologist accuracy rose from 85% to 90% with AI assistance. The study is retrospective and the independent-test performance is lower than the internal result.",
    "field": "medicine",
    "ai_technology": "Deep-learning CT classifier used as an assistant for radiologists",
    "organization": "Xiangya Hospital, Central South University; Rhode Island Hospital; University of Pennsylvania",
    "source_url": "https://doi.org/10.1148/radiol.2020201491",
    "source_type": "paper",
    "publication_date": "2020-04-27",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/32339081/",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A CT model was tested on an internal split and on an independent test set from separate hospitals, and six radiologists read cases unassisted and then with the model's probabilities.",
      "unknown": "The abstract states no limitations section; performance on independent data was lower (87% vs 96% accuracy) and the sensitivity confidence interval is wide.",
      "ai_role": "The AI classified COVID-19 versus other pneumonia on CT slices and its probabilities were shown to radiologists as decision support.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1148/radiol.2020201491&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-04-penet-pulmonary-embolism-ct",
    "title": "PENet detects pulmonary embolism on volumetric CT angiography",
    "summary": "PENet, a 3D deep-learning model for pulmonary embolism on CT angiography, reached AUROC 0.84 on an internal hold-out test and 0.85 on data from one unnamed external institution. The abstract does not report sensitivity, specificity or study size, and the external validation is limited to a single site.",
    "field": "medicine",
    "ai_technology": "PENet, 3D deep convolutional network on volumetric CT",
    "organization": "Stanford University",
    "source_url": "https://doi.org/10.1038/s41746-020-0266-y",
    "source_type": "paper",
    "publication_date": "2020-04-24",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/32352039/",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "PENet, a 3D CNN, was tested on a hold-out set from the training institution (AUROC 0.84) and on an external institution (AUROC 0.85), and was compared with other state-of-the-art 3D CNN models.",
      "unknown": "The abstract states no limitations; external validation covers one institution and sensitivity, specificity and study size are not reported.",
      "ai_role": "A 77-layer 3D CNN, pretrained on Kinetics-600, detected pulmonary embolism on CT angiography and is proposed as a triage tool.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41746-020-0266-y&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-01-stimulated-raman-brain-tumor-diagnosis",
    "title": "Deep neural networks diagnose brain tumours in minutes from stimulated Raman histology",
    "summary": "CNNs trained on over 2.5 million stimulated Raman histology images produced intraoperative brain tumour diagnoses in under 150 seconds, with 94.6% accuracy versus 93.9% for pathologists. The accuracy comparison comes from a prospective multicentre trial with 278 patients, but the abstract does not report sensitivity or specificity separately.",
    "field": "medicine",
    "ai_technology": "Deep neural networks on stimulated Raman histology images",
    "organization": "University of Michigan; Columbia University; University of Miami",
    "source_url": "https://doi.org/10.1038/s41591-019-0715-9",
    "source_type": "paper",
    "publication_date": "2020-01-06",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/31907460/",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Multicenter prospective clinical trial (n = 278): CNN diagnosis of stimulated Raman images was non-inferior to pathologists reading conventional histology (94.6% versus 93.9%), in under 150 seconds.",
      "unknown": "Long-term patient outcomes are not reported in the abstract.",
      "ai_role": "Convolutional neural networks trained on over 2.5 million stimulated Raman histology images.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41591-019-0715-9"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2020-12-multimodal-ct-ehr-pulmonary-embolism",
    "title": "Fusing CT images and EHR data with deep learning improves pulmonary embolism detection",
    "summary": "A deep multimodal network combining CT imaging with electronic health record data reached AUROC 0.947 (95% CI 0.946-0.948) on a held-out test set for pulmonary embolism. The abstract reports no sensitivity, specificity, study size or external validation sites, so the evidence is limited to retrospective internal testing.",
    "field": "medicine",
    "ai_technology": "Multimodal deep neural network fusing CT and EHR data",
    "organization": "Stanford University",
    "source_url": "https://doi.org/10.1038/s41598-020-78888-w",
    "source_type": "paper",
    "publication_date": "2020-12-17",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://pubmed.ncbi.nlm.nih.gov/33335111/",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Several fusion architectures were compared on a held-out test set, and the best late-fusion model reached AUROC 0.947 against imaging-only and EMR-only models.",
      "unknown": "The abstract reports no sensitivity, specificity, study size or external validation and states no limitations.",
      "ai_role": "Deep neural networks classified pulmonary embolism from CT angiography volumes combined with EHR data.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41598-020-78888-w&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-06-acropora-belize-satellite",
    "title": "Satellite model maps collapse and surviving Acropora corals in Belize",
    "summary": "A supervised Random Tree classifier on satellite images predicted a drop of over 90% in live Acropora cover at Belize from 2011 to 2023 across 25 km². Field searches in June 2025 found live or recently dead colonies at 28 of 42 predicted sites and live colonies at 16 previously undocumented locations. The field check was a search for survivors at sites the model flagged, not a blind test.",
    "field": "biology",
    "ai_technology": "Random Tree supervised classification of satellite imagery (machine learning)",
    "organization": "Washington and Lee University; Macalester College",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42230658/",
    "source_type": "paper",
    "publication_date": "2026-06-02",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41598-026-47081-w",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Random Tree classifier's predicted over-90% decline in Acropora cover (2011-2023) was compared with field data from 2012-2025, and field searches at 42 predicted sites found live or recently dead colonies at 28 of them.",
      "unknown": "The abstract states no limitations; the search was a targeted check of flagged sites rather than a blind test, and 14 of 42 predicted sites had no colonies found.",
      "ai_role": "A supervised Random Tree classifier identified Acropora in satellite images and generated the cover predictions that guided the field searches.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41598-026-47081-w&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-european-hamster-yolo-monitoring",
    "title": "YOLO detector tracks endangered European hamster activity in camera traps",
    "summary": "A YOLO object-detection model trained on 2023 camera-trap videos detected European hamsters with an F1-score of 0.93 and accuracy of 0.93. Its activity profile matched human annotation and showed peak activity at dusk and a secondary peak before sunrise. The evidence covers one season at one monitoring setup, so generalization to other sites is untested.",
    "field": "biology",
    "ai_technology": "YOLO object detection (deep learning)",
    "organization": "Goethe University Frankfurt",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42699261/",
    "source_type": "paper",
    "publication_date": "2026-09-04",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/ece3.74267",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A YOLO detector trained on 2023 camera-trap videos was tested for European hamster detection (F1 and accuracy 0.93), and its activity profile was compared with human annotation.",
      "unknown": "The abstract states no limitations; it covers one season and does not name the sites or the number of cameras.",
      "ai_role": "A YOLO object-detection model detected hamsters in camera-trap footage.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1002/ece3.74267&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-06-right-whale-individual-id-birdnet",
    "title": "BirdNET embeddings distinguish individual North Atlantic right whales",
    "summary": "Audio embeddings from the BirdNET model robustly distinguished individual North Atlantic right whales in upcall recordings from 234 samples across 11 individuals at 3 sites. Simulations showed that an extra 7 dB or more of signal excess is needed to identify individuals under ambient noise. The study uses on-animal archival tags on an endangered species, but the sample is small and the model was not trained for whales.",
    "field": "biology",
    "ai_technology": "BirdNET audio embeddings (deep learning)",
    "organization": "Cornell K. Lisa Yang Center for Conservation Bioacoustics; Oregon State University; Syracuse University",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42378267/",
    "source_type": "paper",
    "publication_date": "2026-06-30",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pcbi.1013321",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "BirdNET audio embeddings were tested for distinguishing 11 individual right whales in 234 upcall samples, and noise simulations estimated the extra signal excess needed for individual identification compared with detection.",
      "unknown": "The abstract lists no explicit limitations; noise effects are simulated rather than field-measured, and the dataset is small (11 individuals at 3 sites).",
      "ai_role": "The BirdNET deep-learning model generated audio embeddings used to classify individual identity.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1371/journal.pcbi.1013321&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-fluembed-h3n2-antigenic-drift",
    "title": "FluEmbed predicts H3N2 antigenic drift from sequence alone",
    "summary": "FluEmbed, a protein language model framework, predicted influenza A (H3N2) antigenicity from RNA sequence with Spearman correlations of 0.67 to 0.80 against hemagglutination inhibition titers. It outperformed Hamming, BLOSUM62 and phylogenetic baselines, and in-silico mutagenesis showed unconstrained mutations carry greater escape potential outside dominant lineages. Validation uses existing HI assay data and retrospective sequences; no new wet-lab tests were run.",
    "field": "biology",
    "ai_technology": "FluEmbed (protein language model embeddings)",
    "organization": "University of Georgia",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42579720/",
    "source_type": "paper",
    "publication_date": "2026-08-11",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pcbi.1014628",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "FluEmbed predictions from H3N2 sequences were correlated with existing hemagglutination inhibition titers (Spearman 0.67-0.80) and compared with Hamming, BLOSUM62 and phylogenetic baselines, with in-silico mutagenesis for escape potential.",
      "unknown": "The abstract states no limitations and does not discuss generalization beyond H3N2 or beyond the data used; no wet-lab tests were run.",
      "ai_role": "A protein language model framework predicted H3N2 antigenicity from RNA sequence.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1371/journal.pcbi.1014628&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-11-palladium-dinuclear-ml",
    "title": "Unsupervised machine learning identifies palladium catalysts that form dinuclear complexes",
    "summary": "A workflow built on only five experimental data points screened 348 phosphine ligands and predicted ligands that favour dinuclear Pd(I) complexes over Pd(0) and Pd(II) species. The authors experimentally verified the predictions, including ligands that had never been synthesized before. The evidence is limited to the Pd speciation system studied, not a general catalyst performance gain.",
    "field": "chemistry",
    "ai_technology": "Unsupervised machine learning workflow with clustering on generalized parameter databases and in silico data",
    "organization": "RWTH Aachen University; University of Jyväskylä",
    "source_url": "https://doi.org/10.1126/science.abj0999",
    "source_type": "paper",
    "publication_date": "2021-11-25",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors experimentally verified the predicted phosphine ligands, including ligands never synthesized before, and they gave dinuclear Pd(I) complexes. The ligand predictions were checked by laboratory experiments.",
      "unknown": "Not reported.",
      "ai_role": "Unsupervised machine learning workflow built on five experimental data points, screened 348 phosphine ligands and predicted those favouring dinuclear Pd(I) complexes.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.abj0999&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-12-bi2o3-autonomous-phase-diagram",
    "title": "Autonomous robotic system maps Bi2O3 synthesis phase diagram with active learning",
    "summary": "The SARA agent combined robotic laser spike annealing, optical characterization and nested active learning to map synthesis phase boundaries of Bi2O3. It reached orders-of-magnitude faster establishment of the phase diagram and identified conditions that stabilize metastable delta-Bi2O3. Validation is experimental but restricted to one oxide system.",
    "field": "materials",
    "ai_technology": "Scientific Autonomous Reasoning Agent (SARA) with hierarchical active learning and uncertainty quantification",
    "organization": "Cornell University; University of Bern; California Institute of Technology",
    "source_url": "https://doi.org/10.1126/sciadv.abg4930",
    "source_type": "paper",
    "publication_date": "2021-12-17",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The SARA system autonomously mapped Bi2O3 synthesis phase boundaries with robotic synthesis and optical characterization, and the phase diagram was established experimentally.",
      "unknown": "Not reported.",
      "ai_role": "Nested active learning cycles directed robotic laser spike annealing and optical characterization to map the phase diagram.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/sciadv.abg4930&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-12-umifenovir-retrosynthesis-routes",
    "title": "Retrosynthetic software finds experimentally validated supply routes for umifenovir",
    "summary": "Retrosynthesis software proposed alternative chemical supply routes for the antiviral umifenovir and eleven other antiviral and anti-inflammatory drugs. The authors experimentally validated four routes to umifenovir and one route to bromhexine, one of which converts six C-H bonds. The work is a proof of route feasibility for supply-chain resilience, not a new therapeutic claim.",
    "field": "chemistry",
    "ai_technology": "Computer-aided retrosynthesis software with exclusion of known starting materials",
    "organization": "University of Michigan; MilliporeSigma; Entos, Inc.",
    "source_url": "https://doi.org/10.1038/s41467-021-27547-3",
    "source_type": "paper",
    "publication_date": "2021-12-16",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors experimentally validated four routes to umifenovir and one route to bromhexine proposed by the software.",
      "unknown": "Not reported.",
      "ai_role": "Retrosynthesis software proposed alternative supply routes for umifenovir and other antiviral and anti-inflammatory drugs.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-021-27547-3&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-04-cupd-nanocube-nitrate-reduction",
    "title": "Machine-learned design of CuPd nanocubes for electrocatalytic nitrate reduction to ammonia",
    "summary": "Interpretable machine learning identified ordered B2 CuPd intermetallics whose adsorption energies break the usual scaling relations. The team synthesized monodisperse CuPd nanocubes that reached a 92.5% Faradaic efficiency at -0.5 V RHE for nitrate reduction to ammonia, with a yield of 6.25 mol h-1 g-1 at -0.6 V. Performance was measured in the authors' own electrochemical tests, and the design rules come from a limited set of predicted intermetallics.",
    "field": "materials",
    "ai_technology": "Interpretable machine learning on d-state Pauli repulsion descriptors (DFT-based)",
    "organization": "Virginia Tech; National University of Singapore; Argonne National Laboratory",
    "source_url": "https://doi.org/10.1038/s41467-022-29926-w",
    "source_type": "paper",
    "publication_date": "2022-04-29",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors synthesized monodisperse ordered B2 CuPd nanocubes and measured nitrate reduction performance in their own electrochemical tests (92.5% Faradaic efficiency at -0.5 V RHE).",
      "unknown": "Not reported.",
      "ai_role": "Interpretable machine learning identified ordered B2 CuPd intermetallics whose adsorption energies break scaling relations and provided design rules.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-022-29926-w&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-03-ni3mo-ethane-dehydrogenation",
    "title": "Earth-abundant Ni3Mo alloy outperforms platinum for ethane dehydrogenation",
    "summary": "A microkinetic and high-throughput screen of 1,998 bimetallic alloys, combined with machine-learning concepts, shortlisted noble-metal-free catalysts for selective ethane dehydrogenation. The predicted Ni3Mo alloy was synthesized and experimentally shown to outperform Pt in selective ethylene production. The machine-learning role is secondary to the descriptor-based workflow, and the evidence is a single catalyst test.",
    "field": "materials",
    "ai_technology": "Descriptor-based microkinetic modeling with high-throughput DFT and machine learning",
    "organization": "SLAC National Accelerator Laboratory (SUNCAT); Westlake University; Lanzhou Institute of Chemical Physics, Chinese Academy of Sciences",
    "source_url": "https://doi.org/10.1073/pnas.2024666118",
    "source_type": "paper",
    "publication_date": "2021-03-12",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The predicted Ni3Mo alloy was synthesized and experimentally shown to outperform Pt in selective ethylene production from ethane dehydrogenation.",
      "unknown": "Not reported.",
      "ai_role": "A microkinetic and high-throughput screen of 1,998 bimetallic alloys, based on calculated adsorption energies, shortlisted noble-metal-free catalysts; the abstract describes machine learning concepts as secondary.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1073/pnas.2024666118&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-06-mpro-synthesis-directed-design",
    "title": "COVID Moonshot uses synthesis-directed de novo design for SARS-CoV-2 main protease inhibitors",
    "summary": "A machine-learning de novo design model was coupled with synthesis route prediction in the COVID Moonshot campaign against the SARS-CoV-2 main protease. The authors report novel chemical scaffolds that are active in biochemical and live-virus assays and were synthesized along model-generated routes. The abstract gives no potency figures, so the strength of the result should be checked in the full text.",
    "field": "medicine",
    "ai_technology": "Synthesis-directed de novo design model with synthesis route prediction",
    "organization": "PostEra Inc.; University of Cambridge; The COVID Moonshot Consortium; University of Chicago; Memorial Sloan Kettering Cancer Center; Weizmann Institute of Science",
    "source_url": "https://doi.org/10.1039/d1cc00050k",
    "source_type": "paper",
    "publication_date": "2021-06-01",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Novel scaffolds designed with machine learning and synthesized along model-generated routes were tested in biochemical and live-virus assays.",
      "unknown": "Not reported.",
      "ai_role": "Machine-learning de novo design model coupled with synthesis route prediction in the COVID Moonshot campaign.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1039/d1cc00050k&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-10-19f-mri-copolymer-ml",
    "title": "Machine learning guides automated synthesis of 19F MRI copolymer agents",
    "summary": "An automated continuous-flow polymer platform synthesized 397 copolymer compositions in a six-variable space, in iterative cycles guided by machine learning. Exploring under 0.9% of the compositional space, the models identified more than ten copolymers that outperformed state-of-the-art 19F MRI agents. Validation is experimental across hundreds of real syntheses, but the imaging performance claims are limited to the compositions tested.",
    "field": "materials",
    "ai_technology": "Machine learning models guiding automated flow copolymer synthesis",
    "organization": "University of North Carolina at Chapel Hill; Carnegie Mellon University",
    "source_url": "https://doi.org/10.1021/jacs.1c08181",
    "source_type": "paper",
    "publication_date": "2021-10-12",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An automated flow-synthesis platform made 397 copolymer compositions in a six-variable space; 10 compositions outperformed state-of-the-art 19F MRI agents.",
      "unknown": "In vivo imaging performance is not reported in the abstract.",
      "ai_role": "Machine learning guided iterative synthesis cycles and identified non-intuitive design criteria.",
      "evidence_urls": [
        "https://doi.org/10.1021/jacs.1c08181"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-07-polymer-gas-separation-ml",
    "title": "Interpretable machine learning discovers ultrapermeable polymer membranes for gas separation",
    "summary": "Multitask machine-learning models trained on experimental gas permeability data screened over 9 million hypothetical polymers. They identified hundreds of never-before-seen ultrapermeable membranes, with molecular dynamics simulations confirming the predicted permeabilities of the top candidates. The validation is computational rather than fabricated membranes, so the candidates remain unproven experimentally.",
    "field": "materials",
    "ai_technology": "Multitask machine learning with model interpretation, plus molecular dynamics",
    "organization": "California Institute of Technology; University of Connecticut",
    "source_url": "https://doi.org/10.1126/sciadv.abn9545",
    "source_type": "paper",
    "publication_date": "2022-07-20",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Molecular dynamics simulations checked the predicted permeabilities of the top candidates among over 9 million screened hypothetical polymers; no membranes were fabricated.",
      "unknown": "The authors suggest many candidates could be translated to reality, which is not tested; no limitations section is reported.",
      "ai_role": "Interpretable multitask machine-learning models trained on experimental gas permeability data screened hypothetical polymers.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/sciadv.abn9545&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-maplemit-methane-point-sources",
    "title": "Global methane point sources mapped from EMIT imaging spectroscopy with a vision transformer",
    "summary": "MAPL-EMIT used a vision transformer trained on 3.6 million synthetic plumes embedded in EMIT radiance data to map methane sources. Beyond synthetic tests, the paper abstract reports recovery of 84% of known annotated plume complexes across 1,084 EMIT granules. Comparisons with airborne data, landfills and controlled releases supported previously uncaptured sources. This is source-reported validation, not an independent emissions audit.",
    "field": "climate",
    "ai_technology": "MAPL-EMIT (end-to-end vision transformer on EMIT hyperspectral radiance)",
    "organization": "Google Research; Carnegie Institution for Science; Jet Propulsion Laboratory, California Institute of Technology",
    "source_url": "https://doi.org/10.1073/pnas.2612145123",
    "source_type": "paper",
    "publication_date": "2026-09-01",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Synthetic tests plus known real plumes, coincident airborne observations, landfill comparisons and controlled-release data.",
      "unknown": "Review scope: complete indexed abstract and metadata; full uncertainty budgets and individual source emissions were not assessed.",
      "ai_role": "An end-to-end vision transformer jointly estimated methane enhancements, plume outlines and source locations from hyperspectral radiance.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1073%2Fpnas.2612145123%22&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-05-methane-plumes-sentinel2-vision-transformer",
    "title": "Vision transformer detects methane point-source plumes in Sentinel-2 imagery at global scale",
    "summary": "A vision transformer applied to multispectral Sentinel-2 data detects methane plumes as small as 0.01 km2, corresponding to sources of about 200 to 300 kg CH4 per hour, and is reported to improve on the prior state of the art by an order of magnitude. Detections were compared with airborne methane measurement campaigns; the claims rest on that comparison and the abstract gives no formal detection-rate metrics.",
    "field": "climate",
    "ai_technology": "Vision transformer for multispectral methane plume detection (Sentinel-2)",
    "organization": "Kyoto University (Disaster Prevention Research Institute)",
    "source_url": "https://doi.org/10.1038/s41467-024-47754-y",
    "source_type": "paper",
    "publication_date": "2024-05-14",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Detections were compared with airborne methane measurement campaigns, and the method is reported to detect plumes down to 0.01 km2 (about 200 to 300 kg CH4 per hour); the authors describe this as supporting evidence rather than a definitive validation.",
      "unknown": "Not reported; the page lists no explicit limitations and gives no formal detection-rate metrics.",
      "ai_role": "Vision transformer applied to multispectral Sentinel-2 data to detect methane plumes, reported to improve on prior methods by an order of magnitude.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-024-47754-y&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-07-quake-neural-operator-geothermal",
    "title": "Quake Neural Operator builds microseismic catalogues without phase picking",
    "summary": "The Quake Neural Operator detects and locates microseismic events directly from continuous data without explicit phase picking. Applied to the Geysers geothermal field, it finds nearly an order of magnitude more seismic detections than routine catalogues, checked against the Phase Neural Operator and visual inspection rather than an independent ground-truth catalogue.",
    "field": "other",
    "ai_technology": "Quake Neural Operator (QNO), multi-task neural operator for classification and regression",
    "organization": "University of Texas at El Paso, Department of Earth, Environmental and Resource Sciences",
    "source_url": "https://doi.org/10.1038/s41467-026-73965-6",
    "source_type": "paper",
    "publication_date": "2026-07-06",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Detections at the Geysers geothermal field were compared with the Phase Neural Operator and visual inspection, and the method was reported to find nearly an order of magnitude more detections than routine catalogs.",
      "unknown": "Not reported; the detections are checked only against the Phase Neural Operator and visual inspection.",
      "ai_role": "Quake Neural Operator detects and locates microseismic events directly from continuous data without explicit phase picking.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-026-73965-6&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-04-simulated-data-archaeological-sites-ml",
    "title": "Simulated training data locates archaeological tar-kiln-like sites in Louisiana, field check finds WWII features",
    "summary": "Machine learning models trained on simulated landscape data flagged 9 of 12 known targets (model 1) and all 12 (model 2, with 686 false positives among 709 remaining predictions) in Kisatchie National Forest, Louisiana; a model trained on real South Carolina tar kilns found 11 of 12. Field visits to 11 targets and augering of 2 found none of the expected tar-kiln features, and the authors concluded the structures more likely relate to a WWII military training site. Evidence caveat: only 12 known targets, and field validation is qualitative.",
    "field": "humanities",
    "ai_technology": "Machine learning classifiers trained on simulated archaeological sites (with a real tar-kiln-trained comparison model)",
    "organization": "University of New Mexico; New Mexico Consortium; Utah State University; Louisiana State University",
    "source_url": "https://www.cambridge.org/core/product/identifier/S2326376825101307/type/journal_article",
    "source_type": "paper",
    "publication_date": "2026-04-27",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1017/aap.2025.10130",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Models trained on simulated objects flagged known tar-kiln-like targets, and 11 targets were visited in the field with 2 augered; no expected tar-kiln features were found, and the authors concluded the features are most likely associated with military training exercises.",
      "unknown": "Access to all targets was limited by dense understory, and further historical research is needed to confirm the function of the objects.",
      "ai_role": "Mask R-CNN models trained on simulated lidar-embedded objects produced predictions, a rule-based script removed likely natural mounds, and human archaeologists reviewed and chose field targets.",
      "evidence_urls": [
        "https://www.cambridge.org/core/product/identifier/S2326376825101307/type/journal_article"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-google-era-empirical-research-assistance",
    "title": "Empirical Research Assistance writes expert-level scientific code across six domains",
    "summary": "Google's Empirical Research Assistance (ERA), a Gemini-based tree-search system, reaches expert-level performance on six benchmarks spanning genomics, public health, satellite imagery, neuroscience, time-series forecasting and mathematics. Its state-level U.S. hospital admission forecasts for flu, COVID-19 and RSV rank at or near the top of CDC leaderboards. Results are computational and checked against public benchmarks and leaderboards, with no wet-lab validation; exact benchmark scores were not visible in the sources read.",
    "field": "multidisciplinary",
    "ai_technology": "Empirical Research Assistance (ERA), Gemini-based tree search",
    "organization": "Google Research",
    "source_url": "https://www.nature.com/articles/s41586-026-10658-6",
    "source_type": "paper",
    "publication_date": "2026-05-19",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "ERA-generated methods and models were compared against human-developed methods on a public leaderboard for bioinformatics and against the CDC ensemble and other individual models for COVID-19 hospitalization forecasts.",
      "unknown": "Not reported.",
      "ai_role": "A Gemini-based large language model with tree search generated scientific software, including 40 single-cell data analysis methods and 14 forecasting models.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-026-10658-6&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-microsoft-quine-pancreatic-cancer",
    "title": "Quine ranks compounds that shift pancreatic cancer cells toward a basal state",
    "summary": "Microsoft's Quine, a multimodal biology world model, prioritized thousands of compounds for shifting tumor cell states, and a handful were tested in pancreatic ductal adenocarcinoma cell lines. The top-ranked compounds produced the largest classical-to-basal shifts, as validated by wet-lab experiments, and some moved cells toward a distinct third phenotype. The post gives no quantitative wet-lab figures and links no paper, and it measures cell-state changes rather than clinical efficacy.",
    "field": "biology",
    "ai_technology": "Quine, multimodal biology world model with interactive harness",
    "organization": "Microsoft Research; Broad Institute of Harvard and MIT",
    "source_url": "https://microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology",
    "source_type": "blog",
    "publication_date": "2026-09-29",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "A handful of top-ranked compounds predicted by Quine were tested in patient-derived pancreatic ductal adenocarcinoma cell lines, where the largest classical-to-basal transcriptional shifts were reported.",
      "unknown": "The page calls the drug-repurposing findings early evidence and does not report how many compounds were tested or the assay statistics.",
      "ai_role": "Quine, a multimodal biology world model with a harness, ranked thousands of compounds by their predicted ability to shift tumor cell states.",
      "evidence_urls": [
        "https://microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-knuth-claude-cycles",
    "title": "Claude's Cycles: a Hamiltonian cycle decomposition found by Claude",
    "summary": "Knuth's note reports that Claude Opus 4.6 found an explicit decomposition of the arcs of the m^3-vertex digraph into three directed Hamiltonian cycles for odd m, an open problem Knuth posed. Knuth's proof of the construction was reportedly formally verified in Lean by Kim Morrison. Caveat: sources disagree on the even case, which an independent arXiv paper (2603.24708) and a GPT-5.4 Pro proof treat differently.",
    "field": "computer science",
    "ai_technology": "Claude Opus 4.6",
    "organization": "Stanford University; Anthropic",
    "source_url": "https://cs.stanford.edu/~knuth/papers/claude-cycles.pdf",
    "source_type": "other",
    "publication_date": "2026-02-28",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 5
    }
  },
  {
    "id": "2026-05-zarankiewicz-openevolve",
    "title": "New Zarankiewicz numbers and bounds from OpenEvolve search",
    "summary": "Using the open-source OpenEvolve LLM-driven evolutionary framework with a custom reward, the authors find exact values Z(11,21,3,3)=116, Z(11,22,3,3)=121 and Z(12,22,3,3)=132, and lower bounds for 41 further parameter sets, at under $30 per case. Caveats: arXiv preprint not peer reviewed; a separate NeurIPS 2026 listing by overlapping authors reports different counts.",
    "field": "computer science",
    "ai_technology": "OpenEvolve (LLM-guided evolutionary search, AlphaEvolve-style)",
    "organization": "Massachusetts Institute of Technology; ETH Zurich; Stanford University",
    "source_url": "https://arxiv.org/abs/2605.01120",
    "source_type": "paper",
    "publication_date": "2026-05-01",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The exact values of three Zarankiewicz numbers and lower bounds for 41 further parameter sets were obtained by OpenEvolve search with a tailored reward.",
      "unknown": "The page does not describe how the exact values or bounds were verified.",
      "ai_role": "OpenEvolve, an open-source LLM-based evolutionary framework, searched for mathematical constructions at under $30 per parameter combination.",
      "evidence_urls": [
        "https://arxiv.org/abs/2605.01120"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-01-co-funsearch-bin-packing",
    "title": "Co-FunSearch finds adversarial instances raising the Best-Fit bin packing bound",
    "summary": "A human-LLM pipeline (Co-FunSearch) finds adversarial instances for heuristics, raising the lower bound for Best-Fit bin packing in the random-order model from 1.3 to 1.5. The authors prove the 3/2 ratio analytically and report a search-found instance scoring 1.497 over 10,000 random shuffles against Packingsolver optima. Caveat: arXiv preprint; human analysis supplies the final proofs.",
    "field": "computer science",
    "ai_technology": "Co-FunSearch (FunSearch-based LLM program search)",
    "organization": "University of Bonn; Google DeepMind; University of Manitoba",
    "source_url": "https://arxiv.org/abs/2601.16849",
    "source_type": "paper",
    "publication_date": "2026-01-23",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Human-refined FunSearch outputs were used to derive improved adversarial lower bounds for standard heuristics; the page gives no details of the bin-packing test setup.",
      "unknown": "Not reported on the abstract page; the full paper is needed for the bin-packing evaluation details.",
      "ai_role": "LLM-based evolutionary search (FunSearch) supplied initial patterns, which human experts refined and extrapolated into constructions.",
      "evidence_urls": [
        "https://arxiv.org/abs/2601.16849"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-04-anderson-conjecture-rethlas-archon",
    "title": "Anderson's 2014 commutative algebra problem resolved by Rethlas and Archon",
    "summary": "An informal reasoning agent (Rethlas) and a formal verification agent (Archon) resolve an open problem in commutative algebra posed by D. D. Anderson in 2014, as a counterexample. The Lean 4 formalization has about 19,450 lines and was produced in roughly 80 agent hours, with the code public on GitHub. Caveat: arXiv preprint; the Lean check covers the formal statement as written.",
    "field": "mathematics",
    "ai_technology": "Rethlas (informal agent) with Archon (formal Lean agent)",
    "organization": "Peking University; Westlake University; Kyoto University; Tianjin University",
    "source_url": "https://arxiv.org/abs/2604.03789",
    "source_type": "paper",
    "publication_date": "2026-04-04",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The informal proof was formalized and machine-checked in Lean 4 for the formal statement as written.",
      "unknown": "The page describes the case studies as illustrations of capabilities rather than broad evaluations.",
      "ai_role": "Rethlas, an informal reasoning agent, built candidate proofs; Archon, a formal verification agent, translated arguments into Lean 4 projects.",
      "evidence_urls": [
        "https://arxiv.org/abs/2604.03789"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-03-erdos-650-aristotle",
    "title": "Erdős Problem #650 solved with ChatGPT strategy and Aristotle Lean proof",
    "summary": "The authors prove f(m) = min(m, ceil(2 sqrt(m))) for all m, resolving Erdős Problem #650 on Bloom's site, where f(m) counts disjoint divisibility pairs in an interval for an m-element set. The detailed argument was formally verified in Lean by Harmonic's Aristotle, following a strategy proposed by ChatGPT. Caveat: arXiv preprint; the AI contribution is the strategy and formalization, not an independently checked autonomous solution.",
    "field": "mathematics",
    "ai_technology": "ChatGPT (strategy) and Aristotle (Lean formalization)",
    "organization": "Southeast University; Xi'an Jiaotong University",
    "source_url": "https://arxiv.org/abs/2603.28636",
    "source_type": "paper",
    "publication_date": "2026-03-30",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The detailed argument for f(m) = min(m, ceil(2 sqrt(m))) was formally verified in Lean by Harmonic's Aristotle.",
      "unknown": "The abstract lists no explicit limitations; the authors state the exposition and final proofs are entirely human-written.",
      "ai_role": "ChatGPT proposed the proof strategy and Aristotle produced the Lean formalization.",
      "evidence_urls": [
        "https://arxiv.org/abs/2603.28636"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2020-03-echonet-dynamic",
    "title": "EchoNet-Dynamic: video-based AI for left ventricular ejection fraction",
    "summary": "A deep learning model trained on echocardiogram videos segmented the left ventricle with a Dice coefficient of 0.92 and estimated ejection fraction with a mean absolute error of 4.1% (6.0% on an external dataset from another health system). Prospective evaluation with repeated human measurements found its variance comparable to or lower than that of human experts. Caveat: training data come from one institution's echocardiograms, with external testing on a single additional system.",
    "field": "medicine",
    "ai_technology": "EchoNet-Dynamic (video deep learning model)",
    "organization": "Stanford University; Cedars-Sinai Medical Center",
    "source_url": "https://www.nature.com/articles/s41586-020-2145-8",
    "source_type": "paper",
    "publication_date": "2020-03-25",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-020-2145-8",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Segmentation and ejection fraction were tested on held-out internal data (Dice 0.92, MAE 4.1%) and an external dataset from another health system (MAE 6.0%), and repeated human measurements were compared with model variance.",
      "unknown": "Not reported; the page states no limitations section.",
      "ai_role": "Deep learning model trained on echocardiogram videos to segment the left ventricle and estimate ejection fraction.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-020-2145-8&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-05-prenatal-chd-ensemble",
    "title": "Neural network ensemble for prenatal detection of complex congenital heart disease",
    "summary": "An ensemble of neural networks trained on 107,823 fetal ultrasound images reached an AUC of 0.99, 95% sensitivity and 96% specificity in an internal test set of 4,108 fetal surveys with 0.9% CHD. Its sensitivity was comparable to that of clinicians and it stayed robust on outside-hospital and lower-quality images. Caveat: retrospective study with a low-prevalence internal test set.",
    "field": "medicine",
    "ai_technology": "Ensemble of convolutional neural networks with segmentation models",
    "organization": "University of California, San Francisco; Harvard Medical School",
    "source_url": "https://www.nature.com/articles/s41591-021-01342-5",
    "source_type": "paper",
    "publication_date": "2021-05-14",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-021-01342-5",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Trained on 107,823 images from 1,326 retrospective studies; on 4,108 fetal surveys (0.9% CHD), AUC 0.99, 95% sensitivity, 96% specificity, 100% negative predictive value; comparable to clinicians and robust on outside-hospital images.",
      "unknown": "Prospective screening performance is not reported in the abstract.",
      "ai_role": "An ensemble of neural networks identifies cardiac views and classifies normal versus complex congenital heart disease.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41591-021-01342-5"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-09-exam-federated-covid",
    "title": "Federated learning model predicts oxygen needs in COVID-19 patients",
    "summary": "The EXAM federated model, trained on data from 20 institutes without sharing patient data, achieved an average AUC above 0.92 for predicting oxygen requirements at 24 and 72 hours. At the largest independent test site it reached sensitivity 0.950 and specificity 0.882, with a 16% average AUC gain over single-site models. Caveat: retrospective data from the pandemic period and unharmonized sites.",
    "field": "medicine",
    "ai_technology": "EXAM federated learning model (EMR data and chest X-rays)",
    "organization": "University of Toronto; NVIDIA; Icahn School of Medicine at Mount Sinai; NYU Grossman School of Medicine",
    "source_url": "https://www.nature.com/articles/s41591-021-01506-3",
    "source_type": "paper",
    "publication_date": "2021-09-15",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-021-01506-3",
    "relevance": {
      "impact": 5,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Data from 20 institutes; average AUC above 0.92 for oxygen-requirement outcomes at 24 and 72 hours, 16% higher average AUC than single-site models; at the largest independent site, sensitivity 0.950 and specificity 0.882 for ventilation or death at 24 hours.",
      "unknown": "Effect on patient care when used in practice is not reported in the abstract.",
      "ai_role": "EXAM, a model trained by federated learning on vital signs, laboratory data and chest X-rays without exchanging data between sites.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41591-021-01506-3"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-06-hiv-field-rapid-tests-deep-learning",
    "title": "Deep learning reads HIV rapid tests photographed in rural South Africa",
    "summary": "Deep learning models trained on 11,374 smartphone-tablet images of HIV lateral-flow tests, captured by 60 fieldworkers, classified results with 97.8% sensitivity and 100% specificity in a pilot field study compared with human visual reading. The approach also reduced false positives and false negatives relative to human readers. Caveat: a pilot deployment in one rural setting, so wider field validation is still needed.",
    "field": "medicine",
    "ai_technology": "Deep learning image classifier on mobile-captured rapid test images",
    "organization": "Africa Health Research Institute; University College London",
    "source_url": "https://www.nature.com/articles/s41591-021-01384-9",
    "source_type": "paper",
    "publication_date": "2021-06-17",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-021-01384-9",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a pilot field study, deep learning classification of photographed HIV rapid tests reached 97.8% sensitivity and 100% specificity, compared with human visual reading by experienced nurses.",
      "unknown": "Not reported; the page describes the study only as a pilot field study.",
      "ai_role": "Deep learning models trained on 11,374 smartphone images of lateral-flow tests classified results in a mobile application.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-021-01384-9&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-03-deep-learning-cardiac-allograft-rejection",
    "title": "Deep learning assesses heart allograft rejection from biopsy whole-slide images",
    "summary": "A deep learning system analysing gigapixel endomyocardial biopsy slides detected allograft rejection with an AUC of 0.962, distinguished cellular and antibody-mediated rejection (AUCs 0.958 and 0.874) and graded rejection with an AUC of 0.833. In a human reader study it was non-inferior to conventional assessment and reduced interobserver variability and assessment time. Caveat: test cohorts from the US, Turkey and Switzerland are retrospective and the authors call for clinical trials.",
    "field": "medicine",
    "ai_technology": "Deep learning on whole-slide images for endomyocardial biopsies",
    "organization": "Harvard Medical School; Ege University; Broad Institute of MIT and Harvard; University of California San Diego",
    "source_url": "https://www.nature.com/articles/s41591-022-01709-2",
    "source_type": "paper",
    "publication_date": "2022-03-21",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-022-01709-2",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Rejection detection (AUC 0.962) was measured on a curated US dataset and independent test cohorts from Turkey and Switzerland, and a human reader study found the system non-inferior to conventional assessment.",
      "unknown": "The authors state that clinical trials are needed to establish the efficacy of AI-assisted biopsy assessment.",
      "ai_role": "Deep learning system analysed gigapixel endomyocardial biopsy whole-slide images to detect, subtype and grade rejection.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-022-01709-2&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-04-swarm-learning-histopathology",
    "title": "Swarm learning predicts colorectal cancer mutations from histopathology slides",
    "summary": "Swarm learning, which trains models across partner sites without moving patient data, was applied to over 5,000 patients to predict BRAF status and microsatellite instability from H&E colorectal cancer slides. Swarm-trained models outperformed most locally trained models and matched models trained on merged datasets, and were validated on two independent UK datasets. Caveat: the outcome is a research-grade prediction from slides, not a clinical test.",
    "field": "medicine",
    "ai_technology": "Swarm learning with deep learning on histopathology images",
    "organization": "University of Oxford; Belfast Health and Social Care Trust; University of Leeds",
    "source_url": "https://www.nature.com/articles/s41591-022-01768-5",
    "source_type": "paper",
    "publication_date": "2022-04-25",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-022-01768-5",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Swarm-trained models for BRAF status and microsatellite instability from H&E slides were compared with local and merged-data models and validated on two independent UK datasets.",
      "unknown": "Not reported; the abstract includes no limitations statement.",
      "ai_role": "Swarm learning trained AI models across partner sites without moving patient data, predicting molecular alterations from colorectal cancer histopathology slides.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-022-01768-5&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-05-mental-health-crisis-ehr-model",
    "title": "Machine learning model predicts mental health crises from health records",
    "summary": "A model using electronic health records to flag mental health crises over 28 days reached an AUROC of 0.797, with 58% sensitivity at 85% specificity. In a six-month prospective workflow study at one NHS trust, clinicians rated predictions useful for managing caseloads or mitigating crisis risk in 64% of cases. This measures clinician-reported usefulness, not a demonstrated reduction in crises; generalizability remains limited by the single-trust cohort, and precision was modest (AUPRC 0.159). The study includes authors affiliated with the developer Koa Health.",
    "field": "medicine",
    "ai_technology": "Machine learning model on electronic health records",
    "organization": "Koa Health; University of Warwick; Universitat Pompeu Fabra; Birmingham and Solihull Mental Health NHS Foundation Trust",
    "source_url": "https://www.nature.com/articles/s41591-022-01811-5",
    "source_type": "paper",
    "publication_date": "2022-05-16",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-022-01811-5",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A six-month prospective workflow study at one NHS trust asked clinicians to rate predictions, which were judged useful for managing caseloads or mitigating crisis risk in 64% of cases; the AUROC was 0.797.",
      "unknown": "Not reported; the page gives no rating scale or AUROC for the prospective usefulness measure.",
      "ai_role": "Model using electronic health records to monitor patients for risk of a mental health crisis over 28 days.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-022-01811-5&resultType=core&format=json"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2022-08-parkinson-nocturnal-breathing-ai",
    "title": "AI detects Parkinson's disease from nocturnal breathing signals",
    "summary": "An attention-based AI model extracting breathing from radio waves during sleep detected Parkinson's disease with an AUC of 0.90 on held-out data and 0.85 on external sets, across 7,671 individuals. Its severity estimates correlated with MDS-UPDRS scores (R = 0.94), and it worked for at-home assessment. Caveat: largely retrospective data and no prospective diagnostic trial reported.",
    "field": "medicine",
    "ai_technology": "Attention-based deep learning on contactless radio-wave breathing signals",
    "organization": "Mayo Clinic; University of Rochester Medical Center; Massachusetts Institute of Technology",
    "source_url": "https://www.nature.com/articles/s41591-022-01932-x",
    "source_type": "paper",
    "publication_date": "2022-08-22",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-022-01932-x",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Parkinson's detection was tested on held-out data (AUC 0.90) and external test sets (AUC 0.85) across 7,671 individuals, and severity estimates were compared with MDS-UPDRS scores (R = 0.94).",
      "unknown": "The authors describe the findings as initial evidence that the model may be useful for risk assessment before clinical diagnosis.",
      "ai_role": "Attention-based AI model extracting breathing from radio waves during sleep to detect and grade Parkinson's disease.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-022-01932-x&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-09-desi-strong-lens-foundry",
    "title": "Spectroscopic confirmation of neural-network lens candidates from DESI imaging",
    "summary": "A residual neural network selected strong gravitational lens candidates in the DESI Legacy Imaging Surveys. VLT/MUSE spectroscopy of 76 candidates (223 spectra) confirmed 55 lenses with both lens and source redshifts, 15 with lens redshift only, and found 6 non-lenses. Caveat: the source is an arXiv preprint (submitted 22 September 2025) that does not name the network architecture or list institutions on its abstract page.",
    "field": "physics",
    "ai_technology": "residual neural network (convolutional classifier for strong lens candidates)",
    "organization": "DESI Collaboration",
    "source_url": "https://arxiv.org/abs/2509.18078",
    "source_type": "paper",
    "publication_date": "2025-09-22",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "VLT/MUSE spectroscopy of 76 neural-network-selected candidates (223 spectra) confirmed 55 lenses with both redshifts, 15 with lens redshift only, and 6 non-lenses.",
      "unknown": "Redshifts were obtained by manual identification of spectral features, and background source redshifts could not be determined for 15 targets.",
      "ai_role": "A residual neural network selected strong lens candidates in the DESI Legacy Imaging Surveys; the architecture is not described on the page.",
      "evidence_urls": [
        "https://arxiv.org/abs/2509.18078"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-02-polybot-electronic-polymers",
    "title": "Polybot: an autonomous lab finds processing recipes for conductive polymer films",
    "summary": "Polybot used importance-guided Bayesian optimization to select and execute experiments across seven film-processing variables. The study reports recipes for transparent conductive polymer films with mean conductivity above 4,500 S/cm; morphology and feature-importance analyses identified influential processing factors. These are laboratory fabrication results, not demonstrated industrial-scale performance.",
    "field": "materials",
    "ai_technology": "Importance-guided Bayesian optimization with automated synthesis and characterization",
    "organization": "Argonne National Laboratory; University of Chicago",
    "source_url": "https://doi.org/10.1038/s41467-024-55655-3",
    "source_type": "paper",
    "publication_date": "2025-02-17",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Automated processing experiments on conductive polymer films; recipes yielded reported mean conductivity exceeding 4,500 S/cm.",
      "unknown": "Abstract and publication metadata checked. Detailed campaign counts, baseline comparisons and industrial durability were not assessed.",
      "ai_role": "Importance-guided Bayesian optimization selected processing conditions and balanced multiple measured objectives.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41467-024-55655-3%22&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-flower-reaction-mechanisms",
    "title": "FlowER predicts ten-step pyrazole pathways consistent with known product selectivity",
    "summary": "FlowER predicted two ten-step pathways to pyrazole products already reported in 2024 patents. Human researchers supplied acidic aqueous conditions omitted from the reaction database. Quantum-chemical calculations of intermediate energies favored the experimentally major product, consistent with the reported 8:2 product ratio; the kinetic interpretation assumes the Bell–Evans–Polanyi principle. This is a computational explanation of existing outcomes, with no new reaction experiment or claim of a previously unknown mechanism. Review covered the February 18, 2025 preprint results and the August 20 journal paper’s abstract, metadata and extended-data captions; the journal main text was not assessed.",
    "field": "chemistry",
    "ai_technology": "FlowER (flow matching on bond-electron matrices)",
    "organization": "Massachusetts Institute of Technology; Kookmin University",
    "source_url": "https://doi.org/10.1038/s41586-025-09426-9",
    "source_type": "paper",
    "publication_date": "2025-08-20",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Specific pyrazole-pathway predictions compared with known patent products and supplemented by calculated intermediate free energies.",
      "unknown": "No new experimental confirmation; mechanistic priority not established. The kinetic argument assumes Bell–Evans–Polanyi reasoning; intermediate free energies are not measured transition-state barriers.",
      "ai_role": "FlowER generated candidate electron-redistribution pathways; authors supplied conditions and conducted quantum-chemical analysis.",
      "evidence_urls": [
        "https://arxiv.org/html/2502.12979v1",
        "https://www.nature.com/articles/s41586-025-09426-9?error=cookies_not_supported"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-01-glavitu-glacier-mapping",
    "title": "GlaViTU estimates decadal glacier-area decline from satellite imagery",
    "summary": "GlaViTU combined convolutional and transformer models to map glacier outlines from satellite images. The original supplement applies it to the Aletsch complex between 1998 and 2023 and Hardangerjøkulen between 1996 and 2022, estimating area declines in both. The reported uncertainty bands are broad and overlap across dates. These are retrospective, model-derived geophysical estimates, not independent field measurements. Review covered the paper abstract, metadata, relevant main-text discussion and the first supplementary page.",
    "field": "climate",
    "ai_technology": "GlaViTU (Glacier-VisionTransformer-U-Net)",
    "organization": "University of Twente; University of Oslo",
    "source_url": "https://doi.org/10.1038/s41467-024-54956-x",
    "source_type": "paper",
    "publication_date": "2025-01-02",
    "added_date": "2026-11-14",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Retrospective satellite-image mapping and regional glacier-area estimates at two periods; model-generalization and expert-outline comparisons.",
      "unknown": "Supplementary area estimates have broad, overlapping 95% uncertainty bands. Only the relevant first supplementary page was reviewed; no field campaign or independent glacier-area estimate was performed in this review.",
      "ai_role": "GlaViTU mapped glacier outlines with convolutional-transformer models and inference-time bias optimization.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11695715/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11695715/supplementaryFiles"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-02-fragfold-identifies-inhibitory-fragments-within-bacterial-proteins",
    "title": "FragFold identifies inhibitory fragments within bacterial proteins",
    "summary": "MIT researchers used FragFold, an AlphaFold-based approach, to predict fragments that bind to and inhibit full-length proteins in E. coli. The researchers experimentally confirmed more than half of its binding or inhibition predictions, including cases without prior structural interaction data.",
    "field": "biology",
    "ai_technology": "FragFold, using AlphaFold",
    "organization": "Massachusetts Institute of Technology",
    "source_url": "https://news.mit.edu/2025/ai-system-fragfold-predicts-protein-fragments-0220",
    "source_type": "press release",
    "publication_date": "2025-02-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2322412122",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The researchers experimentally confirmed more than half of its binding or inhibition predictions, including cases without prior structural interaction data.",
      "unknown": "The institutional report says more than half of predictions were accurate; this is a study-specific validation result, not a general success rate across proteins or organisms.",
      "ai_role": "Researchers computationally fragmented bacterial proteins and used AlphaFold predictions to nominate inhibitory fragments; experiments tested binding and inhibition. The AI supplied structural predictions, not experimental verification.",
      "evidence_urls": [
        "https://news.mit.edu/2025/ai-system-fragfold-predicts-protein-fragments-0220"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-slide-uncovers-a-keratinocyte-centered-signal-in-systemic-sclerosis",
    "title": "SLIDE uncovers a keratinocyte-centered signal in systemic sclerosis",
    "summary": "An interpretable machine-learning analysis of multi-omic systemic-sclerosis data identified an unexplored keratinocyte-centered signature, validated by protein staining, and a previously unreported interaction between myeloid-cell and fibroblast states and macrophage HLA signaling. The same study also analyzed CD4 T-cell clonal expansion, but this record treats the paper's disease-mechanism findings as one discovery.",
    "field": "biology",
    "ai_technology": "SLIDE (Significant Latent Factor Interaction Discovery and Exploration)",
    "organization": "University of Pittsburgh; and Cornell University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11588359/",
    "source_type": "paper",
    "publication_date": "2024-02-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-024-02175-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An interpretable machine-learning analysis of multi-omic systemic-sclerosis data identified an unexplored keratinocyte-centered signature, validated by protein staining, and a previously unreported interaction between myeloid-cell and fibroblast states and macrophage HLA signaling.",
      "unknown": "The paper reports computationally identified relationships and validation of one signature by protein staining; these findings do not by themselves establish causal mechanisms or clinical utility.",
      "ai_role": "SLIDE integrated single-cell and spatial multi-omic data, identified interacting latent factors associated with systemic-sclerosis disease severity, and surfaced candidate mechanisms; the keratinocyte signal was checked with protein staining by the study team.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11588359/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-ai-helps-resolve-an-unknown-regulatory-mechanism-in-bacterial-acetyl-coa-synthetase",
    "title": "AI helps resolve an unknown regulatory mechanism in bacterial acetyl-CoA synthetase",
    "summary": "Researchers at the University of Greifswald resolved how a post-translational modification regulates Bacillus subtilis acetyl-CoA synthetase, a central metabolic enzyme. The study used AI-based structural prediction to explain the molecular mechanism at near-atomic detail.",
    "field": "biology",
    "ai_technology": "AlphaFold2 and AI-assisted structural analysis",
    "organization": "University of Greifswald",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11255334/",
    "source_type": "paper",
    "publication_date": "2024-07-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-49952-0",
    "relevance": {
      "impact": 3,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers at the University of Greifswald resolved how a post-translational modification regulates Bacillus subtilis acetyl-CoA synthetase, a central metabolic enzyme.",
      "unknown": "The institutional announcement says the regulatory mechanism is evolutionarily conserved and may matter in higher organisms; that implication is not itself a demonstration in those organisms.",
      "ai_role": "AlphaFold2 structural predictions supported the researchers' mechanistic interpretation of how the enzyme is regulated; the source describes a collaboration with biochemical and structural experiments, not an AI-only result.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11255334/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2025-03-ai-structural-modeling-helps-map-nearly-300-metabolite-binding-interactions-in-e-coli",
    "title": "AI structural modeling helps map nearly 300 metabolite-binding interactions in E. coli",
    "summary": "A cross-institutional study combined chemoproteomics with AI-driven structural modeling to identify nearly 300 small-molecule ligands and their binding sites on proteins in E. coli. The map adds experimentally grounded candidate links between bacterial metabolites and proteins involved in core cellular processes.",
    "field": "biology",
    "ai_technology": "AI-driven structural modeling (Kozakov laboratory; specific model not named in institutional announcement)",
    "organization": "Oregon Health & Science University; and Stony Brook University",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39862855/",
    "source_type": "paper",
    "publication_date": "2025-03-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2025.01.003",
    "relevance": {
      "impact": 3,
      "validation": 1,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The map adds experimentally grounded candidate links between bacterial metabolites and proteins involved in core cellular processes.",
      "unknown": "The cited count is nearly 300 ligands and binding sites; the announcement does not establish that every interaction is a newly discovered physiological mechanism. No named AI model was verified.",
      "ai_role": "AI structural modeling contributed to identifying and localizing metabolite-protein interactions alongside chemoproteomic experiments; the institutional source does not name the exact AI model.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39862855/"
      ],
      "test_type": "not reported"
    }
  },
  {
    "id": "2025-02-protgps-reveals-sequence-signals-that-direct-proteins-to-subcellular-compartments",
    "title": "ProtGPS reveals sequence signals that direct proteins to subcellular compartments",
    "summary": "ProtGPS, a protein language model and generative sequence-design system, helped identify amino-acid sequence features that guide proteins to subcellular compartments. The paper applies the model to predict disease-mutation-associated localization shifts and reports cellular testing of 20 examples.",
    "field": "biology",
    "ai_technology": "ProtGPS protein language model and generative design algorithm",
    "organization": "Massachusetts Institute of Technology; and Whitehead Institute",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12034300/",
    "source_type": "paper",
    "publication_date": "2025-02-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adq2634",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper applies the model to predict disease-mutation-associated localization shifts and reports cellular testing of 20 examples.",
      "unknown": "The 20 tested mutations are a sample of a much larger computational screen; the predictions do not establish causal disease mechanisms in patients.",
      "ai_role": "ProtGPS predicted compartment localization and mutation-associated shifts, and its generative component designed sequences for target localization. Researchers experimentally tested 20 mutation predictions in cells and reported confirmation.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12034300/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-celltransformer-maps-previously-uncatalogued-regions-of-the-mouse-brain",
    "title": "CellTransformer maps previously uncatalogued regions of the mouse brain",
    "summary": "A transformer model applied to spatial transcriptomics mapped fine-grained mouse-brain tissue domains and revealed hundreds of regions absent from existing anatomical atlases. The study evaluated the data-driven domains against the Allen Mouse Brain Common Coordinate Framework and reported strong alignment alongside the novel subdivisions.",
    "field": "biology",
    "ai_technology": "CellTransformer",
    "organization": "University of California, San Francisco; and Allen Institute",
    "source_url": "https://www.nature.com/articles/s41467-025-64259-4",
    "source_type": "paper",
    "publication_date": "2025-10-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-64259-4",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study evaluated the data-driven domains against the Allen Mouse Brain Common Coordinate Framework and reported strong alignment alongside the novel subdivisions.",
      "unknown": "The new domains are data-driven subdivisions; their biological meaning and generalization beyond the analyzed mouse-brain dataset require further study. Nature's paper page was accessible in search and as a PDF abstract, but its HTML reader view redirected to a cookie wall.",
      "ai_role": "The transformer analyzed cell-neighborhood patterns in spatial transcriptomics and assigned tissue domains without manual annotation; human researchers compared outputs with the Allen reference atlas.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-64259-4"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-sparcs-screen-identifies-a-golgi-ph-regulator-of-sting-signaling",
    "title": "SPARCS screen identifies a Golgi-pH regulator of STING signaling",
    "summary": "A genome-wide imaging and CRISPR screen of about 70 million cells identified GPR89A and GPR89B (encoding GPHR) as important for maintaining Golgi acidity and early STING trafficking and activation. The study also found a previously undescribed autophagy phenotype associated with EI24 and linked it to altered autophagosome formation.",
    "field": "biology",
    "ai_technology": "SPARCS platform with machine-learning image analysis",
    "organization": "Ludwig Maximilian University of Munich; Helmholtz Munich, Max Planck Institute of Biochemistry and Technical University of Munich",
    "source_url": "https://www.helmholtz-munich.de/en/icb/news-detail/new-method-links-cell-images-to-genetic-changes",
    "source_type": "press release",
    "publication_date": "2026-10-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2026.09.021",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A genome-wide imaging and CRISPR screen of about 70 million cells identified GPR89A and GPR89B (encoding GPHR) as important for maintaining Golgi acidity and early STING trafficking and activation.",
      "unknown": "The study used pools of at least 500 excised cells for sequencing, so it could not map individual images to individual guide RNAs. The paper describes GPHR as a regulator of STING trafficking and signaling; therapeutic implications remain untested.",
      "ai_role": "Machine-learning image analysis classified phenotypes in genome-wide CRISPR screens; SPARCS recovered known autophagy genes and surfaced EI24 and GPHR-linked findings. Follow-up molecular profiling and experiments by the researchers examined these hits.",
      "evidence_urls": [
        "https://www.helmholtz-munich.de/en/icb/news-detail/new-method-links-cell-images-to-genetic-changes"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-ai-guided-design-enables-e-coli-ribosomal-proteins-without-isoleucine",
    "title": "AI-guided design enables E. coli ribosomal proteins without isoleucine",
    "summary": "Researchers used computational protein design and generative AI to recode ribosomal proteins in E. coli so they function without isoleucine, then assembled an engineered strain that remained viable and genomically stable over more than 450 laboratory generations. The result shows a reduced amino-acid alphabet can support core translation machinery in this bacterium; the organism still uses isoleucine elsewhere.",
    "field": "biology",
    "ai_technology": "Protein design models including ESM2, MSATr, ProteinMPNN and AfDesign",
    "organization": "Columbia University; Massachusetts Institute of Technology and Harvard University",
    "source_url": "https://magazine.columbia.edu/article/could-life-be-made-simpler",
    "source_type": "other",
    "publication_date": "2026-04-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.aeb5171",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers used computational protein design and generative AI to recode ribosomal proteins in E. coli so they function without isoleucine, then assembled an engineered strain that remained viable and genomically stable over more than 450 laboratory generations.",
      "unknown": "This is not a whole 19-amino-acid organism: isoleucine was eliminated from the engineered ribosomal proteins, while the rest of the cell still uses it. Growth was slower than wild type according to the paper.",
      "ai_role": "The researchers used AI-guided computational design to propose amino-acid substitutions that allowed selected ribosomal proteins to function after removing isoleucine; the strain was assembled and experimentally tested by the team.",
      "evidence_urls": [
        "https://magazine.columbia.edu/article/could-life-be-made-simpler"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-ai-designed-enhancers-control-gene-expression-in-healthy-mammalian-cells",
    "title": "AI-designed enhancers control gene expression in healthy mammalian cells",
    "summary": "CRG researchers measured more than 64,000 synthetic enhancers across seven stages of blood-cell development and found that enhancer combinations can activate genes in one cell type but repress them in another. The AI model used those measurements to design previously unseen regulatory DNA sequences that drove the intended cell-selective reporter expression in mouse blood cells.",
    "field": "biology",
    "ai_technology": "Machine-learning enhancer design model developed by Lars Velten's group",
    "organization": "Centre for Genomic Regulation; and Barcelona Collaboratorium",
    "source_url": "https://www.crg.eu/en/news/ai-designed-dna-controls-genes-healthy-mammalian-cells-first-time",
    "source_type": "press release",
    "publication_date": "2025-05-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2025.04.017",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "CRG researchers measured more than 64,000 synthetic enhancers across seven stages of blood-cell development and found that enhancer combinations can activate genes in one cell type but repress them in another.",
      "unknown": "The enhancer designs are proof-of-concept reporter regulators, tested in mouse blood cells; this is not evidence of a safe or effective gene therapy. The paper's Cell page redirected to a cookie wall, so detailed findings rely on the CRG's linked official summary.",
      "ai_role": "Researchers used large-scale enhancer activity measurements in healthy blood-cell models to train the model, which generated synthetic DNA enhancers for specified cell-type expression patterns; experiments tested the designs in mouse blood cells.",
      "evidence_urls": [
        "https://www.crg.eu/en/news/ai-designed-dna-controls-genes-healthy-mammalian-cells-first-time"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-machine-learning-identifies-the-first-enzyme-for-phenazinethiol-conjugation",
    "title": "Machine learning identifies the first enzyme for phenazine–thiol conjugation",
    "summary": "A genome-informed machine-learning workflow (ML-CITO) used contrastive learning over protein-language-model representations and genomic context, starting from 14 known phenazine-modifying sequences, to identify PTC. Biochemical characterization showed PTC catalyzes glutathione-dependent phenazine modification, a reaction previously assumed to occur non-enzymatically.",
    "field": "biology",
    "ai_technology": "ML-CITO (genome-context-informed data augmentation plus protein-language-space contrastive learning)",
    "organization": "California Institute of Technology; University of California, Los Angeles; AITHYRA / Austrian Academy of Sciences",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13419639/",
    "source_type": "paper",
    "publication_date": "2026-08-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2607571123",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A genome-informed machine-learning workflow (ML-CITO) used contrastive learning over protein-language-model representations and genomic context, starting from 14 known phenazine-modifying sequences, to identify PTC.",
      "unknown": "The paper reports that deleting the PTC gene did not cause a strong fitness disadvantage under tested conditions. The demonstrated chemistry and substrate context should not be generalized to other phenazine reactions.",
      "ai_role": "The model ranked candidate phenazine-interacting proteins from sparse known examples; the team then expressed and biochemically tested the candidate enzyme. AI nomination was a direct step in discovering the previously unknown enzymatic reaction.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13419639/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-defensepredictor-uncovers-dozens-of-previously-unknown-bacterial-anti-phage-systems",
    "title": "DefensePredictor uncovers dozens of previously unknown bacterial anti-phage systems",
    "summary": "DefensePredictor, a machine-learning classifier using protein-language-model embeddings, nominated candidate defense genes across diverse E. coli strains. Of 94 predicted gene clusters selected for testing, experiments validated 42 previously unknown systems; analyses of 1,000 prokaryotic genomes also revealed thousands of uncharacterized protein clusters without homology to known systems.",
    "field": "biology",
    "ai_technology": "DefensePredictor, gradient-boosted classification over protein-language-model embeddings and genomic context",
    "organization": "Massachusetts Institute of Technology; and Johns Hopkins University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13092281/",
    "source_type": "paper",
    "publication_date": "2026-04-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adv7924",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Of 94 predicted gene clusters selected for testing, experiments validated 42 previously unknown systems; analyses of 1,000 prokaryotic genomes also revealed thousands of uncharacterized protein clusters without homology to known systems.",
      "unknown": "The 42 validated systems came from a selected subset of predictions, not all computationally proposed systems; genome-wide counts are predicted protein clusters, not experimentally tested systems. The source reports a 45% validation rate for 94 tested units.",
      "ai_role": "The machine-learning model identified likely defense proteins beyond conventional defense-island co-localization methods, guiding experimental cloning and phage-challenge validation. Validation confirmed 42 new systems among 94 tested candidates.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13092281/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-01-interpretable-multi-omics-machine-learning-reveals-unexpected-drug-mechanisms-in-huntington-disease-models",
    "title": "Interpretable multi-omics machine learning reveals unexpected drug mechanisms in Huntington disease models",
    "summary": "An interpretable machine-learning model integrated transcriptomic, epigenomic, metabolomic and proteomic data from Huntington-disease cell models to identify previously unknown mechanisms for protective compounds. The analysis predicted autophagy activation for one compound group and mitochondrial-respiration inhibition for another; follow-up imaging, biochemical and energetics assays experimentally confirmed these effects in the cell model.",
    "field": "biology",
    "ai_technology": "Interpretable network-optimization model integrating multi-omics over a molecular-interaction network",
    "organization": "University of California San Diego; and collaborators",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC6976599/",
    "source_type": "paper",
    "publication_date": "2020-01-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41598-020-57691-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The analysis predicted autophagy activation for one compound group and mitochondrial-respiration inhibition for another; follow-up imaging, biochemical and energetics assays experimentally confirmed these effects in the cell model.",
      "unknown": "Findings were experimentally established in STHdhQ111 murine striatal cells, with autophagy effects also checked in human SH-SY5Y and HEK293 cell lines. The paper notes that the model prioritizes pathways but does not resolve every causal step or demonstrate therapeutic efficacy in patients.",
      "ai_role": "The model integrated multiple omics and prioritized cellular pathways as candidate mechanisms of action; researchers tested those predictions with cellular imaging, biochemical assays and energetics measurements. It revealed unexpected mechanisms for compounds with unrelated known targets.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC6976599/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-progeni-predicts-candidate-melanoma-and-colorectal-cancer-targets-validated-in-cell-models",
    "title": "Progeni predicts candidate melanoma and colorectal-cancer targets validated in cell models",
    "summary": "Progeni combines a probabilistic knowledge graph with graph neural networks to rank disease-associated molecular targets. The authors selected six candidates lacking literature support and reported that knockdown of HSP90AB1 and MME reduced proliferation in a murine melanoma line, while knockdown of ADCY5, ADRA2A and EEF2 reduced proliferation in a murine colorectal-cancer line.",
    "field": "biology",
    "ai_technology": "Progeni probabilistic knowledge graph and graph neural network",
    "organization": "Tsinghua University; Westlake University; Jilin University; Silexon AI Technology",
    "source_url": "https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1011945",
    "source_type": "paper",
    "publication_date": "2024-04-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pcbi.1011945",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors selected six candidates lacking literature support and reported that knockdown of HSP90AB1 and MME reduced proliferation in a murine melanoma line, while knockdown of ADCY5, ADRA2A and EEF2 reduced proliferation in a murine colorectal-cancer line.",
      "unknown": "The functional validations were in B16F10 and MC38 murine cell lines, not human tumor samples or clinical settings. The paper evaluates a model and validates selected candidates; it does not establish these as clinically effective therapeutic targets.",
      "ai_role": "The graph neural network ranked new target hypotheses from heterogeneous biological networks and literature co-occurrence evidence. Researchers then tested selected predictions by shRNA knockdown in representative murine cancer cell lines.",
      "evidence_urls": [
        "https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1011945"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-10-lighthouse-identifies-riboflavin-5-monophosphate-as-a-ppat-inhibitor",
    "title": "LIGHTHOUSE identifies riboflavin 5′-monophosphate as a PPAT inhibitor",
    "summary": "LIGHTHOUSE, a deep-learning model using protein sequence and chemical representations, nominated riboflavin 5′-monophosphate as an inhibitor of PPAT, a target without reported inhibitors in the paper. A biochemical assay measured PPAT inhibition with an IC50 of 7.5 μM, close to the model’s predicted range; the same study also tested predicted antiviral compounds in cell models.",
    "field": "chemistry",
    "ai_technology": "LIGHTHOUSE (ensemble graph-based deep-learning platform)",
    "organization": "Kyushu University; and collaborators",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9549714/",
    "source_type": "paper",
    "publication_date": "2022-10-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.isci.2022.105314",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "LIGHTHOUSE, a deep-learning model using protein sequence and chemical representations, nominated riboflavin 5′-monophosphate as an inhibitor of PPAT, a target without reported inhibitors in the paper.",
      "unknown": "The result is an in-vitro enzyme assay, not a validated anticancer treatment. The authors note LIGHTHOUSE can produce false positives; COVID-19 results were also cell-based and had no clinical evidence.",
      "ai_role": "The model scored protein–compound relationships without requiring a 3D target structure and prioritized the compound from >333,000 candidates for experimental follow-up. The authors assayed PPAT activity and confirmed inhibition in vitro.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9549714/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-01-deepdtnet-repurposes-topotecan-as-an-rort-inhibitor-for-multiple-sclerosis-models",
    "title": "deepDTnet repurposes topotecan as an RORγt inhibitor for multiple-sclerosis models",
    "summary": "The deep-learning network deepDTnet predicted retinoic-acid-receptor-related orphan receptor gamma-t (RORγt) as a previously unrecognized target of topotecan. Biochemical assays showed direct inhibition (IC50 0.43 μM), and topotecan improved outcomes in a mouse experimental autoimmune encephalomyelitis model of multiple sclerosis.",
    "field": "medicine",
    "ai_technology": "deepDTnet, heterogeneous drug–gene–disease network embedding with deep learning",
    "organization": "Cleveland Clinic; Case Western Reserve University, Harvard Medical School and collaborators",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/34123272/",
    "source_type": "paper",
    "publication_date": "2020-01-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/c9sc04336e",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Biochemical assays showed direct inhibition (IC50 0.43 μM), and topotecan improved outcomes in a mouse experimental autoimmune encephalomyelitis model of multiple sclerosis.",
      "unknown": "Evidence is preclinical: direct target and mouse-model assays do not establish safety or effectiveness in people with multiple sclerosis. Topotecan is an established anticancer drug, and its new target/indication remains a repurposing hypothesis.",
      "ai_role": "The model integrated 15 types of chemical, genomic, phenotypic and cellular network profiles to rank new drug targets. It nominated RORγt for topotecan, which the team tested using direct binding/transcriptional assays and a mouse EAE disease model.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/34123272/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-03-clean-experimentally-identifies-sam-hydrolase-activity-in-previously-uncharacterized-proteins",
    "title": "CLEAN experimentally identifies SAM hydrolase activity in previously uncharacterized proteins",
    "summary": "Researchers at the University of Illinois Urbana–Champaign developed CLEAN, a contrastive-learning enzyme-function predictor, and used it to identify the SAM hydrolase activity of the previously uncharacterized protein TTHA0338. In vitro HPLC–MS and enzyme-kinetic experiments also corrected the annotation of MJ1651 and confirmed the multiple activities of SsFIA.",
    "field": "biology",
    "ai_technology": "CLEAN (contrastive learning-based enzyme annotation model)",
    "organization": "University of Illinois Urbana-Champaign; Cornell University; Georgia Institute of Technology",
    "source_url": "https://doi.org/10.1126/science.adf2465",
    "source_type": "paper",
    "publication_date": "2023-03-31",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In vitro HPLC–MS and enzyme-kinetic experiments also corrected the annotation of MJ1651 and confirmed the multiple activities of SsFIA.",
      "unknown": "The study validated a small, selected set of proteins, and also showed CLEAN can assign functions to non-enzymes; the demonstrated result is specific to the tested proteins. TTHA0338 is one of multiple results reported by this paper; record describes the previously unknown function rather than treating every model benchmark as a separate discovery.",
      "ai_role": "CLEAN predicted enzyme commission functions from protein sequences; researchers then validated the predicted activity experimentally. It was the model that nominated TTHA0338 as a SAM hydrolase, not an autonomous discovery or validation system.",
      "evidence_urls": [
        "https://doi.org/10.1126/science.adf2465"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-08-deepamp-designs-broad-spectrum-antimicrobial-peptides-active-against-resistant-bacteria",
    "title": "deepAMP designs broad-spectrum antimicrobial peptides active against resistant bacteria",
    "summary": "Researchers at Hunan University and collaborators used the peptide language model deepAMP to optimize antimicrobial peptides and experimentally tested 29 candidates. More than 90% inhibited Gram-positive and Gram-negative bacteria better than the reference peptide penetratin; three candidates showed reduced resistance emergence in serial-passage assays and activity in a mouse Pseudomonas wound-infection model.",
    "field": "medicine",
    "ai_technology": "deepAMP peptide language model (deepAMP-general, deepAMP-AOM, deepAMP-POM and deepAMP-predict)",
    "organization": "Hunan University; Cleveland Clinic; National Cancer Institute; Tel Aviv University",
    "source_url": "https://doi.org/10.1038/s41467-024-51933-2",
    "source_type": "paper",
    "publication_date": "2024-08-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers at Hunan University and collaborators used the peptide language model deepAMP to optimize antimicrobial peptides and experimentally tested 29 candidates.",
      "unknown": "This is a separate discovery study from the AMPSphere microbiome-mining paper already in the tracker: deepAMP optimizes peptide sequences, whereas AMPSphere mined microbiomes. Testing involved 29 candidates and one female mouse wound model; this is preclinical evidence, not clinical efficacy. The article itself calls for more animal-model validation.",
      "ai_role": "The generative and ranking models optimized candidate peptide sequences for antimicrobial potency and membrane-disrupting activity. Researchers synthesized and tested selected sequences in vitro and in a mouse wound infection experiment.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41467-024-51933-2"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-06-alphafold-assisted-structural-clustering-discovers-new-deaminase-functions-and-compact-base-editing-enzymes",
    "title": "AlphaFold-assisted structural clustering discovers new deaminase functions and compact base-editing enzymes",
    "summary": "The team used AlphaFold2 structure predictions to cluster the deaminase family and found that most proteins in a DddA-like clade were not double-stranded DNA deaminases. The work characterized smaller single-strand-specific deaminases, developed a compact cytidine deaminase suitable for packaging a base editor into one AAV, and demonstrated robust editing by a clade member in soybean.",
    "field": "biology",
    "ai_technology": "AlphaFold2-assisted structure-based protein-family clustering",
    "organization": "Chinese Academy of Sciences; Qi Biodesign; BGI-Qingdao; Aarhus University",
    "source_url": "https://doi.org/10.1016/j.cell.2023.05.041",
    "source_type": "paper",
    "publication_date": "2023-06-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The team used AlphaFold2 structure predictions to cluster the deaminase family and found that most proteins in a DddA-like clade were not double-stranded DNA deaminases.",
      "unknown": "This paper is an AI-assisted protein-discovery and engineering study; experimental base editing demonstrates the utility of selected deaminases, while therapeutic application remains prospective. The study was followed by a 2024 correction/erratum; the record references the original research article, DOI 10.1016/j.cell.2023.05.041, not the erratum.",
      "ai_role": "AlphaFold2 predicted protein structures that enabled structure-similarity clustering of a full deaminase family. The clustering guided discovery of functional diversity and selection of enzymes, which the researchers characterized experimentally and used in cell, animal and plant base-editing tests.",
      "evidence_urls": [
        "https://doi.org/10.1016/j.cell.2023.05.041"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-03-catapro-assisted-enzyme-mining-identifies-a-high-activity-sphingobium-enzyme-for-vanillin-production",
    "title": "CataPro-assisted enzyme mining identifies a high-activity Sphingobium enzyme for vanillin production",
    "summary": "CataPro, a deep-learning model using pretrained protein-language representations and molecular fingerprints to predict enzyme kinetics, was combined with sequence and structure-based filtering to screen candidate enzymes for converting 4-vinylguaiacol to vanillin. The team experimentally identified Sphingobium sp. CSO (SsCSO), whose measured activity was 19.53 times that of the initial enzyme CSO2, and then used CataPro-guided engineering to increase SsCSO activity a further 3.34-fold.",
    "field": "biology",
    "ai_technology": "CataPro",
    "organization": "Shandong University; Shanghai Zelixir Biotech; Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences",
    "source_url": "https://doi.org/10.1038/s41467-025-58038-4",
    "source_type": "paper",
    "publication_date": "2025-03-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The team experimentally identified Sphingobium sp.",
      "unknown": "The reported 19.53× and 3.34× figures are relative activities for the tested 4-vinylguaiacol oxidation assay; they do not establish industrial-scale yield or process performance. Model predictions were used alongside conventional sequence/structure methods, not as the sole discovery method.",
      "ai_role": "CataPro predicted kinetic parameters to prioritize candidate carotenoid-cleavage-oxygenase family enzymes and rank variants; researchers combined its predictions with BLAST, structural similarity and laboratory enzyme assays. It materially narrowed the enzyme search and guided mutations, while experiments established measured activity.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41467-025-58038-4"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-01-ai-photoswitchable-crispri-screening-identifies-new-regulators-of-tfeb-localization-during-starvation",
    "title": "AI-photoswitchable CRISPRi screening identifies new regulators of TFEB localization during starvation",
    "summary": "An image-classification-guided pooled CRISPRi screen mapped factors that affect TFEB movement from nucleus to cytosol during prolonged starvation. The screen identified multiple candidate genes, and follow-up assays independently validated 21 of 64 neural-network-called hits; experiments on CREB5 further linked it to TFEB retention and starvation-associated autophagy/lysosomal protein expression.",
    "field": "biology",
    "ai_technology": "AI-photoswitchable screening (AI-PS), using SVM classifiers and a convolutional neural network",
    "organization": "Janelia Research Campus, Howard Hughes Medical Institute; National Institute of Neurological Disorders and Stroke, NIH",
    "source_url": "https://rupress.org/jcb/article/220/2/e202006180/211694/Image-based-pooled-whole-genome-CRISPRi-screening",
    "source_type": "paper",
    "publication_date": "2021-01-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1083/jcb.202006180",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The screen identified multiple candidate genes, and follow-up assays independently validated 21 of 64 neural-network-called hits; experiments on CREB5 further linked it to TFEB retention and starvation-associated autophagy/lysosomal protein expression.",
      "unknown": "Only 21 of 64 neural-network-called TFEB hits were independently validated in secondary assays (about 32%); candidate effectors should not be treated as all confirmed. Results use cultured U2OS cells and prolonged starvation, and do not establish a physiological role in an organism. CREB5 received targeted follow-up but the overall screen has a modest validation rate.",
      "ai_role": "SVM and CNN models classified single-cell images by subcellular protein localization; the system photoactivated cells with the target phenotype for FACS isolation and gRNA sequencing in pooled CRISPRi screening. AI enabled the genome-scale genotype-to-subcellular-phenotype screen.",
      "evidence_urls": [
        "https://rupress.org/jcb/article/220/2/e202006180/211694/Image-based-pooled-whole-genome-CRISPRi-screening"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-machine-learning-guided-mining-identifies-91-previously-unknown-pet-hydrolases-including-acid-active-enzymes",
    "title": "Machine-learning-guided mining identifies 91 previously unknown PET hydrolases, including acid-active enzymes",
    "summary": "Researchers iteratively combined sequence mining, machine-learning activity/property predictors and high-throughput assays to screen natural enzyme homologs. They experimentally identified 91 previously unknown PET hydrolases among the 212 candidates assayed; 35 had activity on crystalline PET at pH 4.5, and four matched or exceeded a benchmark enzyme under that screening condition.",
    "field": "biology",
    "ai_technology": "Three-round iterative machine-learning pipeline (including VAE/rank model, supervised activity predictors, ESM embeddings and property predictors)",
    "organization": "National Renewable Energy Laboratory; BOTTLE Consortium; Agile BioFoundry; SLAC National Accelerator Laboratory; Harvard Medical School",
    "source_url": "https://doi.org/10.1021/acscatal.5c03460",
    "source_type": "paper",
    "publication_date": "2025-09-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers iteratively combined sequence mining, machine-learning activity/property predictors and high-throughput assays to screen natural enzyme homologs.",
      "unknown": "The article defines PET hydrolase by demonstrated PET-depolymerizing activity and notes that activity may be promiscuous relative to the enzyme’s native role. Assays are small-scale and 48 hours; the authors say further enzyme engineering is needed for meaningful bioreactor studies. This is distinct from the catalogued AI-engineered FAST-PETase: it discovers natural homologs and diverse acid/temperature activity profiles rather than optimizing one known PETase scaffold.",
      "ai_role": "Machine-learning activity, thermal-stability and acid-tolerance predictions prioritized natural homologs across successive mining rounds; new assay results updated later predictors. Researchers expressed/purified over 200 candidates and measured PET hydrolysis to verify activity.",
      "evidence_urls": [
        "https://doi.org/10.1021/acscatal.5c03460"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-structure-based-virtual-screening-identifies-mammalian-enzymes-catalyzing-the-second-step-of-carnitine-biosynthesis",
    "title": "Structure-based virtual screening identifies mammalian enzymes catalyzing the second step of carnitine biosynthesis",
    "summary": "The OSMES workflow screened AlphaFold-modeled vitamin-B6-dependent enzymes by docking the substrate into active sites and ranking catalytically favorable conformations. It identified human SHMT1/2 as enzymes with hydroxytrimethyllysine-aldolase activity, validated in vitro; mouse Tha1 showed much higher catalytic efficiency, supporting a species difference in the pathway. The authors infer that humans may have lost a dedicated gene, with SHMT activity partly compensating.",
    "field": "biology",
    "ai_technology": "OSMES (One Substrate–Many Enzymes Screening), automated structure-based docking/ranking workflow using AlphaFold models",
    "organization": "University of Parma; University of Padua; Sapienza University of Rome; CNR Institute of Molecular Biology and Pathology",
    "source_url": "https://www.nature.com/articles/s41467-024-47466-3.pdf",
    "source_type": "paper",
    "publication_date": "2024-04-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-47466-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "It identified human SHMT1/2 as enzymes with hydroxytrimethyllysine-aldolase activity, validated in vitro; mouse Tha1 showed much higher catalytic efficiency, supporting a species difference in the pathway.",
      "unknown": "Enzyme activity was validated in vitro. The claim about loss of a dedicated human gene is an evolutionary inference, not a direct gene-knockout demonstration; the authors note unresolved pathway questions.",
      "ai_role": "Ranked candidate enzymes for a specified metabolic reaction using docked substrate conformations and catalytically favorable active-site geometry; guided biochemical testing that verified SHMT1/2 activity.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-47466-3.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-03-machine-learning-and-peptide-array-experiments-reveal-new-substrates-for-protein-modifying-enzymes",
    "title": "Machine learning and peptide-array experiments reveal new substrates for protein-modifying enzymes",
    "summary": "A hybrid workflow coupled machine-learning models with enzyme-treated peptide arrays to predict previously unexplored protein post-translational-modification sites. Experimental validation confirmed 37–43% of proposed sites across SET8 and SIRT1–7; mass spectrometry confirmed dynamic methylation for several SET8 substrates and deacetylation at 64 unique SIRT2 sites. The work also found altered SET8 substrate networks associated with breast-cancer missense mutations.",
    "field": "biology",
    "ai_technology": "Enzyme-specific ML-hybrid substrate prediction models trained from high-throughput peptide-array assays",
    "organization": "Carleton University; Ben-Gurion University of the Negev",
    "source_url": "https://www.nature.com/articles/s42004-025-01717-6.pdf",
    "source_type": "paper",
    "publication_date": "2025-03-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42004-025-01717-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimental validation confirmed 37–43% of proposed sites across SET8 and SIRT1–7; mass spectrometry confirmed dynamic methylation for several SET8 substrates and deacetylation at 64 unique SIRT2 sites.",
      "unknown": "Reported validation rate is 37–43% of proposed sites, so predictions are not uniformly confirmed. Results concern in-vitro enzyme assays and selected sites; disease relevance of the altered SET8 network is not established as a clinical mechanism.",
      "ai_role": "Learned enzyme-specific substrate preferences from experimental array measurements and prioritized novel PTM sites for experimental follow-up; this enabled the reported new SET8/SIRT substrate assignments.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42004-025-01717-6.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-machine-learning-guided-cell-free-engineering-creates-specialized-amide-synthetases-for-pharmaceutical-molecules",
    "title": "Machine learning-guided cell-free engineering creates specialized amide synthetases for pharmaceutical molecules",
    "summary": "A cell-free design-build-test-learn platform generated sequence-function data from 1,217 McbA variants across 10,953 reactions. Ridge-regression ML guided enzyme variants that produced nine small-molecule pharmaceuticals with activity 1.6- to 42-fold above the parent; initial substrate-promiscuity assays found McbA could make 11 pharmaceutical compounds plus hybrid molecules. This is experimental enzyme engineering, not evidence of therapeutic efficacy of the products.",
    "field": "biology",
    "ai_technology": "Augmented ridge-regression ML models within a cell-free sequence-function mapping and enzyme-engineering workflow",
    "organization": "Northwestern University; Stanford University",
    "source_url": "https://www.nature.com/articles/s41467-024-55399-0.pdf",
    "source_type": "paper",
    "publication_date": "2025-01-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-55399-0",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A cell-free design-build-test-learn platform generated sequence-function data from 1,217 McbA variants across 10,953 reactions.",
      "unknown": "The platform optimizes in-vitro enzyme production of molecules and does not establish drug efficacy, manufacturing readiness, or broad transferability beyond tested reactions. The nine-product gains are relative to the parent enzyme under the reported assay conditions.",
      "ai_role": "Used measured variant fitness data to predict and select higher-activity McbA variants for distinct amide-forming reactions; selected variants were synthesized and assayed.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-55399-0.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-deepmet-anticipates-and-helps-identify-dozens-of-previously-unrecognized-mammalian-metabolites",
    "title": "DeepMet anticipates and helps identify dozens of previously unrecognized mammalian metabolites",
    "summary": "DeepMet, an LSTM chemical language model trained on known human metabolites, generated a ranked space of metabolite-like structures and was integrated with mass spectrometry data. Researchers report identifying several dozen previously unrecognized mammalian metabolites across standards, human biofluids, mouse tissues and repository data. Some standard-supported assignments still have isomer ambiguity; the study also found known metabolites missing or misannotated in HMDB.",
    "field": "chemistry",
    "ai_technology": "DeepMet chemical language model (three-layer LSTM over SMILES), integrated with MS/MS-spectrum prediction and meta-learning",
    "organization": "Princeton University; University of Alberta; University of British Columbia; University of Toronto; Baylor College of Medicine; collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41586-025-09969-x.pdf",
    "source_type": "paper",
    "publication_date": "2026-01-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09969-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers report identifying several dozen previously unrecognized mammalian metabolites across standards, human biofluids, mouse tissues and repository data.",
      "unknown": "A subset of assigned structures retain regioisomer/stereoisomer ambiguity; human oversight helped prioritize structures for synthesis. The model explores chemical space similar to its training set and may miss divergent biosynthetic chemistry. The paper distinguishes novel metabolites from known compounds missing in HMDB.",
      "ai_role": "Learned metabolite-like chemical structures and prioritized structures likely to exist; combined with accurate mass/MS/MS and reference standards to support identification of previously unrecognized metabolites.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-025-09969-x.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-machine-learning-guided-screening-validates-14-previously-unreported-druggpcr-interactions",
    "title": "Machine-learning-guided screening validates 14 previously unreported drug–GPCR interactions",
    "summary": "Chemical Space Neural Networks used network-neighborhood labels to predict drug–target interactions and were evaluated with a yeast biosensor screen of 3,773 compound–GPCR pairs (539 compounds and seven human GPCRs). The authors report 14 experimentally significant interactions with no prior literature/database evidence in the relevant chemical neighborhoods, tentatively linking some to adverse effects. This is an in-vitro receptor-screen discovery, not clinical proof of drug side effects.",
    "field": "chemistry",
    "ai_technology": "Chemical Space Neural Networks (CSNN), neighborhood-to-prediction model using network homophily and graph neural networks with labels as features",
    "organization": "Technical University of Denmark; Biomia ApS; Joint BioEnergy Institute; Lawrence Berkeley National Laboratory; University of California, Berkeley",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12048553/",
    "source_type": "paper",
    "publication_date": "2025-05-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-59418-6",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors report 14 experimentally significant interactions with no prior literature/database evidence in the relevant chemical neighborhoods, tentatively linking some to adverse effects.",
      "unknown": "Yeast assay is a heterologous in-vitro system and does not establish physiological effects in humans. The article describes the 14 interactions as potentially linked to adverse effects, not confirmed clinical adverse events. Some assay limitations and false positives are discussed; precision for predicted positives is limited.",
      "ai_role": "Used chemical-neighborhood labels to prioritize/annotate compound–GPCR interactions; model outputs were compared with high-throughput yeast biosensor measurements to identify previously unreported hits.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12048553/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-machine-learning-prioritizes-microbiome-metabolites-that-stimulate-interleukin-8-secretion",
    "title": "Machine learning prioritizes microbiome metabolites that stimulate interleukin-8 secretion",
    "summary": "Researchers trained ML models on public drug-development data to prioritize biological properties of microbiome metabolites, then tested predictions prospectively. In vitro experiments identified previously unreported IL-8 secretion stimulation by spermine and spermidine, metabolites previously regarded as anti-inflammatory. The paper reports average accuracy of 92.86% in prospective experiments, while cautioning that sample sizes were small.",
    "field": "biology",
    "ai_technology": "Machine-learning property prediction models implemented in KNIME and trained on public drug-development data",
    "organization": "Duke University; Duke Microbiome Center",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13059116/",
    "source_type": "paper",
    "publication_date": "2026-03-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.isci.2026.115282",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers trained ML models on public drug-development data to prioritize biological properties of microbiome metabolites, then tested predictions prospectively. The selected metabolites were tested in new in-house assays; this is direct experimental testing within the discovery study.",
      "unknown": "The unexpected activity is demonstrated in vitro and does not establish in-vivo host effects. The article explicitly cautions that prospective validation sample sizes were small; reported model metrics vary by task.",
      "ai_role": "Predicted and triaged microbiome metabolites for target assays, including IL-8 activation; prospective laboratory assays then established the unexpected cytokine-stimulatory activity of spermine and spermidine.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13059116/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-03-kcml-links-olfactory-receptor-perturbations-to-multicellular-organization-and-colorectal-cancer-signaling",
    "title": "KCML links olfactory-receptor perturbations to multicellular organization and colorectal-cancer signaling",
    "summary": "Knowledge- and Context-driven Machine Learning (KCML) analyzed perturbation phenotypes across molecular, cellular and population scales. In colorectal-cancer cells, it associated depletion of olfactory receptors and TGFβ/WNT pathway genes with abnormal multicellular organization. The authors integrated the predictions with patient data and found olfactory-receptor expression correlated with WNT/TGFβ signatures and worse outcomes in higher-grade tumors.",
    "field": "biology",
    "ai_technology": "KCML, weakly supervised machine-learning framework using support-vector-machine classifiers trained on gene-ontology annotations and high-throughput perturbation phenotypes",
    "organization": "University of Oxford; University of Zurich",
    "source_url": "https://ora.ox.ac.uk/objects/uuid%3Abbef6c79-e4e7-43a3-a819-c00f09a588d1/files/sv979v315p",
    "source_type": "paper",
    "publication_date": "2020-03-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.15252/msb.20199083",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Knowledge- and Context-driven Machine Learning (KCML) analyzed perturbation phenotypes across molecular, cellular and population scales.",
      "unknown": "Patient data validate expression associations and outcome stratification, not a causal olfactory-receptor mechanism in tumors. The authors describe KCML as hypothesis-generating; predictions can be indirect through interacting genes. This is a distinct paper/result from later AI cancer models.",
      "ai_role": "Mapped gene-perturbation morphology to context-specific gene functions and nominated olfactory-receptor links to multicellular organization; patient-expression analyses provided an independent association check.",
      "evidence_urls": [
        "https://ora.ox.ac.uk/objects/uuid%3Abbef6c79-e4e7-43a3-a819-c00f09a588d1/files/sv979v315p"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-08-mitoreid-identifies-epicatechin-as-a-cyclooxygenase-2-inhibitor",
    "title": "MitoReID identifies epicatechin as a cyclooxygenase-2 inhibitor",
    "summary": "A deep-learning system called MitoReID classified drug mechanisms from time-resolved mitochondrial morphology and membrane-potential images. Among its natural-compound predictions, it nominated epicatechin as a COX-2 inhibitor; the authors report in-vitro follow-up including COX-2 activity, cellular thermal-shift and surface-plasmon-resonance assays. This is a target-mechanism finding, not evidence of clinical efficacy.",
    "field": "chemistry",
    "ai_technology": "MitoReID, deep-learning re-identification model using an Inflated 3D ResNet backbone on temporal mitochondrial images",
    "organization": "Zhejiang University; Zhejiang Lab; Zhejiang Chinese Medical University; Harvard University; Karolinska Institute; Stanford University",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39169261/",
    "source_type": "paper",
    "publication_date": "2024-08-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43588-024-00679-4",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Among its natural-compound predictions, it nominated epicatechin as a COX-2 inhibitor; the authors report in-vitro follow-up including COX-2 activity, cellular thermal-shift and surface-plasmon-resonance assays.",
      "unknown": "Evidence is biochemical/in-vitro and does not establish therapeutic efficacy or safety. MitoReID's six unseen-drug mechanism predictions are a separate task; the epicatechin COX-2 result is the specific new mechanistic finding recorded here. Full publisher page was cookie-blocked, but the primary PubMed record and publisher-indexed article text/extended-data caption were read.",
      "ai_role": "Ranked mechanisms of action for compounds by matching learned time-dependent mitochondrial phenotypes against reference drugs; the model nominated COX-2 inhibition for epicatechin, which was tested in vitro.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39169261/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-beaut-uncovers-new-gut-microbial-bile-acid-enzymes-and-the-novel-bile-acid-3-acetodca",
    "title": "BEAUT uncovers new gut-microbial bile-acid enzymes and the novel bile acid 3-acetoDCA",
    "summary": "The BEAUT workflow combines ESM-2 protein-language-model embeddings with a neural classifier and active-site-pocket-similarity data augmentation to search gut microbial proteins for bile-acid enzymes. The authors predicted over 600,000 candidate enzymes and experimentally tested selected predictions; assays identified new bile-acid activities including MABH and ADS. ADS produces 3-acetoDCA, a previously undescribed bile-acid carbon skeleton, reported in human feces and shown to alter microbial composition and promote Lactobacillus-associated effects in experimental studies.",
    "field": "biology",
    "ai_technology": "BEAUT, ESM-2 protein-language-model embeddings plus dense neural-network classifier, augmented using active-site-pocket similarity",
    "organization": "Peking University; Peking University Third Hospital; Tsinghua–Peking University Joint Center for Life Sciences; Nanchang University",
    "source_url": "https://cqb.pku.edu.cn/info/1101/3119.htm",
    "source_type": "press release",
    "publication_date": "2025-08-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2025.07.017",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors predicted over 600,000 candidate enzymes and experimentally tested selected predictions; assays identified new bile-acid activities including MABH and ADS.",
      "unknown": "The 600,000 figure is a prediction set, not experimentally characterized enzymes; authors tested selected subsets. Physiological claims are based on study assays and animal/microbial experiments, not demonstrated human clinical effects. The journal full text was inaccessible in this reader; the Peking University research team's detailed release summarizes its Cell paper, methods and experiments.",
      "ai_role": "Screened microbial protein sequences for likely bile-acid-metabolizing function and prioritized candidate enzymes; experimental characterization of selected AI predictions identified enzyme activities and the ADS/3-acetoDCA pathway.",
      "evidence_urls": [
        "https://cqb.pku.edu.cn/info/1101/3119.htm"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-horizyn-1-experimentally-identifies-enzymes-for-orphan-and-nonnatural-reactions",
    "title": "Horizyn-1 experimentally identifies enzymes for orphan and nonnatural reactions",
    "summary": "Researchers used the reaction–protein dual-encoder model Horizyn-1 to prioritize enzyme candidates, then experimentally validated activity in orphan-reaction deorphaning, enzyme-promiscuity tests and nonnatural transformations. For an uncharacterized Thermus aquaticus protein, LC–MS confirmed dCMP kinase activity; a selected transaminase produced noncanonical amino acids from lysine-derived donors, with isotope tracing supporting predominant alpha-amine transfer.",
    "field": "biology",
    "ai_technology": "Horizyn-1",
    "organization": "Dayhoff Labs; and collaborating researchers",
    "source_url": "https://doi.org/10.1073/pnas.2520070123",
    "source_type": "paper",
    "publication_date": "2026-03-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers used the reaction–protein dual-encoder model Horizyn-1 to prioritize enzyme candidates, then experimentally validated activity in orphan-reaction deorphaning, enzyme-promiscuity tests and nonnatural transformations.",
      "unknown": "Evidence is from specific in-vitro assays and tested sequences/reactions; model retrieval performance and the tested enzyme activities do not establish broad generalization. The paper reports that the predicted CMP kinase activity for one protein was not confirmed, while dCMP activity was.",
      "ai_role": "A contrastive model jointly embedded reaction fingerprints and ProtT5 protein-sequence representations, ranked candidate enzyme–reaction pairs, and informed experimental selection. The authors assayed predictions with biochemical assays, LC–MS and isotope tracing; some fine-tuning results are predictive-performance findings rather than biological discoveries.",
      "evidence_urls": [
        "https://doi.org/10.1073/pnas.2520070123"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-cellpaint-posh-screening-links-aurkaip1-knockout-with-reduced-mitochondrial-abundance-and-mtorc1-signaling",
    "title": "CellPaint-POSH screening links AURKAIP1 knockout with reduced mitochondrial abundance and mTORC1 signaling",
    "summary": "Self-supervised CP-DINO analysis of pooled Cell Painting CRISPR profiles nominated AURKAIP1 among genes with mitochondrial and mTORC1-related phenotypes. Follow-up AURKAIP1 knockout in A549 cells reduced the mitochondrial marker TOMM20 and pS6, an mTORC1 activity readout, relative to control. The authors propose that mitochondrial translation/metabolism may influence mTORC1, while the causal pathway remains unresolved.",
    "field": "biology",
    "ai_technology": "CellPaint-POSH with CP-DINO-1640 self-supervised embeddings and a vision transformer",
    "organization": "Broad Institute; and collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41467-025-66778-6",
    "source_type": "paper",
    "publication_date": "2025-12-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-66778-6",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Self-supervised CP-DINO analysis of pooled Cell Painting CRISPR profiles nominated AURKAIP1 among genes with mitochondrial and mTORC1-related phenotypes.",
      "unknown": "The result is from an A549 cell-line knockout study and supports an association between AURKAIP1 loss, reduced mitochondrial abundance and reduced pS6; it does not establish the molecular mechanism or direct causal pathway between mitochondrial function and mTORC1. The arrayed follow-up used two technical replicate wells per knockout, was not randomized or blinded, and is not an independent replication. The paper identifies HSD17B10 as a previously reported mTORC1 regulator; this record focuses on AURKAIP1.",
      "ai_role": "CP-DINO self-supervised embeddings and morphological similarity analysis of pooled CRISPR knockout profiles identified AURKAIP1 in a cluster with mitochondrial and pS6 phenotypes, prioritizing it for follow-up. The authors then used arrayed RNP knockout and TOMM20/pS6 staining; those assays showed reduced mitochondrial abundance and pS6 after AURKAIP1 knockout. The model enabled candidate prioritization, while the perturbation and marker measurements supplied the specific experimental evidence.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-66778-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-01-alphafold-multimer-guided-experiments-corroborate-six-novel-human-protein-interfaces",
    "title": "AlphaFold-Multimer-guided experiments corroborate six novel human protein interfaces",
    "summary": "Researchers developed a fragmentation strategy for AlphaFold-Multimer and applied it to 62 human protein interactions associated with neurological disorders. Eighteen predicted structures were judged correct or likely correct, and experiments corroborated six previously unreported interfaces: FBXO23–STX1B, STX1B–VAMP2, ESRRG–PSMC5, PEX3–PEX19, PEX3–PEX16 and SNRPB–GIGYF1.",
    "field": "biology",
    "ai_technology": "AlphaFold-Multimer with a protein-fragmentation strategy",
    "organization": "Institute of Molecular Biology Mainz; Hebrew University of Jerusalem and collaborators",
    "source_url": "https://link.springer.com/article/10.1038/s44320-023-00005-6",
    "source_type": "paper",
    "publication_date": "2024-01-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44320-023-00005-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Eighteen predicted structures were judged correct or likely correct, and experiments corroborated six previously unreported interfaces: FBXO23–STX1B, STX1B–VAMP2, ESRRG–PSMC5, PEX3–PEX19, PEX3–PEX16 and SNRPB–GIGYF1.",
      "unknown": "The article reports 18 correct or likely correct models among 62 interactions and experimental support for six interfaces; these are not a comprehensive interactome map. AlphaFold-Multimer showed limited specificity and was sensitive to input-fragment length, as the source explicitly notes.",
      "ai_role": "AlphaFold-Multimer predicted domain–motif interface structures; researchers tested predictions and used the model-informed fragmentation strategy to prioritize interaction interfaces. Wet-lab assays corroborated six novel contacts, while the tool had limited specificity and performance declined with longer/full-length input sequences.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1038/s44320-023-00005-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-machine-learning-prioritizes-compounds-active-against-heterogeneous-glioblastoma-stem-cell-models",
    "title": "Machine learning prioritizes compounds active against heterogeneous glioblastoma stem-cell models",
    "summary": "Researchers trained cell-survival predictors on image-based screening data from six patient-derived glioblastoma stem-cell lines and screened more than 12,000 compounds. Experimental follow-up identified three compounds with activity across the GBM phenotype; HSP90 inhibitor XL-888 eliminated all six tested lines at nanomolar IC50 values, with analyses consistent with HSP90 inhibition and downstream client-protein disruption.",
    "field": "medicine",
    "ai_technology": "Machine-learning cell-survival predictors trained on high-throughput phenotypic screening",
    "organization": "University of Edinburgh; and Cancer Research UK Scotland Centre",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42158926/",
    "source_type": "paper",
    "publication_date": "2026-05-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/d5dd00190k",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimental follow-up identified three compounds with activity across the GBM phenotype; HSP90 inhibitor XL-888 eliminated all six tested lines at nanomolar IC50 values, with analyses consistent with HSP90 inhibition and downstream client-protein disruption.",
      "unknown": "This is an in-vitro lead-discovery study across six patient-derived cell lines, not clinical efficacy. XL-888 is an existing HSP90 inhibitor and its proposed development as a GBM lead needs medicinal-chemistry optimization, including improved blood–brain-barrier penetration; no patient outcomes are reported.",
      "ai_role": "ML models trained on in-house imaging/compound-response data predicted cell survival and prioritized compounds from a >12,000-compound screen; experimental assays across patient-derived GBM stem-cell models validated three leads and tested mechanisms. Researchers performed the wet-lab work and mechanism analyses.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42158926/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-xunzi-prioritizes-chk2-as-a-parkinsons-disease-mechanism-and-therapeutic-target",
    "title": "XunZi prioritizes CHK2 as a Parkinson’s disease mechanism and therapeutic target",
    "summary": "The XunZi AI-biologist integrated literature reasoning with multimodal molecular data and nominated CHK2 and IRAK4 kinase activation in Parkinson’s disease models. Follow-up experiments found that pharmacological or genetic Chk2 inhibition rescued dopaminergic-neuron loss and motor deficits in PD mouse models, supporting CHK2 as a disease-associated mediator in the tested systems.",
    "field": "medicine",
    "ai_technology": "XunZi AI biologist",
    "organization": "Huazhong University of Science and Technology; and Sichuan University",
    "source_url": "https://www.nature.com/articles/s41551-026-01769-6",
    "source_type": "paper",
    "publication_date": "2026-08-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41551-026-01769-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Follow-up experiments found that pharmacological or genetic Chk2 inhibition rescued dopaminergic-neuron loss and motor deficits in PD mouse models, supporting CHK2 as a disease-associated mediator in the tested systems.",
      "unknown": "The neuroprotective evidence is from experimental Parkinson’s disease models and mice, not human patients or clinical efficacy. XunZi also nominated IRAK4, but the abstract specifically reports Chk2 inhibition rescue; do not imply equivalent causal validation for every AI-nominated target. The Nature article’s complete indexed text was read; direct open may be blocked by a cookie redirect.",
      "ai_role": "XunZi integrated logical reasoning with multimodal data fusion across biomedical literature and molecular datasets to generate and rank de novo disease-target hypotheses. Researchers then tested the CHK2 hypothesis with cellular and mouse experiments, including pharmacological/genetic inhibition; the AI nominated the target, while experiments established the reported effects.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41551-026-01769-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-02-machine-learning-network-analysis-identifies-a-conserved-cell-mechanics-gene-module",
    "title": "Machine-learning network analysis identifies a conserved cell-mechanics gene module",
    "summary": "Researchers combined transcriptomic profiles with single-cell mechanical measurements from mouse and human systems and used machine-learning-based discriminative network analysis to infer a conserved five-gene module associated with cell mechanical properties. In perturbation experiments, changing CAV1 expression by silencing or overexpression altered measured cell stiffness, providing experimental support for a selected marker’s role.",
    "field": "biology",
    "ai_technology": "PC-corr machine-learning-based discriminative network analysis",
    "organization": "Technische Universität Dresden; Max Planck Institute for the Science of Light and collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11832173/",
    "source_type": "paper",
    "publication_date": "2025-02-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.7554/eLife.87930.3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers combined transcriptomic profiles with single-cell mechanical measurements from mouse and human systems and used machine-learning-based discriminative network analysis to infer a conserved five-gene module associated with cell mechanical properties.",
      "unknown": "Only CAV1 was experimentally tested as a selected target in the reported perturbation experiments; the other genes are candidate regulators. Experiments show changes in measured stiffness in specific cell models, not broad physiological or clinical effects. Paper’s eLife assessment notes further validation, especially of genes with previously unknown roles, would strengthen the network.",
      "ai_role": "PC-corr analyzed transcriptomic data linked to measured mechanical phenotypes to identify a conserved five-gene signature and prioritize candidate regulators; selected genes were computationally assessed across six validation datasets, and CAV1 was experimentally perturbed to test effects on cell mechanics.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11832173/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-machine-learning-screening-reveals-abbv-744-as-a-modulator-of-p2y-purinergic-receptors",
    "title": "Machine learning screening reveals ABBV-744 as a modulator of P2Y purinergic receptors",
    "summary": "Researchers used consensus machine-learning classifiers trained on known P2Y6 agonists to select compounds for pharmacological testing. The anticancer drug ABBV-744 showed non-surmountable inhibition of UDP-induced P2Y6 receptor activation and enhanced P2Y1 receptor activation in cellular assays, revealing previously unreported purinergic-receptor activity for this molecule.",
    "field": "chemistry",
    "ai_technology": "Consensus of deep learning, AdaBoost, naive Bayes, k-nearest neighbors, logistic regression, random forest, SVM and XGBoost classifiers",
    "organization": "National Institute of Diabetes and Digestive and Kidney Diseases; and Collaborations Pharmaceuticals",
    "source_url": "https://link.springer.com/article/10.1007/s11302-024-10003-4",
    "source_type": "paper",
    "publication_date": "2024-03-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1007/s11302-024-10003-4",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers used consensus machine-learning classifiers trained on known P2Y6 agonists to select compounds for pharmacological testing.",
      "unknown": "The paper found no novel P2Y6 agonists, which was the initial aim. ABBV-744 inhibited P2Y6 activation at relatively high micromolar concentration (reported IC50 about 75.7 µM); its precise inhibitory mechanism was not determined and requires further assays. P2Y1 enhancement and weak P2Y14 binding effects are assay-specific observations, not clinical pharmacology or therapeutic benefit.",
      "ai_role": "Models trained on curated P2Y6 agonist activity data prioritized molecules from compound libraries for experimental screening. Functional calcium-transient assays in receptor-expressing cells and fluorescent binding assays were used to test predictions; ABBV-744 was the principal confirmed hit with unanticipated activity across P2Y receptor subtypes.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1007/s11302-024-10003-4"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-knowledge-graph-ai-prioritizes-tpp1-as-a-parkinsons-disease-target-linked-to-synuclein-homeostasis",
    "title": "Knowledge-graph AI prioritizes TPP1 as a Parkinson’s disease target linked to α-synuclein homeostasis",
    "summary": "Standigm ASK’s knowledge-graph model ranked Parkinson’s disease target associations; subgraph enrichment and expression analyses narrowed 2,527 predicted targets to five candidates, including TPP1. In a human cell-based α-synuclein seeding model, TPP1 knockdown increased aggregate accumulation and insoluble α-synuclein, experimentally supporting a TPP1–α-synuclein relationship and a possible protective role.",
    "field": "biology",
    "ai_technology": "Standigm ASK knowledge-graph path-based deep-learning model with subgraph-level over-representation re-ranking",
    "organization": "Standigm; and Ajou University School of Medicine",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12814083/",
    "source_type": "paper",
    "publication_date": "2026-01-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.csbj.2025.12.035",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a human cell-based α-synuclein seeding model, TPP1 knockdown increased aggregate accumulation and insoluble α-synuclein, experimentally supporting a TPP1–α-synuclein relationship and a possible protective role.",
      "unknown": "The functional experiment is a human neuroblastoma cell model exposed to α-synuclein preformed fibrils, not a clinical or in-vivo therapeutic demonstration. Knockdown increasing aggregation supports a protective association; direct enzymatic cleavage of α-synuclein is proposed from structural modeling, not established by the reported work. TPP1 is a candidate therapeutic target, not a validated treatment.",
      "ai_role": "The model integrated heterogeneous biomedical databases, encoded relational paths using a convolutional network and attention aggregation, and produced disease-target scores. Researchers applied enrichment and transcriptomic filters to prioritize TPP1, then validated TPP1’s relationship to α-synuclein aggregation via siRNA knockdown and cellular assays; AlphaFold-Multimer modeling suggested, but did not prove, direct proteolysis.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12814083/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-machine-learning-guided-genome-mining-links-an-amycolatopsis-bgc-to-dipyrimicins",
    "title": "Machine-learning-guided genome mining links an Amycolatopsis BGC to dipyrimicins",
    "summary": "A machine-learning model that predicted antibacterial activity from biosynthetic-gene-cluster sequences prioritized Amycolatopsis azurea DSM 43854. Bioactivity-guided fractionation recovered dipyrimicins A and B from this strain, and genome analysis linked them to a previously uncharacterized dip BGC with a 76% predicted antibacterial probability; broth-microdilution assays confirmed activity against several test strains.",
    "field": "biology",
    "ai_technology": "Machine-learning predictor of antibacterial activity from biosynthetic gene clusters",
    "organization": "Vanderbilt University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13122635/",
    "source_type": "paper",
    "publication_date": "2026-03-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jnatprod.6c00057",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Bioactivity-guided fractionation recovered dipyrimicins A and B from this strain, and genome analysis linked them to a previously uncharacterized dip BGC with a 76% predicted antibacterial probability; broth-microdilution assays confirmed activity against several test strains.",
      "unknown": "Dipyrimicins themselves had been reported previously from a related Amycolatopsis strain; the paper’s new result is their isolation from A. azurea DSM 43854 and first linkage to its BGC, not discovery of the molecules as novel compounds. Reported antibacterial potency was moderate and tested against a few strains (for dipyrimicin A, MICs included 32 µg/mL for B. spizizenii and 64 µg/mL for S. aureus; >128 µg/mL for P. aeruginosa). Proposed functions of some tailoring genes remain speculative.",
      "ai_role": "The model ranked biosynthetic clusters and prioritized A. azurea DSM 43854 for experimental work. Researchers performed bioactivity-guided fractionation, isolated the compounds, linked them to the previously uncharacterized dip BGC, and tested antibacterial activity. The dip BGC was absent from the model’s training set, providing an out-of-training-set test of the prioritization approach.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13122635/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-10-machine-learning-screening-finds-previously-unreported-antibacterial-compounds-active-against-burkholderia-and-eskape-pathogens",
    "title": "Machine-learning screening finds previously unreported antibacterial compounds active against Burkholderia and ESKAPE pathogens",
    "summary": "A directed-message-passing neural network trained on a 29,537-compound Burkholderia cenocepacia screen prioritized compounds from a natural-product library. Laboratory testing found five previously uncharacterized compounds that inhibited B. cenocepacia; 22 of 43 tested prioritized compounds inhibited at least one of five ESKAPE pathogens.",
    "field": "medicine",
    "ai_technology": "Directed-message-passing neural network (D-MPNN) implemented with Chemprop",
    "organization": "University of Manitoba",
    "source_url": "https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1010613",
    "source_type": "paper",
    "publication_date": "2022-10-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pcbi.1010613",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Laboratory testing found five previously uncharacterized compounds that inhibited B. cenocepacia; 22 of 43 tested prioritized compounds inhibited at least one of five ESKAPE pathogens.",
      "unknown": "The compounds were active in vitro; no clinical effect, safety or therapeutic readiness is established. Several confirmed compounds were known antimicrobials, while five natural-product hits were described as previously uncharacterized. An inferred RpoB mechanism for STL558147 was suggested by CRISPRi sensitivity rather than directly demonstrated binding.",
      "ai_role": "The D-MPNN ranked FDA-approved and natural-product library molecules for predicted antibacterial activity; researchers tested high-ranked compounds in bacterial growth assays and confirmed activity. It was a screening-prioritization tool, not evidence of clinical efficacy.",
      "evidence_urls": [
        "https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1010613"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-metgenx-annotates-two-previously-uncharacterized-metabolites-in-mouse-liver",
    "title": "MetGenX annotates two previously uncharacterized metabolites in mouse liver",
    "summary": "MetGenX, a structure-informed neural-network workflow for generating structures from tandem mass spectra, was applied to untargeted mouse-liver metabolomics and reported two previously uncharacterized metabolites, adenosyl-cysteine and adenosine 5′-diphosphate 5′-inosine (Ap2I). The paper reports validation against chemical standards using retention time and MS2 spectral matches.",
    "field": "chemistry",
    "ai_technology": "MetGenX structure-informed encoder-decoder neural network with LightGBM reranking",
    "organization": "Shanghai Institute of Organic Chemistry, Chinese Academy of Sciences, and collaborators",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42009661/",
    "source_type": "paper",
    "publication_date": "2026-04-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-72149-6",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "MetGenX, a structure-informed neural-network workflow for generating structures from tandem mass spectra, was applied to untargeted mouse-liver metabolomics and reported two previously uncharacterized metabolites, adenosyl-cysteine and adenosine 5′-diphosphate 5′-inosine (Ap2I).",
      "unknown": "The original Nature full-text route was inaccessible in this session; the primary-source PubMed article page provides the paper abstract and figure captions describing standard-based retention-time/MS2 matches for the two structures. The finding is metabolite annotation in mouse tissue, not evidence of human biological function. Independent replication was not checked.",
      "ai_role": "The model generated and reranked candidate molecular structures from MS2 spectra; researchers applied a multi-step workflow to mouse-liver samples and compared the two reported metabolite candidates with standards. AI proposed annotations, while standard-based chromatographic and spectral matching supported the assignments.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42009661/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-07-machine-learning-reveals-a-nitrogen-assimilation-strategy-in-bacteroides",
    "title": "Machine learning reveals a nitrogen-assimilation strategy in Bacteroides",
    "summary": "A microbiome metabolomics study used random-forest models to associate bacterial metabolite profiles with taxonomy and found a previously undescribed nitrogen-assimilation pattern in Bacteroides/Bacteroidetes. Follow-up growth experiments tested 60 Bacteroidetes strains with asparagine or glutamine as the sole nitrogen source and supported the predicted phenotype; the study also reports a distinct reduced-sulfur utilization result in B. thetaiotaomicron.",
    "field": "biology",
    "ai_technology": "Random-forest models over metabolomics profiles",
    "organization": "Stanford University; and Chan Zuckerberg Biohub collaborators",
    "source_url": "https://www.nature.com/articles/s41586-021-03707-9",
    "source_type": "paper",
    "publication_date": "2021-07-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-021-03707-9",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Follow-up growth experiments tested 60 Bacteroidetes strains with asparagine or glutamine as the sole nitrogen source and supported the predicted phenotype; the study also reports a distinct reduced-sulfur utilization result in B. thetaiotaomicron.",
      "unknown": "The study's abstract reports machine-learning discovery of a new metabolism and the figure captions document follow-up culture tests; the article's full text is paywalled in this session, so the exact mechanistic pathway is not asserted. Mouse and culture evidence do not establish a human health effect.",
      "ai_role": "The models identified metabolite features that distinguished taxa and highlighted a previously undescribed Bacteroides metabolic phenotype; researchers then tested bacterial growth under defined nitrogen sources. Comparative-genomics pathways remained candidate explanations rather than proven pathways.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-021-03707-9"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-lambdazero-generates-an-experimentally-active-soluble-epoxide-hydrolase-inhibitor-scaffold",
    "title": "LAMBDAZERO generates an experimentally active soluble epoxide hydrolase inhibitor scaffold",
    "summary": "The LAMBDAZERO generative active-learning workflow searched for synthesizable soluble epoxide hydrolase (sEH/EPHX2) inhibitors and produced a quinazoline scaffold absent from the authors' comparison set of known sEH inhibitors. A synthesized analog, UM0152893, inhibited sEH in vitro with sub-micromolar potency.",
    "field": "chemistry",
    "ai_technology": "LAMBDAZERO generative active learning with deep reinforcement learning and docking",
    "organization": "Mila – Quebec AI Institute; Université de Montréal; University of Toronto; collaborators",
    "source_url": "https://arxiv.org/abs/2405.01616",
    "source_type": "paper",
    "publication_date": "2024-05-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2405.01616",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The LAMBDAZERO generative active-learning workflow searched for synthesizable soluble epoxide hydrolase (sEH/EPHX2) inhibitors and produced a quinazoline scaffold absent from the authors' comparison set of known sEH inhibitors.",
      "unknown": "This is a workshop paper/preprint rather than a clinical study. The reported lead is a biochemical inhibitor of purified sEH; no cellular, animal, safety or therapeutic outcome is established. The authors report 24/35 synthesized analogs active in vitro with potency spanning roughly 90 μM to 0.4 μM; one best lead is sub-micromolar.",
      "ai_role": "The generative model searched molecular space using docking, synthesizability and drug-likeness constraints and proposed the scaffold; researchers synthesized analogs and measured enzyme inhibition in biochemical assays.",
      "evidence_urls": [
        "https://arxiv.org/abs/2405.01616"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-09-cross-species-machine-learning-identifies-transcription-factors-that-negatively-regulate-plant-nitrogen-use-efficiency",
    "title": "Cross-species machine learning identifies transcription factors that negatively regulate plant nitrogen-use efficiency",
    "summary": "Evolutionarily informed XGBoost and GENIE3 analyses of conserved nitrogen-response expression patterns prioritized transcription factors associated with nitrogen-use efficiency in Arabidopsis and maize. Loss-of-function tests found higher NUE in Arabidopsis myb33 and tcp2 mutants under both low- and high-nitrogen conditions, and in six other tested TF mutants specifically under low nitrogen; a maize nfya3 mutant also showed higher nitrogen-use efficiency under sufficient nitrogen in field tests.",
    "field": "biology",
    "ai_technology": "Evolutionarily informed XGBoost models and GENIE3 random-forest gene-regulatory-network inference",
    "organization": "New York University; Purdue University and University of Illinois Urbana-Champaign",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8463701/",
    "source_type": "paper",
    "publication_date": "2021-09-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-021-25893-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Loss-of-function tests found higher NUE in Arabidopsis myb33 and tcp2 mutants under both low- and high-nitrogen conditions, and in six other tested TF mutants specifically under low nitrogen; a maize nfya3 mutant also showed higher nitrogen-use efficiency under sufficient nitrogen in field tests.",
      "unknown": "The machine-learning prediction scores are not themselves causal evidence; this paper adds functional evidence from specific mutants. Some tested maize candidate roles were assessed through Arabidopsis homologs, and the maize NFYA3 phenotype described here was under sufficient-N field conditions, not a general yield benefit across conditions.",
      "ai_role": "The models ranked conserved nitrogen-responsive genes and transcription factors for contribution or network connectivity to NUE; researchers selected the candidates and tested loss-of-function mutants in Arabidopsis and maize, including field-grown maize.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8463701/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-active-learning-repurposes-bms-262084-as-a-potent-tmprss2-inhibitor-that-blocks-coronavirus-entry",
    "title": "Active learning repurposes BMS-262084 as a potent TMPRSS2 inhibitor that blocks coronavirus entry",
    "summary": "A molecular-dynamics and active-learning workflow prioritized the existing compound BMS-262084 as a nanomolar inhibitor of human TMPRSS2 (IC50 1.82 nM). Cell assays reported inhibition of SARS-CoV-2 variants and other coronavirus entry into Calu-3 lung cells.",
    "field": "medicine",
    "ai_technology": "Molecular-dynamics-informed active learning with a target-specific scoring function",
    "organization": "Freie Universität Berlin; Microsoft Research AI for Science, NIH NCATS, German Primate Center and collaborators",
    "source_url": "https://www.nature.com/articles/s41467-025-62139-5.pdf",
    "source_type": "paper",
    "publication_date": "2025-07-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-62139-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Cell assays reported inhibition of SARS-CoV-2 variants and other coronavirus entry into Calu-3 lung cells.",
      "unknown": "BMS-262084 is a known small molecule repurposed here for TMPRSS2; the discovery claim is its target/antiviral activity under tested conditions. Evidence is biochemical and cell-based, not clinical efficacy or safety. Assay concentrations, viral variants and cell systems limit generalization.",
      "ai_role": "The active-learning workflow used receptor ensembles and target-specific features to iteratively prioritize compounds for testing; researchers measured purified-enzyme inhibition and viral-entry effects in cell assays. This is AI-assisted repurposing of an existing compound, not de novo molecule design.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-62139-5.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-03-machine-learning-selection-uncovers-plant-enzymes-for-alkaloid-pathway-branches",
    "title": "Machine-learning selection uncovers plant enzymes for alkaloid pathway branches",
    "summary": "Support-vector-machine ranking selected poppy homologs whose experimental characterization identified aromatic acetaldehyde synthase activity in PsTyDC6 and phenylpyruvate-decarboxylase activity in PsPDC1. In vitro and engineered E. coli metabolomics detected pathway products including norcoclaurine and reticuline, demonstrating alternative routes from tyrosine into benzylisoquinoline alkaloids.",
    "field": "biology",
    "ai_technology": "High-dimensional support vector machine (SVM) enzyme-sequence classifiers",
    "organization": "Kobe University; Tokyo Denki University, and collaborators",
    "source_url": "https://www.nature.com/articles/s41467-022-28883-8",
    "source_type": "paper",
    "publication_date": "2022-03-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-022-28883-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Support-vector-machine ranking selected poppy homologs whose experimental characterization identified aromatic acetaldehyde synthase activity in PsTyDC6 and phenylpyruvate-decarboxylase activity in PsPDC1.",
      "unknown": "The study demonstrates enzyme activity and pathway products in vitro and engineered E. coli; it does not establish the exact physiological roles of these enzymes in living poppy plants. The paper describes several enzyme activities as potential and says natural pathway roles need further elucidation.",
      "ai_role": "SVM models ranked more than 100 homologous candidate sequences for specialized aromatic-acetaldehyde-synthase and phenylpyruvate-decarboxylase functions. The team selected high-scoring sequences for in-vitro enzyme assays and metabolomics of reconstructed E. coli pathways.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-022-28883-8"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-08-machine-learning-treg-framework-identifies-smarcc1-as-a-previously-unrecognized-regulator",
    "title": "Machine-learning Treg framework identifies SMARCC1 as a previously unrecognized regulator",
    "summary": "A framework combining single-cell ATAC-seq, motif analysis, topic modeling, machine learning and genetic variation identified SMARCC1 as a previously unrecognized controller of regulatory T-cell chromatin programs. CRISPR-Cas9 deletion of Smarcc1 in Tregs produced the predicted loss of accessibility in topic-3 regulatory regions, supporting the computationally inferred connection.",
    "field": "biology",
    "ai_technology": "Integrated machine-learning framework using topic modeling and high-density genetic variation over single-cell ATAC-seq profiles",
    "organization": "Harvard Medical School; and Imagine Institute, University of Paris",
    "source_url": "https://cbdm.hms.harvard.edu/assets/Publications/2024pub/chowdharyPNAS.pdf",
    "source_type": "paper",
    "publication_date": "2024-08-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2411301121",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A framework combining single-cell ATAC-seq, motif analysis, topic modeling, machine learning and genetic variation identified SMARCC1 as a previously unrecognized controller of regulatory T-cell chromatin programs.",
      "unknown": "CRISPR validation supported the predicted SMARCC1 effect for topic 3; topic 10 was not represented and the authors note possible redundancy with SMARCC2. The result concerns regulatory chromatin programs in Tregs, not clinical or therapeutic outcomes.",
      "ai_role": "The framework organized co-varying open chromatin regions into Treg regulatory modules, inferred transcription-factor effects on those programs, and nominated SMARCC1 as a regulator of topic-3/10 accessibility. The team tested the prediction by CRISPR-Cas9 editing Tregs and profiling chromatin accessibility.",
      "evidence_urls": [
        "https://cbdm.hms.harvard.edu/assets/Publications/2024pub/chowdharyPNAS.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-machine-learning-microscopy-isolates-micronucleated-cells-and-reveals-chromosome-abnormality-formation",
    "title": "Machine-learning microscopy isolates micronucleated cells and reveals chromosome-abnormality formation",
    "summary": "The MAGIC platform used an XGBoost image classifier to identify and photolabel micronucleated cells for single-cell genome sequencing. In near-diploid human cell models, the selected cells exposed new chromosome-abnormality formation patterns, including dicentric-associated breakage-fusion-bridge events; measured baseline abnormalities approximately doubled in TP53-deficient cells, and chromosome losses outnumbered gains.",
    "field": "biology",
    "ai_technology": "MAGIC adaptive imaging platform with an XGBoost micronuclei classifier",
    "organization": "European Molecular Biology Laboratory; and German Cancer Research Center (DKFZ), with collaborators",
    "source_url": "https://www.nature.com/articles/s41586-025-09632-5",
    "source_type": "paper",
    "publication_date": "2025-10-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09632-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In near-diploid human cell models, the selected cells exposed new chromosome-abnormality formation patterns, including dicentric-associated breakage-fusion-bridge events; measured baseline abnormalities approximately doubled in TP53-deficient cells, and chromosome losses outnumbered gains.",
      "unknown": "Findings come from near-diploid non-transformed cell lines and sequencing limited mainly to chromosome abnormalities larger than 200 kb in dividing cells. Micronucleated RPE-1 cells did not show significant CA enrichment, unlike MCF10A; spontaneous cell-line results do not establish cancer outcomes in patients.",
      "ai_role": "Machine learning examined live confocal images, identified micronucleated cells and drove targeted photolabelling so selected cells could be sorted for downstream single-cell sequencing. The model enabled targeted sampling; it did not itself infer chromosome-abnormality mechanisms from sequence data.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-025-09632-5"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-01-mechanistic-machine-learning-links-pi3k-to-src-inhibitor-effects-in-cardiac-fibroblasts",
    "title": "Mechanistic machine learning links PI3K to Src-inhibitor effects in cardiac fibroblasts",
    "summary": "LogiMML, trained on high-content images of human cardiac fibroblasts, predicted that Src inhibition suppresses actin–myosin stress fibers through PI3K signaling. Follow-up PI3K inhibition reduced stress-fiber formation and procollagen I production in the fibroblasts, supporting the proposed pathway mechanism.",
    "field": "biology",
    "ai_technology": "LogiMML, a logic-based mechanistic network coupled to ridge regression",
    "organization": "University of Virginia School of Medicine; and Brigham Young University",
    "source_url": "https://www.pnas.org/doi/pdf/10.1073/pnas.2303513121",
    "source_type": "paper",
    "publication_date": "2024-01-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2303513121",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "LogiMML, trained on high-content images of human cardiac fibroblasts, predicted that Src inhibition suppresses actin–myosin stress fibers through PI3K signaling.",
      "unknown": "The validation is in vitro in primary human ventricular cardiac fibroblasts; it supports a pathway contribution to measured cell phenotypes, not therapeutic efficacy in an animal model or patients. The model's PI3K effect is described as partial.",
      "ai_role": "LogiMML fitted the relationship between high-content microscopy phenotypes and perturbations to an experimentally informed fibroblast signaling network, then predicted pathway mechanisms for pirfenidone and the Src inhibitor WH-4-023. Researchers tested the PI3K component of the Src-inhibitor prediction with PI3K inhibition in human cardiac fibroblasts.",
      "evidence_urls": [
        "https://www.pnas.org/doi/pdf/10.1073/pnas.2303513121"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-01-machine-learned-multiscale-simulations-reveal-lipid-dependent-ras-organization-at-membranes",
    "title": "Machine-learned multiscale simulations reveal lipid-dependent RAS organization at membranes",
    "summary": "The MuMMI machine-learning multiscale simulation ensemble found that local membrane lipid fingerprints correlate with wild-type KRAS clustering and orientation, with many simulated RAS–RAS interfaces appearing nonspecific. Complementary liposome and diffusion experiments supported lipid-dependent RAS adhesion and dynamics, motivating the authors' conclusion that lipids can mediate RAS multimer organization.",
    "field": "biology",
    "ai_technology": "MuMMI (Multiscale Machine-learned Modeling Infrastructure)",
    "organization": "Lawrence Livermore National Laboratory; Los Alamos National Laboratory, and the NCI RAS Initiative",
    "source_url": "https://www.pnas.org/content/pnas/119/1/e2113297119.full.pdf",
    "source_type": "paper",
    "publication_date": "2022-01-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2113297119",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Complementary liposome and diffusion experiments supported lipid-dependent RAS adhesion and dynamics, motivating the authors' conclusion that lipids can mediate RAS multimer organization.",
      "unknown": "Most specific orientation and interface conclusions are simulation-derived; experimental comparisons provide validation of selected membrane adhesion/diffusion behaviors, not direct demonstration of these mechanisms in intact cells or cancer outcomes. Authors describe absolute diffusion values as semiquantitative and dependent on methods/force fields.",
      "ai_role": "Machine learning selects diverse local membrane patches from a macroscale lipid model to launch and couple atomistic/coarse-grained simulations, producing an ensemble of more than 100,000 simulations. The resulting simulations identified lipid-composition relationships with RAS orientation and multimer organization; selected predictions were compared with surface-plasmon resonance, fluorescence-correlation and other biophysical experiments.",
      "evidence_urls": [
        "https://www.pnas.org/content/pnas/119/1/e2113297119.full.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-10-bionic-links-the-natural-product-np329-to-yeast-cell-wall-biogenesis-and-1-6-glucan-synthesis",
    "title": "BIONIC links the natural product NP329 to yeast cell-wall biogenesis and β-1,6-glucan synthesis",
    "summary": "The BIONIC graph-convolutional network integrated yeast interaction datasets and predicted essential-gene chemical sensitivities to the uncharacterized compound NP329. Experimental screens supported four BIONIC-predicted sensitive genes in the cell-wall biogenesis process; follow-up tests showed related steroidal alkaloid jervine reduced β-1,6-glucan and KRE6/SKN1 mutations conferred resistance, supporting a pathway-level connection for this compound family.",
    "field": "biology",
    "ai_technology": "BIONIC (Biological Network Integration using Convolutions), a graph-convolutional deep-learning network-integration method",
    "organization": "University of Toronto; RIKEN Center for Sustainable Resource Science, and collaborators",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11236286/",
    "source_type": "paper",
    "publication_date": "2022-10-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-022-01616-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimental screens supported four BIONIC-predicted sensitive genes in the cell-wall biogenesis process; follow-up tests showed related steroidal alkaloid jervine reduced β-1,6-glucan and KRE6/SKN1 mutations conferred resistance, supporting a pathway-level connection for this compound family.",
      "unknown": "BIONIC predicted chemical–genetic sensitivity, not direct molecular binding. NP329 is pseudojervine; jervine is a related compound tested in follow-up. The authors report jervine reduced β-1,6-glucan and KRE6/SKN1 mutations conferred resistance, implicating but not biochemically proving target engagement. Yeast-specific findings.",
      "ai_role": "BIONIC combined yeast protein-interaction, coexpression and genetic-interaction network features to predict essential-gene chemical sensitivities. For NP329 it predicted four of six experimentally sensitive essential genes in the glycosylation/cell-wall biogenesis process, helping identify this cellular process for follow-up. The later jervine experiments are supportive chemical-family follow-up, not direct proof that BIONIC predicted the jervine interaction.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11236286/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-12-a-recurrent-neural-network-uncovers-a-non-canonical-human-grhl1-dna-binding-site",
    "title": "A recurrent neural network uncovers a non-canonical human GRHL1 DNA-binding site",
    "summary": "A convolutional recurrent neural network found 46 GRHL1 ChIP-seq sites missed by a conventional position-weight matrix, including sequences lacking the previously considered obligatory CNNG motif. Isothermal titration calorimetry confirmed GRHL1 binding to a selected non-canonical sequence lacking that core motif; nucleotide mutagenesis experiments also showed that model scores tracked measured affinity across tested variants.",
    "field": "biology",
    "ai_technology": "Convolutional recurrent neural network trained on GRHL1 HT-SELEX binding data",
    "organization": "Charité – Universitätsmedizin Berlin; Max Delbrück Center for Molecular Medicine; Hannover Medical School; Institute of Bioorganic Chemistry, Polish Academy of Sciences",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10696883/",
    "source_type": "paper",
    "publication_date": "2023-12-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s12864-023-09830-3",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Isothermal titration calorimetry confirmed GRHL1 binding to a selected non-canonical sequence lacking that core motif; nucleotide mutagenesis experiments also showed that model scores tracked measured affinity across tested variants.",
      "unknown": "Only one of the 46 predicted non-canonical sites was directly validated by ITC in this study; 46 is a model-nominated set, not 46 experimentally confirmed bindings. Biochemical affinity was tested with the isolated GRHL1 DNA-binding domain and synthetic DNA, not in vivo transcriptional effects for the novel site.",
      "ai_role": "The CNN/RNN learned GRHL1 sequence-binding patterns from HT-SELEX and scored human-cell ChIP-seq peaks. It identified 46 sites not detected by FIMO/PWM; one selected high-scoring non-canonical site was experimentally confirmed by ITC, and predicted variant affinity changes were compared with ITC results.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10696883/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-a-protein-language-model-workflow-discovers-high-activity-kynureninase-enzymes-for-tumor-models",
    "title": "A protein-language-model workflow discovers high-activity kynureninase enzymes for tumor models",
    "summary": "A workflow combining homology search, protein-language-model embeddings and regression prioritized kynureninases by predicted catalytic activity. Researchers experimentally tested four top-ranked candidates; the most active had nearly twice the turnover number of the prior best, and enzyme treatment was associated with a 3.4-fold reduction in tumor weight in a mouse comparison.",
    "field": "medicine",
    "ai_technology": "Protein language model embeddings coupled with regression in the SeekRank enzyme-prioritization workflow",
    "organization": "Seoul National University; and Galux Inc.",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39777462/",
    "source_type": "paper",
    "publication_date": "2025-01-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/nar/gkae1245",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers experimentally tested four top-ranked candidates; the most active had nearly twice the turnover number of the prior best, and enzyme treatment was associated with a 3.4-fold reduction in tumor weight in a mouse comparison.",
      "unknown": "The evidence is a published paper abstract; it describes an enzyme/treatment experiment in a mouse tumor model, not clinical efficacy. The reported tumor-weight reduction is model-specific; no claim of human therapeutic benefit is made.",
      "ai_role": "The model represented homologous kynureninase sequences with protein-language-model embeddings and regressed kinetic activity to rank enzyme candidates. Four high-ranked candidates were tested in vitro and in vivo; the abstract reports one with nearly two-fold improved turnover number and reduced tumor weight in a mouse model.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39777462/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-machine-learning-uncovers-four-new-extracellular-contractile-injection-system-toxins",
    "title": "Machine learning uncovers four new extracellular contractile injection-system toxins",
    "summary": "An XGBoost classifier trained on genomic and biochemical features prioritized eCIS-associated toxins from bacterial and archaeal genomes. Four candidates, EAT14–17, were experimentally shown to be toxic in bacterial and/or yeast cells; EAT14 also inhibited Rac1/Cdc42-mediated mitogenic signaling in human cells.",
    "field": "biology",
    "ai_technology": "XGBoost classifier for extracellular contractile injection-system-associated toxins (EATs)",
    "organization": "Hebrew University of Jerusalem; University of Illinois Urbana-Champaign; collaborators",
    "source_url": "https://link.springer.com/article/10.1038/s44320-024-00053-6",
    "source_type": "paper",
    "publication_date": "2024-07-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44320-024-00053-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Four candidates, EAT14–17, were experimentally shown to be toxic in bacterial and/or yeast cells; EAT14 also inhibited Rac1/Cdc42-mediated mitogenic signaling in human cells.",
      "unknown": "Four of eight selected candidates were toxic under the tested heterologous-expression assays; the authors report an empirical positive predictive value of 50% for this sample and discuss possible host/compartment mismatch. Toxin delivery by native eCIS and physiological target organisms remain incompletely established. EAT14’s signaling effect was shown in human cells, not as a therapeutic effect.",
      "ai_role": "The classifier used genomic and protein biochemical features to score candidate EAT genes across eCIS-positive microbial genomes. The model narrowed candidates for functional tests; heterologous expression assays validated four previously uncharacterized toxins, with follow-up active-site mutations and a human-cell signaling assay.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1038/s44320-024-00053-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-deep-learning-assisted-dip-ms-identifies-polr2e-as-a-constitutive-prefoldin-like-complex-subunit",
    "title": "Deep-learning-assisted DIP-MS identifies POLR2E as a constitutive prefoldin-like-complex subunit",
    "summary": "DIP-MS combines deep-learning-based signal processing with native-gel fractionation and mass spectrometry to resolve complex isoforms. In human prefoldin-like-complex experiments, repeated coelution and affinity-purification data supported POLR2E as a constitutive PFDL component, updating the view that it was merely an associated protein.",
    "field": "biology",
    "ai_technology": "Deep Interactome Profiling by Mass Spectrometry (DIP-MS), including deep-learning-based signal processing",
    "organization": "Institute of Molecular Systems Biology; ETH Zurich and collaborators",
    "source_url": "https://www.nature.com/articles/s41592-024-02211-y",
    "source_type": "paper",
    "publication_date": "2024-03-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-024-02211-y",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In human prefoldin-like-complex experiments, repeated coelution and affinity-purification data supported POLR2E as a constitutive PFDL component, updating the view that it was merely an associated protein.",
      "unknown": "The paper proposes POLR2E as a constitutive PFDL subunit based on coelution and affinity-purification in the tested human-cell system; it does not establish that every cellular PFDL complex contains POLR2E or fully characterize its functional contribution. DIP-MS also recapitulated known ASDURF association, which is not counted here as novel because ASDURF’s PAQosome/PFDL role had been reported earlier.",
      "ai_role": "Deep-learning signal processing helped resolve coexisting protein-complex isoforms from AP-MS/native PAGE/DIA-MS data. DIP-MS on UXT and PFDN2 baits repeatedly detected PFDL coelution with POLR2E; affinity purification of four tagged PFDL-like subunits also recovered POLR2E, supporting the proposed constitutive association.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41592-024-02211-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-04-machine-learning-screening-yields-novel-potent-acetylcholinesterase-and-butyrylcholinesterase-inhibitors",
    "title": "Machine-learning screening yields novel potent acetylcholinesterase and butyrylcholinesterase inhibitors",
    "summary": "NCATS researchers trained machine-learning QSAR models and applied them to a roughly 360,000-compound library. Experimental colorimetric and orthogonal fluorescence assays confirmed new AChE/BChE inhibitors; the study reports 22 AChE and 67 BChE inhibitors with IC50 below 5 μM, including novel structural series, with some submicromolar compounds.",
    "field": "chemistry",
    "ai_technology": "Ensemble comparison of machine-learning QSAR classifiers (including XGBoost, random forest, SVM, neural network and naïve Bayes)",
    "organization": "National Center for Advancing Translational Sciences, National Institutes of Health",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10688023/",
    "source_type": "paper",
    "publication_date": "2023-04-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jcim.3c00230",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimental colorimetric and orthogonal fluorescence assays confirmed new AChE/BChE inhibitors; the study reports 22 AChE and 67 BChE inhibitors with IC50 below 5 μM, including novel structural series, with some submicromolar compounds.",
      "unknown": "These are biochemical inhibitor assays and early medicinal-chemistry leads; they do not establish brain exposure, in-vivo efficacy, safety or clinical benefit. Several known compounds were newly identified as inhibitors, so the count of all assay hits should not be described as entirely novel molecules.",
      "ai_role": "Models trained on quantitative high-throughput screening data prioritized compounds in a ~360,000-compound collection for AChE/BChE inhibition and selectivity assays. Researchers tested 2,762 high-scoring compounds in colorimetric assays and used an orthogonal fluorescence assay to confirm predicted inhibitors; subsequent structure–activity analysis identified novel scaffolds.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10688023/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-enzysfc-discovers-nitrile-hydratase-activity-in-nitrilases-and-evolves-amide-selective-variants",
    "title": "EnzySFC discovers nitrile-hydratase activity in nitrilases and evolves amide-selective variants",
    "summary": "The AI-assisted EnzySFC workflow experimentally validated 10 previously uncharacterized nitrilases selected from 1,113 candidates. A wild-type Phototrophicales nitrilase showed exclusive nitrile hydratase activity; a separate Actinomycetia nitrilase was AI-evolved, with four tested top variants reaching 100% amide yield and complete substrate conversion.",
    "field": "biology",
    "ai_technology": "EnzySFC AI-assisted enzyme discovery and functional-evolution platform",
    "organization": "China State Institute of Pharmaceutical Industry, Research Center for Systems Biosynthesis",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41645784/",
    "source_type": "paper",
    "publication_date": "2026-02-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jafc.5c14148",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The AI-assisted EnzySFC workflow experimentally validated 10 previously uncharacterized nitrilases selected from 1,113 candidates.",
      "unknown": "The reported outcomes are enzyme and substrate assays; the abstract does not establish industrial scale performance or application in a living system. The 10 discovery candidates and 16 mutants were experimental follow-ups to model prioritization, not comprehensive validation of all 1,113 sequences.",
      "ai_role": "EnzySFC combined de novo enzyme discovery and multiobjective activity/specificity optimization to rank candidate nitrilases and propose evolved mutants. The study experimentally tested ten discovery candidates and sixteen mutants from one evolution prediction cycle; the named activity switch and product-selective variants were measured in assays.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41645784/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-machine-learning-metabolomics-identifies-widespread-promiscuous-activities-in-escherichia-coli-enzymes",
    "title": "Machine-learning metabolomics identifies widespread promiscuous activities in Escherichia coli enzymes",
    "summary": "In vitro metabolomics of 667 purified E. coli enzymes, analyzed with a machine-learning-derived multivariate classifier, indicated previously unrecognized promiscuous activities in 11% of enzymes tested. Follow-up reconstruction assigned 22 reaction stoichiometries to 11 enzymes, with four reactions experimentally validated; examples include expanded nucleotide-substrate use by DeoA, cofactor and sugar promiscuity in NanK, and FMN dephosphorylation by CobC. Follow-up assays also found that pyrimidine-specific ribonucleoside hydrolase RihB acted on thymidine, guanosine and deoxyuridine, further extending its measured substrate range.",
    "field": "biology",
    "ai_technology": "Machine-learning-derived multivariate classifier applied to enzyme metabolomics ion-trace data",
    "organization": "ETH Zurich; Weizmann Institute of Science; University Children’s Hospital Zurich and University of Zurich",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13389410/",
    "source_type": "paper",
    "publication_date": "2026-05-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42003-026-10099-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In vitro metabolomics of 667 purified E. coli enzymes, analyzed with a machine-learning-derived multivariate classifier, indicated previously unrecognized promiscuous activities in 11% of enzymes tested.",
      "unknown": "The classifier operated at a 33% false-discovery rate; the 135 ion traces were putatively annotated, and only four of 22 reconstructed reaction stoichiometries were experimentally validated. Therefore the full 11% result is a set of candidate activities, not 11% independently confirmed enzyme functions. The RihB activities were prompted by a low-confidence potassium-thymidine-adduct trace and then tested directly with purified enzyme; they should be distinguished from the classifier’s higher-confidence hit set.",
      "ai_role": "A machine-learning-derived multivariate classifier, with a reported false-discovery rate of 33%, detected unexpected changes in metabolite ion traces in the enzyme assays and helped nominate candidate promiscuous activities. The researchers then reconstructed candidate reaction stoichiometries and experimentally validated four reactions.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13389410/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-11-generative-deep-learning-discovers-the-ripk1-inhibitor-ri-962-with-an-unreported-scaffold",
    "title": "Generative deep learning discovers the RIPK1 inhibitor RI-962 with an unreported scaffold",
    "summary": "A target-conditioned generative deep-learning model based on a distribution-learning conditional recurrent neural network generated a tailored compound library that was screened for RIPK1 inhibition. The study reports RI-962 as a potent inhibitor with a previously unreported scaffold, validated in vitro for protection from necroptosis and in vivo in two inflammatory models.",
    "field": "medicine",
    "ai_technology": "Generative deep-learning distribution-learning conditional recurrent neural network (GDL cRNN)",
    "organization": "State Key Laboratory of Biotherapy and Cancer Center, West China Hospital, Sichuan University; West China School of Pharmacy, Sichuan University",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/36371441/",
    "source_type": "paper",
    "publication_date": "2022-11-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-022-34692-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reports RI-962 as a potent inhibitor with a previously unreported scaffold, validated in vitro for protection from necroptosis and in vivo in two inflammatory models.",
      "unknown": "Evidence is preclinical: cellular necroptosis assays and two inflammatory mouse models do not establish human efficacy or safety. The paper identifies one leading molecule from a generated library; it does not validate generative-model performance across targets generally.",
      "ai_role": "The GDL cRNN generated a target-tailored virtual compound library; virtual screening and subsequent chemical synthesis and bioactivity assays selected RI-962. This provides direct experimental testing of a model-generated molecule rather than a prediction-only hit.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/36371441/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-02-machine-learning-guided-screening-identifies-bcl-2-family-inhibitors-is20-and-is21",
    "title": "Machine-learning-guided screening identifies Bcl-2-family inhibitors IS20 and IS21",
    "summary": "Machine-learning-guided virtual screening followed by surface-plasmon-resonance and biological assays identified IS20 and IS21 as compounds binding anti-apoptotic Bcl-2, Bcl-xL and Mcl-1 proteins. Both affected cancer-cell viability and induced apoptosis/autophagic flux in tested cell models; IS21 also reduced tumor growth in leukemia and melanoma mouse models.",
    "field": "medicine",
    "ai_technology": "Machine-learning-guided virtual screening (model family not named in PubMed abstract)",
    "organization": "IRCCS Regina Elena National Cancer Institute; Sapienza University of Rome; Italian National Research Council and collaborators",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/35265218/",
    "source_type": "paper",
    "publication_date": "2022-02-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.7150/thno.64233",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Machine-learning-guided virtual screening followed by surface-plasmon-resonance and biological assays identified IS20 and IS21 as compounds binding anti-apoptotic Bcl-2, Bcl-xL and Mcl-1 proteins.",
      "unknown": "The article calls the compounds potential/new leads; findings are biochemical, cell-based and mouse-model preclinical evidence. It does not establish clinical effectiveness. The PubMed abstract does not specify the machine-learning algorithm, so no specific model architecture is attributed here.",
      "ai_role": "Machine-learning-guided virtual screening prioritized putative Bcl-2-family inhibitor compounds; the candidate hits were then evaluated by surface plasmon resonance and in vitro cell assays, and IS21 was evaluated in mouse tumor models. The reported outcome is the specific assay characterization of IS20 and IS21.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/35265218/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-deep-learning-virtual-screening-finds-antibacterial-hits-in-over-one-billion-compounds",
    "title": "Deep-learning virtual screening finds antibacterial hits in over one billion compounds",
    "summary": "Genentech and collaborators trained the deep-learning model GNEprop on high-throughput screening data and screened more than 1.4 billion synthetically accessible molecules. In follow-up experiments, 82 selected compounds showed antibacterial activity against the sensitized E. coli ΔtolC strain; the paper reports a 90-fold higher hit rate than the training screen and describes chemically diverse hits, some with activity beyond that strain and validated biological targets.",
    "field": "medicine",
    "ai_technology": "GNEprop deep-learning antibacterial activity predictor",
    "organization": "Genentech Research and Early Development; NVIDIA; Mila–Université de Montréal collaborators",
    "source_url": "https://www.nature.com/articles/s41587-025-02814-6",
    "source_type": "paper",
    "publication_date": "2025-10-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-025-02814-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In follow-up experiments, 82 selected compounds showed antibacterial activity against the sensitized E. coli ΔtolC strain; the paper reports a 90-fold higher hit rate than the training screen and describes chemically diverse hits, some with activity beyond that strain and validated biological targets.",
      "unknown": "The 82 confirmed hits were tested for single-dose activity against the sensitized E. coli ΔtolC screening strain; this is not evidence that all 82 work against wild-type bacteria or in infections. The 90-fold improvement refers to hit rate relative to the HTS training experiment. Some additional compounds were selected for dose-response testing; do not imply all virtual hits were validated. A publisher correction fixed a coauthor’s misspelled first name and did not change results.",
      "ai_role": "GNEprop was trained on approximately two million small molecules screened against E. coli ΔtolC and used to rank a much larger synthetically accessible library for experimental prioritization. The follow-up assays confirmed antibacterial activity for 82 compounds; biological characterization also investigated specific targets.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41587-025-02814-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-10-prospective-machine-learning-screen-identifies-six-previously-unreported-oatp1b1-inhibitors",
    "title": "Prospective machine-learning screen identifies six previously unreported OATP1B1 inhibitors",
    "summary": "Researchers trained and compared seven machine-learning classifiers for OATP1B1 transporter inhibition. In prospective CHO-cell testing, 15 of 19 compounds predicted active by SVC models inhibited the transporter; six (abamectin, asiaticoside, berbamine, doramectin, mobocertinib and umbralisib) were described as previously unreported OATP1B1 inhibitors.",
    "field": "chemistry",
    "ai_technology": "Seven molecular-fingerprint machine-learning classifiers, with prospective selection by support vector classifier (SVC)",
    "organization": "Collaborations Pharmaceuticals; University of Arizona",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9873312/",
    "source_type": "paper",
    "publication_date": "2022-10-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.molpharmaceut.2c00662",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In prospective CHO-cell testing, 15 of 19 compounds predicted active by SVC models inhibited the transporter; six (abamectin, asiaticoside, berbamine, doramectin, mobocertinib and umbralisib) were described as previously unreported OATP1B1 inhibitors. The new candidate screen used contracted Eurofins assays; external execution is not a separate replication study.",
      "unknown": "The 84% refers to a small prospective test set of 19 compounds, and transporter inhibition was measured in vitro in engineered CHO cells. These findings flag possible drug–drug interaction liability; they do not show clinical interaction or patient outcome. The authors qualify six inhibitors as appearing novel at publication.",
      "ai_role": "Models trained on in-vitro OATP1B1 inhibition data ranked compounds for prospective testing. Researchers tested 19 model-selected compounds in transporter-expressing CHO cells and observed inhibition for 15, including six compounds the authors described as novel OATP1B1 inhibitors.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9873312/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-machine-learning-screening-identifies-natural-rort-inhibitors-with-preclinical-psoriasis-activity",
    "title": "Machine-learning screening identifies natural RORγt inhibitors with preclinical psoriasis activity",
    "summary": "An ensemble machine-learning screen of natural products prioritized protoberberine alkaloids as candidate RORγt inhibitors. In vitro assays found that berberine and coptisine inhibited Th17 differentiation and directly bound RORγt by surface plasmon resonance; coptisine also showed efficacy in a preclinical mouse psoriasis model.",
    "field": "medicine",
    "ai_technology": "Ensemble machine-learning models for natural-product RORγt activity and counter-screening, combining graph neural networks and molecular-fingerprint random forests",
    "organization": "Bionsight; Incheon National University; Seoul National University; Kangwon National University; Ajou University; Wright State University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13227160/",
    "source_type": "paper",
    "publication_date": "2025-09-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.jare.2025.09.004",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In vitro assays found that berberine and coptisine inhibited Th17 differentiation and directly bound RORγt by surface plasmon resonance; coptisine also showed efficacy in a preclinical mouse psoriasis model.",
      "unknown": "Berberine and coptisine are known natural products; the novelty is their prioritization and experimental characterization as RORγt inhibitors in this workflow, not discovery of new chemical entities. The mouse result is preclinical and does not establish psoriasis treatment in humans. The catalog already contains an unrelated RORγt result involving topotecan; this paper's compounds and evidence are distinct.",
      "ai_role": "Ensemble models screened a natural-product library and prioritized predicted RORγt inhibitors. Chemotaxonomic analysis narrowed candidates to protoberberines; six were experimentally evaluated, and berberine/coptisine showed in-vitro effects and direct target binding, with coptisine tested in a mouse psoriasis model.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13227160/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-machine-learning-reveals-broad-and-previously-uncharacterized-fungal-polyketide-synthase-activity",
    "title": "Machine learning reveals broad and previously uncharacterized fungal polyketide-synthase activity",
    "summary": "A genome-mining and cell-free enzyme study characterized fungal type III polyketide synthases (T3PKSs), revealing broad and varied substrate/product profiles. An ML model trained on 341 enzyme–substrate pairs predicted substrate specificity and guided tests that found two fungal enzymes, AcaePKS and PansPKS, active on all ten tested substrates, including non-natural substrates absent from training data.",
    "field": "biology",
    "ai_technology": "Feedforward neural-network multilayer perceptron trained on protein-language-model embeddings and substrate fingerprints",
    "organization": "University of Groningen; University of Vienna",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12559476/",
    "source_type": "paper",
    "publication_date": "2025-09-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/anie.202514786",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An ML model trained on 341 enzyme–substrate pairs predicted substrate specificity and guided tests that found two fungal enzymes, AcaePKS and PansPKS, active on all ten tested substrates, including non-natural substrates absent from training data.",
      "unknown": "The enzyme products were characterized in vitro; the study examines product scope and biocatalytic potential, not native ecological roles or medical efficacy. Two candidate proteins expressed poorly, and the model's transfer accuracy was lower on non-fungal enzyme classes. The reported five enzyme sample includes only three well-expressing candidates, of which two showed the broadest tested activity.",
      "ai_role": "The model learned fungal T3PKS enzyme–substrate activity from measured data, was tested on unseen substrates, and predicted candidate promiscuous fungal enzymes. Guided experiments confirmed broad substrate activity for AcaePKS and PansPKS, including substrate pairs outside training data; cell-free assays and LC–MS characterized enzyme products.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12559476/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-caclens-predicts-and-validates-a-zearalenone-degrading-enzyme",
    "title": "CACLENS predicts and validates a zearalenone-degrading enzyme",
    "summary": "The multitask deep-learning enzyme-screening framework CACLENS selected ten candidate enzymes for zearalenone (ZEN) degradation, which were expressed and tested. One candidate degraded more than 90% of both ZEN and its analogue α-zearalenol (α-ZOL) in the reported assay.",
    "field": "biology",
    "ai_technology": "CACLENS (Cross-Attention & Contrastive Learning-enabled Enzyme Selection), a multimodal multitask deep-learning enzyme classifier",
    "organization": "Shanghai Institute of Nutrition and Health, Chinese Academy of Sciences; University of Chinese Academy of Sciences",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12904076/",
    "source_type": "paper",
    "publication_date": "2025-12-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202518063",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The multitask deep-learning enzyme-screening framework CACLENS selected ten candidate enzymes for zearalenone (ZEN) degradation, which were expressed and tested.",
      "unknown": "This is an in-vitro enzyme degradation finding; it does not establish safety of degradation products, remediation performance at environmental scale, or food/feed safety. The abstract does not name the successful enzyme or its degradation-product identity.",
      "ai_role": "CACLENS combined reaction-type classification, EC-number prediction and reaction-feasibility estimates to screen candidate enzymes. The paper then expressed and assayed ten selected enzyme candidates against ZEN; one showed >90% degradation of ZEN and α-ZOL.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12904076/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-active-learning-discovers-eight-experimental-binders-for-the-lrrk2-wdr-domain",
    "title": "Active learning discovers eight experimental binders for the LRRK2 WDR domain",
    "summary": "A free-energy molecular-dynamics and machine-learning active-learning workflow optimized compounds against the understudied LRRK2 WDR domain. Of 35 analogs experimentally tested, eight showed measurable binding confirmed by surface plasmon resonance and, for fluorinated compounds, an orthogonal 19F-NMR assay; measured dissociation constants ranged from 18 to 230 μM.",
    "field": "chemistry",
    "ai_technology": "Active-learning-guided relative-binding-free-energy (AL-RBFE) workflow using molecular-dynamics thermodynamic-integration calculations and an ML affinity-prediction model",
    "organization": "Carnegie Mellon University; University of Toronto Structural Genomics Consortium; University of British Columbia; University of Ottawa and collaborators",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12152950/",
    "source_type": "paper",
    "publication_date": "2025-05-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jcim.5c00588",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Of 35 analogs experimentally tested, eight showed measurable binding confirmed by surface plasmon resonance and, for fluorinated compounds, an orthogonal 19F-NMR assay; measured dissociation constants ranged from 18 to 230 μM.",
      "unknown": "The experiments establish ligand binding, not functional inhibition or a therapeutic effect. Confirmed Kd values (18–230 μM) indicate modest affinity, and the work is an optimization round around two previously confirmed challenge hits rather than an initial target hit discovery. This is not evidence of Parkinson’s disease efficacy.",
      "ai_role": "An iterative ML model selected ligand analogs for expensive free-energy simulations, using resulting calculations to choose compounds for further rounds and experimental validation. Of 35 procured/tested compounds, eight were confirmed as binding to LRRK2 WDR by SPR, with 19F-NMR used for orthogonal confirmation of fluorinated molecules.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12152950/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-02-biological-activity-profile-modeling-identifies-sars-cov-2-antiviral-leads",
    "title": "Biological activity-profile modeling identifies SARS-CoV-2 antiviral leads",
    "summary": "NCATS researchers used biological activity-based modeling (BABM) to infer antiviral activity from compounds’ profiles across multiple assays. Of 311 model-predicted SARS-CoV-2 candidates tested in a live-virus cell-culture assay, 99 were confirmed active (32% hit rate); follow-up assays implicated viral-entry inhibition and/or autophagy modulation among confirmed compounds.",
    "field": "medicine",
    "ai_technology": "Biological activity-based modeling (BABM) models trained on multi-assay compound-activity profiles",
    "organization": "National Center for Advancing Translational Sciences, NIH; Public Health Agency of Canada; Florida State University",
    "source_url": "https://www.nature.com/articles/s41587-021-00839-1",
    "source_type": "paper",
    "publication_date": "2021-02-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-021-00839-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "NCATS researchers used biological activity-based modeling (BABM) to infer antiviral activity from compounds’ profiles across multiple assays.",
      "unknown": "Evidence is from cell-culture live-virus and mechanistic assays, not clinical treatment. 99/311 were active in the primary assay; further potency/secondary confirmations were available for subsets. Most confirmed compounds were reported to inhibit entry and/or modulate autophagy; these data do not establish safe or effective human therapy.",
      "ai_role": "BABM used prior activity signatures across assay panels to predict compounds with potential SARS-CoV-2 activity, including compounds outside conventional similar-structure hit finding. Researchers tested 311 model-predicted compounds in live-virus assays and conducted secondary/mechanistic assays on confirmed actives.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41587-021-00839-1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-mispop-machine-learning-screening-identifies-the-oncolytic-peptide-dermaseptin-s9",
    "title": "MISPOP machine learning screening identifies the oncolytic peptide Dermaseptin-S9",
    "summary": "The MISPOP ensemble screened a natural peptide library of 1,033 sequences, prioritized 16 oncolytic-peptide candidates and selected five for synthesis and testing. Dermaseptin-S9 showed membrane disruption and immunogenic-cell-death readouts in vitro and over 92% tumor-growth inhibition in a B16F10 melanoma mouse model.",
    "field": "medicine",
    "ai_technology": "MISPOP mechanism-informed ensemble combining XGBoost, deep neural networks and transfer learning with physicochemical priors",
    "organization": "Shanghai University of Traditional Chinese Medicine; Sichuan University; Naval Medical University; Chinese Academy of Medical Sciences and Peking Union Medical College",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13336043/",
    "source_type": "paper",
    "publication_date": "2026-05-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.75652",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The MISPOP ensemble screened a natural peptide library of 1,033 sequences, prioritized 16 oncolytic-peptide candidates and selected five for synthesis and testing.",
      "unknown": "The active peptide is a natural sequence rather than a new AI-designed molecule. Findings are preclinical: three tumor cell lines and a B16F10 mouse melanoma model (n=5 per group shown in article figures); they do not establish human efficacy or absence of toxicity beyond the tested model/doses. Only five of 16 model-prioritized candidates were synthesized and evaluated.",
      "ai_role": "The machine-learning ensemble integrated three predictors and mechanism-related physicochemical features to prioritize candidate oncolytic peptides from a natural sequence library. Five shortlisted peptides were synthesized and evaluated; Dermaseptin-S9 was tested in cell-based membrane-disruption/ICD assays and a mouse melanoma model.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13336043/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-06-machine-intelligence-design-yields-selective-synthetic-cox-1-inhibitors-from-a-marine-natural-product-template",
    "title": "Machine-intelligence design yields selective synthetic COX-1 inhibitors from a marine natural-product template",
    "summary": "Using Marinopyrrole A as a template, an automated design workflow generated and synthesized new natural-product-inspired molecules; target-prediction software pointed to cyclooxygenases. In biochemical assays, synthetic mimetics 2 and 2a inhibited COX-1 at 0.10 and 0.16 μM, respectively, and compound 2a was crystallographically observed bound to COX-1. Human platelet assays also showed inhibition.",
    "field": "chemistry",
    "ai_technology": "DOGS rule-based de novo molecule construction and CATS molecular-similarity ranking, coupled with SPiDER self-organizing-map machine-learning target prediction",
    "organization": "ETH Zurich; Thomas Jefferson University; University of Bari; University of Innsbruck; Goethe University Frankfurt; Friedrich Schiller University Jena",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8373093/",
    "source_type": "paper",
    "publication_date": "2021-06-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202100832",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In biochemical assays, synthetic mimetics 2 and 2a inhibited COX-1 at 0.10 and 0.16 μM, respectively, and compound 2a was crystallographically observed bound to COX-1.",
      "unknown": "The study tests a small number of designed molecules; automated structure generation was rule-based and similarity-driven rather than a trained generative neural model. The COX-1 effects were biochemical/platelet assays, not clinical effects; the platelet IC50s for 2, 2a, 2b were 9, 13 and 56 nM, but COX-2 inhibition occurred at higher concentrations and selectivity varies by assay. Do not call these approved drugs or imply therapeutic benefit.",
      "ai_role": "DOGS algorithmically generated natural-product-inspired structures and suggested synthetic routes; CATS ranked candidates by pharmacophore similarity. SPiDER's cascaded self-organizing-map models predicted molecular target families, including COX. Experimental target assays confirmed the previously unrecognized COX-1 activity of selected synthetic compounds, with crystallographic confirmation for compound 2a.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8373093/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-pinal-designs-functional-proteins-from-natural-language-descriptions",
    "title": "Pinal designs functional proteins from natural-language descriptions",
    "summary": "The 16-billion-parameter Pinal model generated protein candidates from plain-language function descriptions. In four reported design tasks—fluorescent protein, PET hydrolase, alcohol dehydrogenase and metabolic H-protein—the selected designs displayed the intended function in laboratory tests; the best H-protein raised product titre 1.7-fold relative to native E. coli H-protein in a reconstituted CO2-fixation pathway.",
    "field": "biology",
    "ai_technology": "Pinal, a 16-billion-parameter protein foundation model with text-to-structure (T2struct) and text-conditioned sequence design (SaProt-T) components",
    "organization": "Westlake University; and collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41586-026-11205-z",
    "source_type": "paper",
    "publication_date": "2026-10-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-11205-z",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In four reported design tasks—fluorescent protein, PET hydrolase, alcohol dehydrogenase and metabolic H-protein—the selected designs displayed the intended function in laboratory tests; the best H-protein raised product titre 1.7-fold relative to native E. coli H-protein in a reconstituted CO2-fixation pathway.",
      "unknown": "This is a primary-paper proof of concept with selected designs, not a broad hit-rate or industrial-performance demonstration. Experimental denominators vary substantially by task; extended data reports one active PETase among 94 screened and a fluorescent design selected from 29. The H-protein titre comparison was in a reconstituted pathway, not an organism or commercial process. No independent replication or therapeutic result is claimed.",
      "ai_role": "Pinal maps natural-language function descriptions to candidate protein structures and sequences. The team filtered and ranked model outputs, selected candidates for synthesis, and tested intended functions. The paper reports lab-observed fluorescence, PET-hydrolysis/esterase activity, alcohol-dehydrogenase catalysis and H-protein activity; the latter improved titre in a reconstituted multi-enzyme pathway.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-026-11205-z"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-06-sequence-conditioned-cogmol-designs-sars-cov-2-spike-and-main-protease-inhibitors",
    "title": "Sequence-conditioned CogMol designs SARS-CoV-2 spike and main-protease inhibitors",
    "summary": "IBM Research and academic collaborators used the CogMol deep generative foundation model to design compounds from target protein sequence information alone for SARS-CoV-2 spike receptor-binding domain and main protease. In vitro inhibition was reported for two of four synthesized candidates for each target; the most potent spike-RBD inhibitor also showed activity against multiple variants in live-virus neutralization assays.",
    "field": "chemistry",
    "ai_technology": "CogMol, a sequence-conditioned deep generative foundation model using a molecular variational autoencoder and protein–ligand binding predictor",
    "organization": "IBM Research; Diamond Light Source; University of Oxford; Enamine Ltd.; associated research centers",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10284550/",
    "source_type": "paper",
    "publication_date": "2023-06-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.adg7865",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In vitro inhibition was reported for two of four synthesized candidates for each target; the most potent spike-RBD inhibitor also showed activity against multiple variants in live-virus neutralization assays.",
      "unknown": "The source describes a proof of concept with a small selected synthesis set from roughly 100 ranked designs. Only in-vitro target assays were performed; the leading RBD inhibitor was also tested in live-virus neutralization against variants. The paper says ADME attributes were not experimentally validated and further work is needed to confirm inhibitor binding mechanisms. These are preclinical leads, not medicines or demonstrated clinical benefit.",
      "ai_role": "The team sampled CogMol conditioned on each target's amino-acid sequence to generate candidate small molecules; candidates were filtered and ranked in silico, synthesized and assayed. The model generated compounds without target-specific fine-tuning or target-structure input at inference; assays found micromolar inhibition for 2/4 synthesized candidates for each target.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10284550/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-prospective-deep-learning-screen-identifies-nanomolar-irak1-inhibitor-scaffolds",
    "title": "Prospective deep-learning screen identifies nanomolar IRAK1 inhibitor scaffolds",
    "summary": "Ro5's prospective HydraScreen structure-based deep-learning screen ranked compounds for IRAK1 and was evaluated against a 47,000-compound robotic high-throughput screen. The study reports that 23.8% of IRAK1 hits were in HydraScreen's top 1% and identifies three nanomolar scaffolds, two described in the abstract as novel IRAK1 candidates.",
    "field": "chemistry",
    "ai_technology": "HydraScreen structure-based deep-learning virtual screening, with SpectraView knowledge-graph target evaluation",
    "organization": "Ro5 (AI Chemistry); and Strateos",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11566907/",
    "source_type": "paper",
    "publication_date": "2024-11-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s13321-024-00914-0",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reports that 23.8% of IRAK1 hits were in HydraScreen's top 1% and identifies three nanomolar scaffolds, two described in the abstract as novel IRAK1 candidates. The new candidate experiments were conducted within the Ro5/Strateos discovery collaboration.",
      "unknown": "This is prospective early-stage hit identification, not drug development or disease efficacy. The assay validated kinase inhibition in vitro. Authors are affiliated with the companies that developed the tools. The abstract describes two scaffolds as novel candidates, while detailed similarity analysis indicates one of three nanomolar scaffolds had a similar known active; do not call all hits unprecedented.",
      "ai_role": "SpectraView selected IRAK1 as the study target; HydraScreen scored docked ligand-pose ensembles to rank compounds before prospective screening. Researchers compared rankings with automated high-throughput assay results from a 47,000-compound library and reported nanomolar IRAK1 hit scaffolds, two described as novel candidates.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11566907/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-machine-learning-prioritization-and-crispr-knockout-identify-oswox13-as-a-rice-heat-tolerance-regulator",
    "title": "Machine-learning prioritization and CRISPR knockout identify OsWOX13 as a rice heat-tolerance regulator",
    "summary": "Meta-transcriptomic comparison of heat-tolerant and susceptible rice genotypes yielded 409 meta-differentially expressed genes. Random-forest and XGBoost feature ranking with SHAP independently prioritized OsWOX13; CRISPR/Cas9 knockout lines showed lower survival under heat stress (~20% versus ~60% in wild type) and delayed flowering, with accompanying changes in ABA, oxidative-stress markers and stress/flowering-associated proteins.",
    "field": "biology",
    "ai_technology": "Random forest and XGBoost models with SHAP feature attribution",
    "organization": "Hainan University; Guizhou University; National Institute of Agricultural Science, Rural Development Administration (Republic of Korea); Pusan National University; Kookmin University",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41424172/",
    "source_type": "paper",
    "publication_date": "2025-12-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1111/ppl.70714",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Meta-transcriptomic comparison of heat-tolerant and susceptible rice genotypes yielded 409 meta-differentially expressed genes.",
      "unknown": "The PubMed primary abstract supports gene prioritization and knockout evidence, but does not establish detailed direct molecular targets or field-level yield effects. Reported survival figures concern the study's heat-stress condition and experimental lines; this is not evidence of breeding performance or resilience in field environments. Mechanistic pathway interpretations remain limited to associated physiological/proteomic measurements.",
      "ai_role": "The models prioritized stress-associated genes from meta-transcriptomic differences; OsWOX13 was among the top-ranked genes in both models. The team tested its predicted positive role using CRISPR/Cas9 knockout, followed by heat-stress survival, flowering, physiology and proteomic assays.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41424172/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-machine-learning-uncovers-new-small-molecule-chloride-transporters-from-drugbank",
    "title": "Machine learning uncovers new small-molecule chloride transporters from DrugBank",
    "summary": "After high-throughput testing 1,894 compounds and confirming 59 chloride transporters, researchers trained and evaluated machine-learning classifiers, then prioritized DrugBank compounds for wet-lab tests. Five of ten selected drugs—bazedoxifene, centhaquine, brilliant green, tamoxifen and nevanimibe—were classified by the authors as active chloride transporters, with an estimated 50–60% hit rate under their assays.",
    "field": "chemistry",
    "ai_technology": "Machine-learning binary classifiers and consensus QSAR models trained on experimentally measured chloride-transport activity",
    "organization": "Tulane University; Rowan University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12368578/",
    "source_type": "paper",
    "publication_date": "2025-08-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/d5dd00140d",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "After high-throughput testing 1,894 compounds and confirming 59 chloride transporters, researchers trained and evaluated machine-learning classifiers, then prioritized DrugBank compounds for wet-lab tests.",
      "unknown": "The paper combines a large primary high-throughput screen with a later ML-led DrugBank virtual screen; do not attribute all 59 initial hits to ML. The five drug-repurposing hits were measured in artificial liposome assays, not cells, animals or patients. The paper discusses assay artifacts: quinacrine/brilliant green interfere with the fluorescent assay, HCl salts can affect electrode readings, and evidence for brilliant green was weaker than for the other confirmed compounds. Authors report 5/10 active among selected candidates (or 7/12 including two known active compounds), not a generalizable clinical hit rate.",
      "ai_role": "After initial experimental discovery data were collected, machine-learning models ranked DrugBank candidates for transport activity. Researchers synthesized no new candidates for the virtual screen; instead, they tested ten prioritized available compounds in liposome chloride-transport assays. Five were classed as active; results extend these compounds' known pharmacology with membrane chloride transport activity.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12368578/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-chooser-uses-protein-foundation-models-to-discover-functional-cas-homologs",
    "title": "CHOOSER uses protein foundation models to discover functional Casλ homologs",
    "summary": "The CHOOSER framework fine-tuned ESM-2 protein-language-model representations for alignment-free CRISPR-Cas discovery and pre-crRNA self-processing predictions. It nominated 11 Casλ homologs, nearly doubling the known Casλ set. Wet-lab work on selected homologs confirmed pre-crRNA processing for five of six tested proteins; EphcCasλ also cleaved target DNA and showed trans-cleavage activity, with a TTR PAM reported.",
    "field": "biology",
    "ai_technology": "CHOOSER, a protein foundation-model framework fine-tuning ESM-2 representations for Cas homolog identification and Cas12 pre-crRNA-processing classification",
    "organization": "Tongji University; Zhejiang Lab; Shanghai East Hospital; collaborating Chinese research institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11576732/",
    "source_type": "paper",
    "publication_date": "2024-11-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-54365-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Wet-lab work on selected homologs confirmed pre-crRNA processing for five of six tested proteins; EphcCasλ also cleaved target DNA and showed trans-cleavage activity, with a TTR PAM reported.",
      "unknown": "This is a specific functional genomics result, not evidence of a ready-to-use diagnostic or gene-editing tool. Only selected candidates were purified and tested; the paper reports pre-crRNA activity in five of six tested Casλ proteins and DNA-cleavage validation for a subset. EphcCasλ activity and PAM characterization are in-vitro findings. The article has a publisher-listed author correction dated 2025-01-09; this record relies on the version-of-record text and retains limited scope.",
      "ai_role": "CHOOSER mined candidate proteins adjacent to CRISPR arrays using learned sequence representations rather than alignment-only detection and prioritized Cas12 homologs by predicted pre-crRNA processing. The study reported 11 Casλ homologs and experimentally tested selected predictions; purified EphcCasλ was validated for self-processing pre-crRNA, DNA cleavage and trans-cleavage.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11576732/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-deep-learning-designs-selective-antimicrobial-peptides-against-cutibacterium-acnes",
    "title": "Deep learning designs selective antimicrobial peptides against Cutibacterium acnes",
    "summary": "A deep-learning design pipeline combining sequence generation and classifiers, transfer learning, pretrained protein embeddings and C. acnes-specific training data produced 42 novel peptides for testing. Five showed high-potency, selective activity against C. acnes, with MICs of 2–4 µg/mL; additional in-vitro assays tested activity against other microbes, hemolysis and keratinocyte cytotoxicity.",
    "field": "medicine",
    "ai_technology": "Deep-learning peptide generators and activity/selectivity classifiers using transfer learning and pretrained protein embeddings",
    "organization": "Shanghai MetaNovas Biotech; Fuzhou University; Metanovas Biotech",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10894229/",
    "source_type": "paper",
    "publication_date": "2024-02-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41598-024-55205-3",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A deep-learning design pipeline combining sequence generation and classifiers, transfer learning, pretrained protein embeddings and C. acnes-specific training data produced 42 novel peptides for testing.",
      "unknown": "All evidence is in vitro, with no animal or human testing. The 42 synthesized candidates are a selected subset of model output, and the abstract's five high-potency peptides have measured C. acnes MICs; activity against other organisms and toxicity is based on separate in-vitro assays. Do not imply topical efficacy or clinical safety.",
      "ai_role": "The models generated peptide sequences, ranked antimicrobial/hemolytic properties and incorporated C. acnes-focused training data. The team synthesized and tested 42 selected designs; five had reported C. acnes MICs of 2–4 µg/mL.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10894229/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-01-amplify-prioritizes-bullfrog-genome-antimicrobial-peptides-active-against-priority-pathogens",
    "title": "AMPlify prioritizes bullfrog-genome antimicrobial peptides active against priority pathogens",
    "summary": "Researchers integrated the attentive deep-learning classifier AMPlify into a genome-mining pipeline for North American bullfrog sequences. Of 101 candidate mature peptides generated from genome-annotated precursors, AMPlify predicted 75 as antimicrobial; 11 candidates were synthesized and tested, with four showing activity against tested bacteria, including a multidrug-resistant carbapenemase-producing E. coli isolate.",
    "field": "medicine",
    "ai_technology": "AMPlify, an attentive deep-learning antimicrobial-peptide sequence classifier",
    "organization": "Canada's Michael Smith Genome Sciences Centre, BC Cancer; University of British Columbia; University of Victoria; British Columbia Centre for Disease Control; Vancouver General Hospital; University of Washington",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8788131/",
    "source_type": "paper",
    "publication_date": "2022-01-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s12864-022-08310-4",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Of 101 candidate mature peptides generated from genome-annotated precursors, AMPlify predicted 75 as antimicrobial; 11 candidates were synthesized and tested, with four showing activity against tested bacteria, including a multidrug-resistant carbapenemase-producing E. coli isolate.",
      "unknown": "This is a small selected validation set: 11 of 75 model-predicted candidates were tested and four were active. Results are in vitro and do not establish therapeutic efficacy or safety. Candidate precursor identification also used homology/genome annotation; attribute AMPlify specifically to mature-peptide prioritization, not the whole discovery workflow.",
      "ai_role": "After bioinformatics identified candidate peptide precursors from the bullfrog genome, AMPlify scored 101 mature peptide sequences and nominated putative AMPs. The team selected 11 predicted peptides for synthesis and antimicrobial susceptibility testing; four showed activity against one or more of six tested bacterial strains, including the carbapenemase-producing E. coli clinical isolate.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8788131/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-03-lstm-designed-peptides-show-antibacterial-activity-and-mixed-efficacy-in-a-mouse-sepsis-model",
    "title": "LSTM-designed peptides show antibacterial activity and mixed efficacy in a mouse sepsis model",
    "summary": "A generative long short-term-memory recurrent neural network produced 198 candidate antimicrobial peptides; five were synthesized. PEP-38 and PEP-137 inhibited carbapenem-resistant Klebsiella isolates in vitro, and PEP-137 also inhibited Pseudomonas aeruginosa. In a small mouse K. pneumoniae sepsis experiment, PEP-137 yielded 50% survival, PEP-38 did not improve survival, and the in-vitro-inactive PEP-36 control unexpectedly yielded 66.7% survival.",
    "field": "medicine",
    "ai_technology": "Generative long short-term memory recurrent neural network trained on antimicrobial peptide sequences",
    "organization": "Stavropol State Medical University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8944797/",
    "source_type": "paper",
    "publication_date": "2022-03-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3390/antibiotics11030411",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a small mouse K. pneumoniae sepsis experiment, PEP-137 yielded 50% survival, PEP-38 did not improve survival, and the in-vitro-inactive PEP-36 control unexpectedly yielded 66.7% survival.",
      "unknown": "Only 5/198 AI-generated peptides were synthesized. The animal experiment used one administration, small groups (12 mice per group) and a simple model; PEP-36's unexpected survival effect despite no in-vitro antibacterial activity complicates interpretation. PEP-38 did not improve survival; no mechanism, pharmacokinetics, pharmacodynamics or safety was established. These are exploratory preclinical results, not a treatment claim.",
      "ai_role": "The LSTM RNN generated 198 peptide sequences, which were further filtered in silico; five selected candidates were synthesized and assayed. Two showed in-vitro activity against carbapenem-resistant isolates, and two were advanced into a mouse sepsis study.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8944797/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-10-machine-learning-guided-screening-discovers-potent-innate-immune-signaling-modulators",
    "title": "Machine-learning-guided screening discovers potent innate immune signaling modulators",
    "summary": "At the University of Chicago, an active-learning pipeline iteratively selected molecules from a 139,998-compound library for in-vitro assays of NF-κB and interferon regulatory factor (IRF) signaling. After screening about 2% of the library, the study reported compounds suppressing NF-κB up to 15-fold or elevating NF-κB and IRF up to 5-fold and 6-fold, respectively. Follow-up cytokine assays found PME-4007 increased IFN-β secretion about threefold when co-delivered with the STING agonist cGAMP.",
    "field": "biology",
    "ai_technology": "Machine-learning active-learning pipeline with variational-autoencoder molecular representations, Gaussian-process regression and Bayesian optimization",
    "organization": "University of Chicago Pritzker School of Molecular Engineering; University of Chicago Cellular Screening Center",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38020385/",
    "source_type": "paper",
    "publication_date": "2023-10-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/d3sc03613h",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "At the University of Chicago, an active-learning pipeline iteratively selected molecules from a 139,998-compound library for in-vitro assays of NF-κB and interferon regulatory factor (IRF) signaling.",
      "unknown": "The work reports in-vitro pathway modulation and primary-cell cytokine assays, not vaccination, immunotherapy efficacy, animal outcomes or clinical benefit. Fold changes are study-specific relative to agonist-only controls. A subset of candidates was further characterized; four tested compounds were flagged as non-viable by the cytostasis/cytotoxicity confluency mask despite strong pathway effects.",
      "ai_role": "Models were iteratively trained on screening measurements and used Bayesian optimization to select subsequent compounds for in-vitro testing. This model-guided selection led to the reported NF-κB/IRF modulators; follow-up cytokine release assays tested a subset.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38020385/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-11-neural-network-guided-autonomous-experiments-reveal-an-unreported-tosmic-reaction",
    "title": "Neural-network-guided autonomous experiments reveal an unreported TosMIC reaction",
    "summary": "A reactivity-seeking neural network coupled to an autonomous reaction platform navigated unknown chemical mixtures using real-time analytical data. Among the reactions it identified, follow-up characterization revealed an unreported photochemical reaction and a new reactivity mode for p-toluenesulfonylmethyl isocyanide (TosMIC): six equivalents of TosMIC underwent a multistep single-substrate cascade yielding a trimeric product in 47% unoptimized yield and forming five C–C bonds.",
    "field": "chemistry",
    "ai_technology": "Reactivity-seeking neural network and junction-tree neural network integrated with an autonomous robotic synthesis and analytics platform",
    "organization": "University of Glasgow School of Chemistry",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/34849401/",
    "source_type": "paper",
    "publication_date": "2021-11-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acscentsci.1c00435",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Among the reactions it identified, follow-up characterization revealed an unreported photochemical reaction and a new reactivity mode for p-toluenesulfonylmethyl isocyanide (TosMIC): six equivalents of TosMIC underwent a multistep single-substrate cascade yielding a trimeric product in 47% unoptimized yield and forming five C–C bonds.",
      "unknown": "The source describes several found reactions, some previously known; this entry is limited to the distinct unreported TosMIC reactivity result. The reported 47% yield was unoptimized. The platform identified candidate reactions, while follow-up analyses and manual chemical investigation established the product and mechanism; no broad claim of fully autonomous mechanistic proof is made.",
      "ai_role": "A convolutional neural network classified reaction signatures from online NMR/IR/analytical data and a junction-tree neural network encoded reagent structures to help prioritize new experiments. The closed-loop system surfaced reactions; chemists then manually investigated and characterized selected outcomes, including the TosMIC trimer-forming cascade.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/34849401/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-machine-learning-guided-optimization-finds-low-loading-conditions-for-platinum-catalyzed-amide-reduction",
    "title": "Machine-learning-guided optimization finds low-loading conditions for platinum-catalyzed amide reduction",
    "summary": "A machine-learning-assisted reaction-optimization study found conditions for reducing N,N-dimethylacetamide to dimethylethylamine using the known platinum complex [Pt(DMS)₂Cl₂]. The selected condition used 0.01 mol% catalyst and reached complete conversion in two hours at 39 °C without solvent, with a reported turnover frequency of 5,002 h⁻¹. The chemical catalyst identity is not claimed as new.",
    "field": "chemistry",
    "ai_technology": "Sunthetics machine-learning reaction-optimization platform",
    "organization": "Ghent University Center for Sustainable Chemistry; Sunthetics; University of Padua",
    "source_url": "https://pubs.rsc.org/en/content/articlehtml/2024/cc/d4cc05273k",
    "source_type": "paper",
    "publication_date": "2024-11-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/d4cc05273k",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The selected condition used 0.01 mol% catalyst and reached complete conversion in two hours at 39 °C without solvent, with a reported turnover frequency of 5,002 h⁻¹.",
      "unknown": "This is a single-substrate reaction-condition optimization on N,N-dimethylacetamide. [Pt(DMS)₂Cl₂] is a known complex; the reported finding is the ML-assisted optimized condition and measured performance, not discovery of a new platinum compound. The best-performing condition is reported after human-guided interpretation as well as model recommendations. The paper does not establish transferability to other amides, industrial scale-up or general performance across catalyst classes.",
      "ai_role": "The ML platform recommended reaction conditions across iterative experiments, using observed reaction outcomes to refine proposed conditions and reduce the number of trials. The team combined those recommendations with chemist interpretation and tested the predicted conditions experimentally; the source reports the optimized platinum-catalyzed reduction as a study-specific result.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlehtml/2024/cc/d4cc05273k"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-machine-learning-and-expert-co-screening-identifies-two-new-silver-catalyzed-cycloadditions",
    "title": "Machine-learning and expert co-screening identifies two new silver-catalyzed cycloadditions",
    "summary": "Researchers at the Zelinsky Institute used unsupervised machine learning to filter and cluster candidate atom-economic cycloaddition reactions generated from quantum-chemical data, then used a digital co-expert and chemist review to prioritize experiments. The team experimentally confirmed two previously undescribed silver-catalyzed cycloaddition reactions yielding new products within about one week.",
    "field": "chemistry",
    "ai_technology": "Machine-learning candidate filtering and clustering combined with the digital co-expert reaction-prioritization workflow",
    "organization": "N. D. Zelinsky Institute of Organic Chemistry, Russian Academy of Sciences",
    "source_url": "https://onlinelibrary.wiley.com/doi/10.1002/ange.202523905",
    "source_type": "paper",
    "publication_date": "2026-02-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/anie.202523905",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers at the Zelinsky Institute used unsupervised machine learning to filter and cluster candidate atom-economic cycloaddition reactions generated from quantum-chemical data, then used a digital co-expert and chemist review to prioritize experiments.",
      "unknown": "Two reactions were confirmed from a heavily filtered set of computational candidates, with final selection and validation by chemists. The reported approximately 180-fold acceleration is a comparison for the expert-screening bottleneck, not a general chemical-discovery speedup. The abstract does not specify individual products or reaction yields; no such details are inferred here.",
      "ai_role": "Unsupervised ML filtered and clustered computationally generated reaction candidates to reduce the set for review; the digital co-expert helped prioritize candidates. Human chemists made the final selection and performed experimental validation. The source reports two newly confirmed cycloadditions; the system was a human-AI pipeline, not autonomous discovery.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/10.1002/ange.202523905"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-bayesian-optimization-finds-mild-solvent-free-mechanochemical-conditions-for-cyclopropanation-and-epoxidation",
    "title": "Bayesian optimization finds mild solvent-free mechanochemical conditions for cyclopropanation and epoxidation",
    "summary": "A Bayesian-optimization-guided mechanochemical study identified conditions for the established Johnson–Corey–Chaykovsky reaction using KOH in place of hazardous NaH, under air-equilibrated conditions and with no solvent or liquid-assisted grinding agent. In the chalcone benchmark, one optimized condition reached cyclopropanation completion in five minutes; the study also demonstrated cyclopropanation and epoxidation across multiple substrates, including yields of 45–75% for four additional substrates.",
    "field": "chemistry",
    "ai_technology": "Bayesian optimization implemented with BoTorch/PyTorch, including multi-objective optimization of product yield and process mass intensity",
    "organization": "University of Parma SynCat Lab; University of Parma IMPLab",
    "source_url": "https://www.nature.com/articles/s41467-026-75499-3",
    "source_type": "paper",
    "publication_date": "2026-07-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-75499-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Bayesian-optimization-guided mechanochemical study identified conditions for the established Johnson–Corey–Chaykovsky reaction using KOH in place of hazardous NaH, under air-equilibrated conditions and with no solvent or liquid-assisted grinding agent.",
      "unknown": "The Johnson–Corey–Chaykovsky transformation itself is established chemistry; the finding is a measured, safer and faster mechanochemical process window rather than discovery of a new reaction class. A 5-minute completion time and other yields apply to the tested substrates/conditions; some prior solvent-minimized conditions could not be reproduced. The scale and substrate scope remain those reported in the paper.",
      "ai_role": "The BO models selected among six reaction variables (base identity/loading, time, milling frequency, liquid-assisted grinding agent and loading) using experimental yields and process-mass-intensity outcomes. Researchers ran the experiments, performed synthetic substrate-scope tests and used time-resolved in-situ X-ray diffraction to study milling effects.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-75499-3"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-10-deep-learning-screening-identifies-covalent-sars-cov-2-main-protease-inhibitor-scaffolds",
    "title": "Deep-learning screening identifies covalent SARS-CoV-2 main-protease inhibitor scaffolds",
    "summary": "A two-stage deep-learning screen selected cysteine-directed compounds for testing against SARS-CoV-2 3CL protease. Of 32 top-ranked compounds experimentally assayed, 12 inhibited the enzyme; six had IC₅₀ values below 12 μM, with the strongest reported at 1.4 μM. Five of those six exhibited time-dependent inhibition consistent with covalent binding.",
    "field": "chemistry",
    "ai_technology": "Deep-learning COVCL and NOVCL classifiers integrating a directed message-passing neural network and feed-forward network",
    "organization": "Peking University BNLMS; and Peking-Tsinghua Center for Life Sciences; Chinese Academy of Medical Sciences Research Unit of Drug Design Method",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/36209629/",
    "source_type": "paper",
    "publication_date": "2022-10-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.ejmech.2022.114803",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A two-stage deep-learning screen selected cysteine-directed compounds for testing against SARS-CoV-2 3CL protease.",
      "unknown": "Evidence in the primary abstract is biochemical enzyme inhibition and inhibition-mechanism assays. It does not establish antiviral efficacy in cells/animals or clinical benefit. Potency values are for the reported enzyme assay; most tested compounds were inactive, and the six sub-12 μM hits are early leads.",
      "ai_role": "The COVCL model first prioritized cysteine-targeted covalent candidates, then NOVCL ranked those candidates for predicted 3CL-protease inhibition. Researchers tested 32 selected molecules in enzyme assays and performed follow-up tests of time-dependent inhibition for active hits.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/36209629/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-ai-and-developmental-profiling-reveal-conserved-rules-for-microrna-strand-selection",
    "title": "AI and developmental profiling reveal conserved rules for microRNA strand selection",
    "summary": "Using a high-throughput qPCR assay, Arizona State University researchers measured microRNA 5p/3p strand use across six Caenorhabditis elegans developmental stages and selected tissues. A machine-learning model trained on experimentally measured strand usage identified compositional and structural sequence features associated with strand preference; the learned relationships predicted strand usage in other vertebrates, including humans, and the authors report that the rules are conserved but context-dependent.",
    "field": "biology",
    "ai_technology": "Predictive machine-learning model using 77 biologically informed features, trained on high-throughput qPCR measurements from the HiTmiSS platform",
    "organization": "Arizona State University Biodesign Institute and School of Life Sciences; Arizona State University School of Computing and Augmented Intelligence",
    "source_url": "https://academic.oup.com/nar/article/54/2/gkaf1510/8425383",
    "source_type": "paper",
    "publication_date": "2026-01-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/nar/gkaf1510",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using a high-throughput qPCR assay, Arizona State University researchers measured microRNA 5p/3p strand use across six Caenorhabditis elegans developmental stages and selected tissues.",
      "unknown": "The dataset measured strand usage in C. elegans across six developmental stages and specific tissues; cross-species including human results are model predictions/validation against available strand-usage data, not direct tests of gene-regulatory outcomes in human cells. The result concerns strand-selection patterns and predictors, not disease causality or therapeutic efficacy.",
      "ai_role": "The model learned strand-preference relationships from the experimental strand-usage dataset, quantified feature contributions, and predicted preference across additional nematode microRNAs and vertebrate species. Experimental qPCR measurements established the C. elegans profiles; predictions across species extend the analysis but are not themselves experimental confirmation of function in each organism.",
      "evidence_urls": [
        "https://academic.oup.com/nar/article/54/2/gkaf1510/8425383"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-biochemical-constrained-generative-ai-yields-experimentally-active-antimicrobial-peptides-against-e-coli",
    "title": "Biochemical-constrained generative AI yields experimentally active antimicrobial peptides against E. coli",
    "summary": "A University of Hong Kong-led team used a conditional variational autoencoder constrained by biochemical properties and a random-forest activity classifier to generate and prioritize antimicrobial peptides. In vitro testing of 54 generated peptides found 18.5% with MIC ≤10 μg/mL and 38.9% with MIC ≤50 μg/mL against the study’s E. coli model; nine peptides were classified as both active and non-toxic in the reported assays.",
    "field": "medicine",
    "ai_technology": "CVAE-BIO: conditional variational autoencoder with biochemical constraints and a random-forest antimicrobial-activity classifier",
    "organization": "The University of Hong Kong; University of Hong Kong-Shenzhen Hospital; Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41849223/",
    "source_type": "paper",
    "publication_date": "2026-03-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/bib/bbag115",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In vitro testing of 54 generated peptides found 18.5% with MIC ≤10 μg/mL and 38.9% with MIC ≤50 μg/mL against the study’s E. coli model; nine peptides were classified as both active and non-toxic in the reported assays.",
      "unknown": "Validation is in vitro, primarily against E. coli ATCC 25922; it is not evidence of treatment efficacy in animals or humans. The reported nine active/non-toxic peptides are assay-defined candidate hits, not validated drugs. This paper generated a different peptide set and used different training/design and testing workflows from the earlier C. acnes deep-learning, bullfrog-genome AMPlify, and LSTM Klebsiella sepsis studies in this ledger; do not treat the general AMP topic as a duplicate.",
      "ai_role": "The CVAE generated peptide sequences subject to predefined biochemical constraints; the random forest ranked activity using 30 biochemical descriptors. Researchers synthesized/tested 54 generated peptides in vitro, measured MIC and hemolysis, and reported the active/non-toxic subset.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41849223/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-01-machine-learning-prioritizes-ruthenium-complexes-with-increased-mrsa-antibacterial-hit-rates",
    "title": "Machine learning prioritizes ruthenium complexes with increased MRSA antibacterial hit rates",
    "summary": "Researchers trained machine-learning models on 288 modularly synthesized ruthenium arene Schiff-base complexes and used them to prioritize 54 new compounds for antibacterial-activity testing. The model-selected set had a 53.7% hit rate against methicillin-resistant Staphylococcus aureus, compared with 9.4% in the original library, an approximately 5.7-fold enrichment.",
    "field": "chemistry",
    "ai_technology": "Machine-learning activity-prediction models trained on ruthenium arene Schiff-base complex structures and antibacterial assay data",
    "organization": "University of Bern Department of Chemistry, Biochemistry & Pharmaceutical Sciences; National University of Singapore Department of Chemistry",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38088924/",
    "source_type": "paper",
    "publication_date": "2024-01-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/anie.202317901",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers trained machine-learning models on 288 modularly synthesized ruthenium arene Schiff-base complexes and used them to prioritize 54 new compounds for antibacterial-activity testing.",
      "unknown": "This is a study-specific in-vitro antibacterial hit-rate enrichment, not animal efficacy, clinical validation, or an approved antibiotic. The comparison is 54 model-selected compounds vs the 288-complex source library, not a randomized prospective comparison with all possible metal complexes.",
      "ai_role": "Models were trained using assay data from 288 modularly synthesized ruthenium complexes and used to predict activity for a new set of 54 compounds. Those prioritized compounds were then assessed for antibacterial activity; the paper reports higher MRSA hit-rate in the model-selected set than in the original library.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38088924/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-12-machine-learning-discovers-d-amino-acid-mixtures-that-boost-anti-biofilm-antibiotic-cocktails",
    "title": "Machine learning discovers D-amino-acid mixtures that boost anti-biofilm antibiotic cocktails",
    "summary": "A machine-learning-guided high-throughput screen identified a five-component D-amino-acid mixture (“D-mix”) with about 78% anti-biofilm activity against Pseudomonas aeruginosa at 100 μM. Combining D-mix with half the lowest effective gentamicin dose retained about 91% bacterial-growth inhibition and about 93% anti-biofilm efficacy in vitro. In a small mouse lung-infection model, gentamicin/D-mix treatment reduced measured P. aeruginosa burden by about 95% in blood and 91% in lung tissue at 72 hours.",
    "field": "medicine",
    "ai_technology": "High-throughput screening coupled to machine-learning regression and iterative global optimization of D-amino-acid mixture ratios",
    "organization": "University of Science and Technology Beijing Institute for Advanced Materials and Technology; National Materials Corrosion and Protection Data Center; Beijing University of Chemical Technology",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38126652/",
    "source_type": "paper",
    "publication_date": "2023-12-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202307173",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a small mouse lung-infection model, gentamicin/D-mix treatment reduced measured P. aeruginosa burden by about 95% in blood and 91% in lung tissue at 72 hours.",
      "unknown": "The in-vivo experiment used a small mouse P. aeruginosa infection model, not human treatment; study-specific bacterial counts and efficacy are reported at 72 hours. The composition is a selected mixture of five D-amino acids, with authors noting the full synergism mechanism remains unclear. D-mix alone was not bactericidal under the reported in-vitro conditions; effects are combination- and assay-dependent.",
      "ai_role": "Experimental measurements of mixture biofilm effects were used to train and iteratively update ML regression models, which guided mixture selection over design rounds. The selected D-mix was then evaluated with different antibiotics in vitro; the lead gentamicin/D-mix combination was tested in a P. aeruginosa lung-infection mouse model.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38126652/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-machine-learning-assisted-screening-yields-an-experimentally-active-ftsz-inhibitor-candidate",
    "title": "Machine-learning-assisted screening yields an experimentally active FtsZ inhibitor candidate",
    "summary": "A fused machine-learning and virtual-screening workflow narrowed compounds from three libraries to five candidate FtsZ inhibitors for bioactivity evaluation. The compound T3995 inhibited Staphylococcus aureus ATCC 6538 and Bacillus subtilis ATCC 9732 in reported assays, with MIC values of 32 and 2 μg/mL, respectively. Morphology and FtsZ-polymerization assays supported FtsZ-related antibacterial activity.",
    "field": "medicine",
    "ai_technology": "ROC-guided machine-learning selection combined with virtual screening, clustering and binding-model analysis",
    "organization": "Jiangxi Provincial Key Laboratory of Drug Design and Evaluation, Jiangxi Science and Technology Normal University; Jiangxi University of Traditional Chinese Medicine Cancer Research Center",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39043879/",
    "source_type": "paper",
    "publication_date": "2024-07-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1007/s00210-024-03276-4",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The compound T3995 inhibited Staphylococcus aureus ATCC 6538 and Bacillus subtilis ATCC 9732 in reported assays, with MIC values of 32 and 2 μg/mL, respectively.",
      "unknown": "Evidence is early in-vitro antibacterial activity and indirect support for an FtsZ-related mechanism; the abstract does not report direct target-binding measurements or in-vivo efficacy. Hemolysis was reported as safe at or below 16 μg/mL, while the MIC against S. aureus was 32 μg/mL, so this does not establish a therapeutic safety margin. The paper’s 2025 issue date follows its 2024 online publication date; the source-publication date here is the documented first online publication (23 July 2024).",
      "ai_role": "ROC-guided ML filtered candidates from three libraries; virtual screening of acquired compounds followed, then clustering and binding-model analysis prioritized five hits. Researchers performed growth, time-kill, cell-morphology, FtsZ-polymerization and hemolysis assays, including testing T3995.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39043879/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-ai-guided-huj3-mutation-improves-hiv-1-gp120-binding-and-neutralization-in-assays",
    "title": "AI-guided HuJ3 mutation improves HIV-1 gp120 binding and neutralization in assays",
    "summary": "The IsAb2.0 workflow, combining AlphaFold-Multimer complex modeling with FlexddG mutation scoring, proposed five substitutions to improve the binding of humanized J3 nanobody HuJ3 to HIV-1 gp120. Of these, E44R experimentally improved binding (ELISA EC50 1.696 nM) and neutralization relative to HuJ3; the other proposed substitutions did not consistently improve binding, and T99R failed to express. The gain did not restore neutralization to the parental llama J3 level.",
    "field": "biology",
    "ai_technology": "IsAb2.0 antibody-design protocol integrating AlphaFold-Multimer 2.3/3.0, SnugDock, alanine scanning and FlexddG",
    "organization": "University of Pittsburgh School of Pharmacy; University of Pittsburgh Center for Antibody Therapeutics; University of Pittsburgh Center for AIDS Research",
    "source_url": "https://academic.oup.com/bib/article/25/5/bbae445/7758758",
    "source_type": "paper",
    "publication_date": "2024-07-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/bib/bbae445",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Of these, E44R experimentally improved binding (ELISA EC50 1.696 nM) and neutralization relative to HuJ3; the other proposed substitutions did not consistently improve binding, and T99R failed to express.",
      "unknown": "This is a single, previously known J3 nanobody humanization/optimization case, not de novo antibody discovery, animal efficacy or clinical evidence. Only one of five proposed mutants improved binding; several predictions failed and the authors note limitations in prediction accuracy and manual steps. E44R remained less potent than parental J3.",
      "ai_role": "The protocol modeled the HuJ3–gp120 complex, nominated interface hotspots and computationally ranked five single-residue mutations. Researchers synthesized/tested variants in binding and HIV-1 neutralization assays; the measured E44R improvement is the experimental outcome, while most predictions did not improve performance.",
      "evidence_urls": [
        "https://academic.oup.com/bib/article/25/5/bbae445/7758758"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-06-generative-machine-learning-nominates-a-chiral-ligand-validated-in-asymmetric-ch-activation",
    "title": "Generative machine learning nominates a chiral ligand validated in asymmetric C–H activation",
    "summary": "A transfer-learned chemical language model generated novel chiral amino-acid ligands for palladium-catalyzed asymmetric β-C(sp³)–H activation, while an ensemble of 30 fine-tuned deep-learning regressors predicted reaction enantioselectivity. From 490 generated ligand candidates, mechanistic and synthetic filters retained 73; the authors tested selected reactions prospectively without disclosing the model-predicted enantioselectivity to the experimentalists. One highlighted reaction using generated ligand L5 gave 94% ee, matching the model prediction of 94%.",
    "field": "chemistry",
    "ai_technology": "Transfer-learned ULMFiT chemical language model (FnG) for ligand generation and a 30-model deep-learning ensemble (EnP) for enantioselectivity prediction",
    "organization": "Indian Institute of Technology Bombay; Scripps Research",
    "source_url": "https://pubs.rsc.org/en/content/articlepdf/2025/sc/d5sc01098e",
    "source_type": "paper",
    "publication_date": "2025-06-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D5SC01098E",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "From 490 generated ligand candidates, mechanistic and synthetic filters retained 73; the authors tested selected reactions prospectively without disclosing the model-predicted enantioselectivity to the experimentalists. The coauthor collaboration tested newly generated candidates; blinding experimentalists to model predictions is distinct from independent follow-up replication.",
      "unknown": "The accepted manuscript reports proof-of-concept prospective chemistry rather than a broadly validated catalyst family: only a small, expert-selected subset was tested and the authors report that some generated ligands were infeasible or failed to form product. The highlighted result is a ligand for an established C–H activation reaction, not a new reaction class; it does not establish scale-up or utility beyond tested conditions. The selected reaction gave 94% ee, but the accessible text table names it L5 and summarizes substrate/coupling partner graphically rather than in searchable prose.",
      "ai_role": "FnG generated candidate chiral ligands from a model fine-tuned on known ligands, while EnP ranked reactions formed by pairing generated ligands with known substrates/coupling partners. The researchers selected a small set for blinded prospective wet-lab testing; the paper reports one highlighted new ligand/reaction at 94% ee, matching prediction.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlepdf/2025/sc/d5sc01098e"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-08-generative-antibody-language-model-yields-a-pseudovirus-neutralizing-xbb-spike-binder",
    "title": "Generative antibody language model yields a pseudovirus-neutralizing XBB spike binder",
    "summary": "Tencent AI Lab and collaborators developed PALM-H3, a pretrained antibody language model that generates heavy-chain CDRH3 sequences, and A2binder to rank antigen-antibody binding. For SARS-CoV-2 XBB, which the paper says was not present during training, the selected generated antibody (Artificial 1) bound spike and neutralized pseudovirus in vitro, with reported KD 0.13 nM and IC50 0.00301 μg/mL. The study also assayed generated antibodies against wild-type, Alpha and Delta; performance varied by antibody and variant.",
    "field": "biology",
    "ai_technology": "PALM-H3 pretrained antibody generative language model and A2binder antigen-antibody affinity predictor",
    "organization": "Tencent AI Lab; Peking University Shenzhen Graduate School; Xijing Hospital of Digestive Diseases, Fourth Military Medical University",
    "source_url": "https://www.nature.com/articles/s41467-024-50903-y",
    "source_type": "paper",
    "publication_date": "2024-08-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-50903-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "For SARS-CoV-2 XBB, which the paper says was not present during training, the selected generated antibody (Artificial 1) bound spike and neutralized pseudovirus in vitro, with reported KD 0.13 nM and IC50 0.00301 μg/mL.",
      "unknown": "The functional evidence is in vitro, using SARS-CoV-2 pseudovirus rather than live-virus or clinical outcomes. The study tested selected antibody candidates; several generated candidates performed substantially worse, so the reported top result is not representative of all generated sequences. The stated novelty is experimental neutralization against the XBB variant by a generated antibody, not proof of a treatment or protection in people.",
      "ai_role": "PALM-H3 generated CDRH3 sequences conditioned on antigen specificity using a RoFormer-based antibody model; A2binder scored antibody-antigen pairs to select candidates for binding, surface-plasmon-resonance and pseudovirus neutralization tests. The paper reports it generated and selected an XBB spike-binding antibody despite XBB being absent from its training set.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-50903-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-proby-screening-uncovers-target-selective-fluorescent-probes-for-tau-bcl-2-and-tdp-43",
    "title": "PROBY screening uncovers target-selective fluorescent probes for tau, BCL-2 and TDP-43",
    "summary": "The PROBY deep-learning pipeline screened 26,416 target-validated molecules for useful optical properties, identified 11 candidates for testing and found target-responsive probes PE859 (tau), obatoclax (BCL-2) and B3 (TDP-43). The study reports selectivity assays, cell/tissue labeling and ex-vivo tumor imaging with obatoclax. AI-guided structural modification of PE859 yielded 859-2, which enabled in-vivo two-photon and cleared-brain imaging of tau pathology in P301S transgenic mice; its measured tau-aggregate dissociation constant was 54.8 nM.",
    "field": "chemistry",
    "ai_technology": "PROBY two-stage message-passing neural network pipeline (PROBY-L1 fluorescence classification and PROBY-L2 photophysical-property prediction)",
    "organization": "Children’s Hospital of Chongqing Medical University; Stanford University School of Medicine; Harvard Medical School and Massachusetts General Hospital; Sichuan University; South-Central Minzu University",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41441857/",
    "source_type": "paper",
    "publication_date": "2025-12-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202515604",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The PROBY deep-learning pipeline screened 26,416 target-validated molecules for useful optical properties, identified 11 candidates for testing and found target-responsive probes PE859 (tau), obatoclax (BCL-2) and B3 (TDP-43).",
      "unknown": "The PROBY hits are repurposed molecules from a target-validated library, not newly synthesized compounds. Target/probe performance is assay-specific; tau in-vivo imaging was demonstrated in a small transgenic-mouse study (n=3), not humans. This entry describes probe function and imaging results, not treatment effects or diagnostic performance in patients.",
      "ai_role": "PROBY ranked target-validated compounds by predicted fluorescence and seven optical properties, enabling selection and experimental verification of target-specific probes; the team then used AI-predicted structural modification alongside synthesis and bioassays to optimize PE859 into 859-1/859-2.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41441857/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-cyclerfdiffusion-designed-tp4-is-a-reproducible-cyclic-peptide-binder-to-trem2",
    "title": "CycleRFdiffusion-designed TP4 is a reproducible cyclic peptide binder to TREM2",
    "summary": "A structure-conditioned workflow combining CycleRFdiffusion, ProteinMPNN and HighFold generated and screened 1,500 cyclic peptide–TREM2 complexes and prioritized four candidates. The lead TP4 showed weak but reproducible TREM2 binding in spectral-shift, microscale-thermophoresis and surface-plasmon-resonance assays. The authors report favorable plasma stability and moderate metabolic stability, presenting the work as an initial proof of concept for TREM2-targeted cyclic binders.",
    "field": "biology",
    "ai_technology": "CycleRFdiffusion backbone design, ProteinMPNN sequence design and HighFold structural prediction/screening",
    "organization": "Weill Cornell Medicine; Macao Polytechnic University; University of Gdańsk",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41435973/",
    "source_type": "paper",
    "publication_date": "2025-12-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.bmcl.2025.130512",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The lead TP4 showed weak but reproducible TREM2 binding in spectral-shift, microscale-thermophoresis and surface-plasmon-resonance assays.",
      "unknown": "The only reported lead, TP4, is described as a weak binder. The evidence is biophysical binding and stability profiling, not demonstrated modulation of TREM2 signaling, microglial function, disease pathology or therapeutic efficacy. Treat this as proof-of-concept binder discovery.",
      "ai_role": "The AI pipeline generated and ranked 1,500 candidate cyclic peptide–TREM2 complexes based on structure, sequence and predicted energetics, from which four were selected for experiments; TP4 was then assessed with multiple orthogonal binding assays.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41435973/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-ai-guided-natural-product-inspired-molecules-inhibit-two-multidrug-resistance-transporters",
    "title": "AI-guided natural-product-inspired molecules inhibit two multidrug-resistance transporters",
    "summary": "Combining marine natural-product fragment mining, AI-based molecular generation and structure-guided selection, the authors designed optimized analogues Ib18 and It12 targeting the ABC transporters P-glycoprotein (ABCB1) and BCRP (ABCG2). Both analogues reversed transporter-mediated resistance in overexpressing cells and outperformed reference inhibitors. Target-engagement assays supported transporter inhibition without expression downregulation. In xenografts, Ib18 increased mitoxantrone antitumor activity and tumor accumulation without a detected increase in systemic toxicity.",
    "field": "medicine",
    "ai_technology": "AI-based molecular generation integrated with marine natural-product fragment mining and structure-guided prioritization",
    "organization": "Zhejiang University of Technology; Zhejiang–Egypt Joint Laboratory on Intelligent Discovery of Marine Drugs; Zhejiang Provincial People’s Hospital; University of Pittsburgh",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42720471/",
    "source_type": "paper",
    "publication_date": "2026-09-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jmedchem.6c02170",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Target-engagement assays supported transporter inhibition without expression downregulation.",
      "unknown": "The evidence is cellular and preclinical xenograft evidence, not patient efficacy or established clinical safety. The abstract reports no systemic toxicity increase in the studied xenograft setting, which should not be generalized. It does not give quantitative potency values or detailed comparator identities in the abstract.",
      "ai_role": "AI-based generation contributed candidates to a natural-product-inspired library; structure-guided prioritization selected optimized analogues for synthesis and functional tests. The authors tested Ib18 and It12 in transporter-overexpressing cells and evaluated Ib18 with mitoxantrone in xenografts.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42720471/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-limits-guided-peptide-design-yields-selective-anti-tuberculosis-host-defense-peptides",
    "title": "LIMITS-guided peptide design yields selective anti-tuberculosis host-defense peptides",
    "summary": "The Penn State/University of Minnesota team used the Lead Informed Machine Interrogation of Therapeutic Sequences (LIMITS) workflow to train random-forest and neural-network predictors on approximately 100 host-defense peptides, using lead mutational scans and biophysical sequence features. Experimentally tested neural-network-designed MAD5.1 and MAD5.3 showed reported MICs of 5 μM against M. tuberculosis H37Rv, selectivity indices >16 against an eight-organism non-mycobacterial panel, and mammalian-cell toxicity indices >100. The study identifies peptide length as a potentially underappreciated activity determinant.",
    "field": "medicine",
    "ai_technology": "LIMITS (Lead Informed Machine Interrogation of Therapeutic Sequences), using random-forest and feed-forward neural-network predictors",
    "organization": "Pennsylvania State University; University of Minnesota",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40310992/",
    "source_type": "paper",
    "publication_date": "2025-05-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.biomac.5c00244",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimentally tested neural-network-designed MAD5.1 and MAD5.3 showed reported MICs of 5 μM against M. tuberculosis H37Rv, selectivity indices >16 against an eight-organism non-mycobacterial panel, and mammalian-cell toxicity indices >100.",
      "unknown": "This is preclinical in-vitro peptide evidence, not animal efficacy or clinical treatment. The paper reports that MAD5.3 did not produce a statistically significant reduction in H37Rv viability after seven days of culture despite its MIC and selectivity/toxicity metrics; do not generalize the headline activity to every candidate or assay. MAD2.2 was moderately hemolytic. The 5 μM H37Rv MIC and high indices apply to selected peptides and stated assay conditions.",
      "ai_role": "Lead-candidate mutational scanning informed feature selection and training on a limited ~100-peptide dataset; the models iteratively selected candidate anti-TB peptide sequences for synthesis and bacteriological testing. NN-predicted MAD5 sequences were assayed against virulent and avirulent M. tuberculosis strains and counter-screened against other bacteria and mammalian cells.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40310992/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-ai-generates-compounds-with-experimentally-validated-pancreatic-cancer-cell-selectivity",
    "title": "AI generates compounds with experimentally validated pancreatic cancer cell selectivity",
    "summary": "The IRB Barcelona team trained cell-line cytotoxicity predictors on data from more than 11,000 compounds assayed across pancreatic-cancer and control cell lines, then integrated the models with a reinforcement-learning generative system to design molecules for cell-selective effects. In follow-up tests, 45 AI-designed molecules or close analogues were assayed; 20 showed the intended cell-specific cytotoxicity, and dose-response testing confirmed 18 of those 20. Across the seven selectivity exercises, at least one hit was found in five; measured IC50 values could be calculated for 12 confirmed hits and ranged from 0.97 to 16.6 μM in the targeted cell lines.",
    "field": "chemistry",
    "ai_technology": "Cell-line-specific cytotoxicity prediction models coupled to a reinforcement-learning molecular generator (REINVENT-based workflow)",
    "organization": "Institute for Research in Biomedicine (IRB Barcelona), The Barcelona Institute of Science and Technology; ICREA",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42162204/",
    "source_type": "paper",
    "publication_date": "2026-05-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42004-026-02071-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The IRB Barcelona team trained cell-line cytotoxicity predictors on data from more than 11,000 compounds assayed across pancreatic-cancer and control cell lines, then integrated the models with a reinforcement-learning generative system to design molecules for cell-selective effects.",
      "unknown": "This is a proof-of-concept hit-discovery study using in-vitro cell-line assays, not evidence of tumor selectivity in patients or an animal efficacy result. Only binary cytotoxicity labels at 10 μM informed the predictors; 45 generated structures or close analogues were tested, with 20 showing desired phenotypes and 18 confirming in dose-response assays. Exact generated structures were not always the synthesized compounds, and measured IC50 spans 0.97–16.6 μM for 12 of the confirmed hits. The paper explicitly characterizes this as an early hit-discovery stage.",
      "ai_role": "The authors trained predictive models on phenotypic screening data from >11,000 compounds, then used their scores as rewards/objectives in a generative workflow to propose molecules intended to kill one pancreatic-cancer cell line while sparing another or control cells. Candidate molecules or close purchasable analogues were experimentally tested and dose-response validated.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42162204/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-08-machine-learning-generates-active-daunomycin-binding-synthetic-nucleic-acid-polymers-outside-selected-sequence-families",
    "title": "Machine learning generates active daunomycin-binding synthetic nucleic-acid polymers outside selected sequence families",
    "summary": "The study combined in-vitro selection with a conditional variational autoencoder trained on selection data to generate high-affinity highly functionalized nucleic-acid polymer (HFNAP) sequences. Experimentally generated CVAE sequences bound daunomycin with reported dissociation constants of 9–26 nM despite having no apparent primary-sequence similarity to experimentally selected polymers; competition selections found selected generated variants performed as well as or better than sequences obtained after multiple rounds of high-stringency selection.",
    "field": "chemistry",
    "ai_technology": "Conditional variational autoencoder (CVAE) trained on in-vitro selection-derived HFNAP sequence/fitness data",
    "organization": "Broad Institute of MIT and Harvard; Harvard University; Massachusetts Institute of Technology; Brigham and Women's Hospital",
    "source_url": "https://www.nature.com/articles/s41467-022-31955-4",
    "source_type": "paper",
    "publication_date": "2022-08-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-022-31955-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimentally generated CVAE sequences bound daunomycin with reported dissociation constants of 9–26 nM despite having no apparent primary-sequence similarity to experimentally selected polymers; competition selections found selected generated variants performed as well as or better than sequences obtained after multiple rounds of high-stringency selection.",
      "unknown": "The finding is target binding by synthetic HFNAPs, not therapeutic efficacy or cellular activity. Binding results are specific to daunomycin and reported assays; the authors describe sequences as having no apparent similarity rather than proving wholly unrelated functional mechanisms. Some competition performance was assessed by in-vitro selection rather than a clinical or organism-level test.",
      "ai_role": "The authors used the CVAE to condition generation of novel HFNAP sequences on predicted daunomycin-binding fitness, then synthesized/tested generated sequences in binding and competition-selection assays. The machine-learning model generated the active sequences; wet-lab selection and competition assays established their binding activity.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-022-31955-4"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-08-deep-learning-designs-abiotic-miniproteins-that-improve-antisense-oligonucleotide-nuclear-delivery",
    "title": "Deep learning designs abiotic miniproteins that improve antisense-oligonucleotide nuclear delivery",
    "summary": "A generator–predictor–optimizer deep-learning workflow trained on quantitative delivery assays was used to design abiotic Mach miniproteins. Twelve selected predicted constructs were synthesized/tested; the strongest library construct improved phosphorodiamidate morpholino oligomer (PMO) delivery nearly 20-fold over unconjugated PMO in the standardized HeLa assay, while the authors describe the best designs as up to 50-fold more effective than prior polypeptide-based variants. PMO–Mach3/4 also produced dose-dependent reporter-protein restoration in skeletal muscle, diaphragm and heart in transgenic mice after a single intravenous dose.",
    "field": "medicine",
    "ai_technology": "Deep-learning generator–predictor–optimizer workflow: nested-LSTM sequence generator, CNN activity predictor on residue chemical fingerprints, and evolutionary optimization",
    "organization": "Massachusetts Institute of Technology; Broad Institute of MIT and Harvard; Harvard Medical School; Sarepta Therapeutics",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8819921/",
    "source_type": "paper",
    "publication_date": "2021-08-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41557-021-00766-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A generator–predictor–optimizer deep-learning workflow trained on quantitative delivery assays was used to design abiotic Mach miniproteins.",
      "unknown": "The paper reports improved delivery in a standardized HeLa reporter assay and transgenic mouse reporter model, not therapeutic correction of a disease. Only selected designs were synthesized (12 from hundreds of predicted sequences); mouse experiments used an EGFP reporter transgene and short seven-day follow-up. The article separately states the library's highest construct was nearly 20-fold over PMO alone and characterizes the designs as 50-fold better than previously known polypeptide variants; these denominators are not interchangeable. Safety evidence is limited to reported assays/doses and short follow-up.",
      "ai_role": "The model generated seed cell-penetrating sequences, predicted quantitative nuclear-delivery activity from peptide fingerprints, and iteratively optimized sequences for higher predicted activity subject to length, charge, similarity and solubility constraints. The team synthesized and experimentally assayed selected model-designed peptides; this linked the AI design to measured enhanced PMO delivery and animal reporter results.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8819921/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-ai-accelerated-virtual-screening-identifies-small-molecule-binders-to-klhdc2-and-nav1-7",
    "title": "AI-accelerated virtual screening identifies small-molecule binders to KLHDC2 and Nav1.7",
    "summary": "The OpenVS platform combined target-specific neural-network active learning with RosettaVS physics-based docking to search billion-scale compound libraries. The study reported seven low-micromolar KLHDC2 degron-pocket binders and four Naᵥ1.7 VSD4 binders. KLHDC2 hit C29 competed with a degron peptide at IC50 about 2.9 μM and its predicted pose was supported by a 2.0 Å crystal structure; the strongest Naᵥ1.7 hit, Z8739902234, had IC50 1.3 μM in whole-cell electrophysiology.",
    "field": "chemistry",
    "ai_technology": "OpenVS active-learning platform with target-specific feed-forward neural-network triage, coupled to RosettaVS physics-based docking and RosettaGenFF-VS scoring",
    "organization": "University of Washington Institute for Protein Design; and Department of Biochemistry; Howard Hughes Medical Institute; University of California, Davis; Korea Institute of Science and Technology; Sungkyunkwan University",
    "source_url": "https://www.nature.com/articles/s41467-024-52061-7.pdf",
    "source_type": "paper",
    "publication_date": "2024-09-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-52061-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reported seven low-micromolar KLHDC2 degron-pocket binders and four Naᵥ1.7 VSD4 binders.",
      "unknown": "Early ligand-hit discovery only: binding/inhibition was measured in biochemical competition, biolayer interferometry and cell electrophysiology assays, not in animals or patients. The paper states further optimization is needed to improve KLHDC2 compounds toward nanomolar potency. KLHDC2 C29’s crystal structure is experimental support for one predicted binding pose; target modulation, therapeutic efficacy and safety were not established. The reported target hit rates apply to manually filtered, synthesized subsets, not the full billion-molecule library.",
      "ai_role": "The target-specific neural network used docking-labeled data to guide which molecules were prioritized for additional expensive docking over iterative multi-billion-compound screens; RosettaVS/RosettaGenFF-VS performed physics-based docking and pose/affinity scoring. The AI-guided screen nominated molecules that were synthesized and experimentally tested. Experiments confirmed KLHDC2 degron displacement and a crystal structure matching the predicted pose, plus Naᵥ1.7 binding/inhibition in patch clamp. The paper attributes discovery to the combined AI-accelerated and physics-based workflow, not to a generative model alone.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-52061-7.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-ai-ranked-sars-cov-2-protein-interactions-guide-discovery-of-an-nsp10nsp16-inhibitor",
    "title": "AI-ranked SARS-CoV-2 protein interactions guide discovery of an NSP10–NSP16 inhibitor",
    "summary": "A machine-learning classifier applied to quantitative PPI assay data and AlphaFold-Multimer predictions to prioritize SARS-CoV-2 protein interactions; 15 of 34 high-confidence interactions were not previously listed in IMEx. The team then targeted the NSP10–NSP16 methyltransferase interface in an ultra-large virtual screen. Compound 459 bound NSP10 (MST affinity about 13 μM), reduced NSP10–NSP16 interaction in cells (BRET IC50 9.2 μM), inhibited complex methyltransferase activity by about 50% at 100 μM, and reduced SARS-CoV-2 reporter replication in cells (IC50 39.5 μM).",
    "field": "chemistry",
    "ai_technology": "Multi-adaptive support-vector-machine (maSVM) classifier for quantitative protein–protein interaction data, coupled with AlphaFold-Multimer complex prediction; VirtualFlow/QuickVina2 and later docking methods performed compound screening",
    "organization": "Max Delbrück Center for Molecular Medicine; University of Luxembourg; University of California, San Francisco; Harvard Medical School; University of California, San Diego; University of Toronto; University of Cambridge; University of Antwerp and collaborators",
    "source_url": "https://link.springer.com/article/10.1038/s44320-024-00019-8",
    "source_type": "paper",
    "publication_date": "2024-03-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44320-024-00019-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning classifier applied to quantitative PPI assay data and AlphaFold-Multimer predictions to prioritize SARS-CoV-2 protein interactions; 15 of 34 high-confidence interactions were not previously listed in IMEx.",
      "unknown": "The paper reports an experimental small-molecule hit, not a drug or clinical effect. Compound 459 has micromolar affinity and cellular antiviral IC50; authors explicitly state it requires extensive optimization. Methyltransferase inhibition was about 50% at 100 μM in one initial assay; subsequent data include cellular interaction and reporter-replication experiments. The NSP10–NSP16 interface was previously known and had prior virtual screening, but this paper reports a different compound scaffold and tested outcomes; shared target does not make this a duplicate. In vitro and cell assays do not demonstrate efficacy in animals or people.",
      "ai_role": "The maSVM algorithm prioritized PPI interactions using quantitative LuTHy binary-assay data and AlphaFold-Multimer predictions, supporting selection and mapping of SARS-CoV-2 interactions, including the NSP10–NSP16 interface and interface hot spots. The drug-like compound hit itself was identified by VirtualFlow and docking against the selected interface, not generated by the classifier. Compound 459 was then experimentally tested for protein binding, interaction disruption, methyltransferase activity and viral replication.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1038/s44320-024-00019-8"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-10-transcriptome-based-machine-learning-reveals-an-mcl1-dependency-in-alveolar-soft-part-sarcoma",
    "title": "Transcriptome-based machine learning reveals an MCL1 dependency in alveolar soft part sarcoma",
    "summary": "A support-vector-regression model trained to infer gene dependencies from RNA-sequencing profiles predicted selective MCL1 dependence in alveolar soft part sarcoma (ASPS), distinct from TFE3-translocation renal cell carcinoma. The authors tested the prediction with CRISPR knockout and the MCL1 inhibitor murizatoclax in two ASPS cell lines; the reported viability effects were greater in ASPS than comparator tRCC lines.",
    "field": "biology",
    "ai_technology": "Transcriptome-based support-vector regression model (RBF kernel; 657 highly predictable gene dependencies)",
    "organization": "Dana-Farber Cancer Institute; Broad Institute of MIT and Harvard",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11527139/",
    "source_type": "paper",
    "publication_date": "2024-10-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1101/2024.10.24.620074",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors tested the prediction with CRISPR knockout and the MCL1 inhibitor murizatoclax in two ASPS cell lines; the reported viability effects were greater in ASPS than comparator tRCC lines.",
      "unknown": "bioRxiv preprint, not peer reviewed as stated by PMC. Validation was in two ASPS cell lines with CRISPR and inhibitor assays; it does not establish patient benefit or clinical efficacy. The model nominates dependencies and is not a functional screen itself.",
      "ai_role": "The model predicted per-gene dependency scores from tumor/cell-line expression profiles and prioritized selective vulnerabilities. The study then tested MCL1 dependence in ASPS using three distinct sgRNAs and murizatoclax; those experiments, rather than the prediction alone, provide the reported validation.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11527139/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-prot-vae-designs-synthetic-enzymes-with-measured-activity-and-high-temperature-stability",
    "title": "ProT-VAE designs synthetic enzymes with measured activity and high-temperature stability",
    "summary": "The transformer–variational-autoencoder ProT-VAE generated synthetic enzyme sequences that were tested experimentally. Among 190 designed human phenylalanine-hydroxylase sequences, 69 had measurable activity and 19 exceeded wild-type activity, with the best at 2.5× wild type. A designed γ-carbonic anhydrase had a measured melting temperature of 116 °C (61 °C above the most thermostable sequence previously reported by the authors) and remained active in 23% methyl diethanolamine at pH 11.25 and 93 °C.",
    "field": "biology",
    "ai_technology": "Protein Transformer Variational AutoEncoder (ProT-VAE), using ProtT5nv transformer representations and a family-specific variational autoencoder",
    "organization": "Evozyne; NVIDIA",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12541330/",
    "source_type": "paper",
    "publication_date": "2025-10-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2408737122",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The transformer–variational-autoencoder ProT-VAE generated synthetic enzyme sequences that were tested experimentally.",
      "unknown": "The reported outcomes are laboratory expression/activity/stability tests on selected synthetic sequences; most generated candidates were not active or did not express. The 0.8× activity comparison is relative to bovine α-carbonic anhydrase under the stated assay, and the 61 °C comparison is against the previously reported thermostable γ-CA referenced by the authors. Industrial relevance is a proposed application, not demonstrated process performance.",
      "ai_role": "The generative model learned protein sequence–function patterns, produced candidate enzyme sequences conditioned in a learned latent space, and guided selection for synthesis. The authors measured enzyme function and stability experimentally; AlphaFold was used only for structure predictions of selected designs.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12541330/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-03-teleprot-guided-engineering-yields-a-nuclease-variant-with-improved-neutral-ph-biofilm-degradation",
    "title": "TeleProt-guided engineering yields a nuclease variant with improved neutral-pH biofilm degradation",
    "summary": "A machine-learning-guided design and high-throughput screening campaign produced a Bacillus licheniformis NucB variant with 19-fold higher measured specific activity than wild type at pH 7. At 1 μM in a laboratory biofilm assay at neutral pH, the selected variant degraded 71% ± 2% of the biofilm versus 13% ± 3% for wild-type NucB.",
    "field": "biology",
    "ai_technology": "TeleProt machine-learning-guided protein-library design framework; best-performing ML3 variant selected by model-based optimization with a deep neural network (MBO-DNN)",
    "organization": "X, the Moonshot Factory; Google DeepMind; Triplebar; University of Cambridge",
    "source_url": "https://doi.org/10.1016/j.cels.2025.101236",
    "source_type": "paper",
    "publication_date": "2025-03-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning-guided design and high-throughput screening campaign produced a Bacillus licheniformis NucB variant with 19-fold higher measured specific activity than wild type at pH 7.",
      "unknown": "The evidence is from enzyme and in-vitro B. licheniformis biofilm assays; it does not show wound healing, animal efficacy or clinical benefit. At the 1 μM dose, the paper reports the stated biofilm comparison; conditions and estimates are study-specific. This is engineering of a known nuclease, not discovery of a naturally new enzyme or organismal mechanism.",
      "ai_role": "TeleProt combined sequence-evolutionary information with experimental genotype–phenotype data to propose diverse NucB variants for high-throughput microfluidic activity screening. The selected ML-guided variant was then purified and assessed in a lower-throughput nuclease assay and a neutral-pH biofilm degradation assay.",
      "evidence_urls": [
        "https://doi.org/10.1016/j.cels.2025.101236"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-frase-bot-screening-identifies-a-small-molecule-ligand-that-selectively-kills-cib1-dependent-cancer-cells",
    "title": "FRASE-bot screening identifies a small-molecule ligand that selectively kills CIB1-dependent cancer cells",
    "summary": "FRASE-bot used a neural-network nativeness score to triage protein–ligand fragments, then used the retained structural features to seed virtual screening for Calcium and Integrin Binding Protein 1 (CIB1). Of 56 purchased virtual-screen hits, 29 showed dose-dependent CIB1 displacement; compound UNC10245380 (UNC5380) had a TR-FRET displacement IC50 of 8.0 ± 2.5 μM and selectively killed CIB1-depletion-sensitive triple-negative breast cancer cells in the reported assays, with no detected killing in an insensitive line.",
    "field": "chemistry",
    "ai_technology": "FRASE-bot: fragment-in-structural-environment screening with a neural-network nativeness scoring model",
    "organization": "University of North Carolina at Chapel Hill; University of Missouri",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11219766/",
    "source_type": "paper",
    "publication_date": "2024-07-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-49892-9",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Of 56 purchased virtual-screen hits, 29 showed dose-dependent CIB1 displacement; compound UNC10245380 (UNC5380) had a TR-FRET displacement IC50 of 8.0 ± 2.5 μM and selectively killed CIB1-depletion-sensitive triple-negative breast cancer cells in the reported assays, with no detected killing in an insensitive line.",
      "unknown": "The authors describe UNC10245380 as a starting chemical probe/lead, not a drug. Binding was inferred from peptide displacement in TR-FRET and a docking pose; the pose is not a solved small-molecule complex. Selective cell killing was in vitro at 30 μM, with no animal or clinical evidence. The earlier Research Square manuscript is the same underlying paper/result and is not counted separately.",
      "ai_role": "The neural network triaged fragments by predicted native-ligand pose likelihood; selected fragments informed pharmacophore construction and virtual screening. The resulting purchased compounds were then tested in TR-FRET binding-displacement and cell-death assays. The model did not perform the wet-lab validation.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11219766/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-chemical-language-models-produce-experimentally-confirmed-dual-ligands-for-three-target-pairs",
    "title": "Chemical language models produce experimentally confirmed dual ligands for three target pairs",
    "summary": "A generative chemical language model was fine-tuned on known ligands for paired targets. Prospective synthesis and in-vitro assays found seven of twelve tested designs modulated both intended targets, with confirmed dual activity across FXR/sEH, FXR/THRβ and PPARδ/sEH. Activities varied substantially and were often unbalanced; the strongest reported sEH inhibitor, compound 7, had IC50 17 nM with micromolar PPARδ agonism.",
    "field": "chemistry",
    "ai_technology": "Target-pair-fine-tuned chemical language model (four-layer LSTM recurrent neural network trained on SMILES, pretrained on 365,000 ChEMBL molecules)",
    "organization": "Goethe University Frankfurt; Ludwig-Maximilians-Universität München; Fraunhofer Institute for Translational Medicine and Pharmacology",
    "source_url": "https://www.nature.com/articles/s41467-024-52060-8",
    "source_type": "paper",
    "publication_date": "2024-09-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-52060-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Prospective synthesis and in-vitro assays found seven of twelve tested designs modulated both intended targets, with confirmed dual activity across FXR/sEH, FXR/THRβ and PPARδ/sEH.",
      "unknown": "This is in-vitro activity of a small set of synthesized molecules, not evidence of therapeutic efficacy or synergistic clinical benefit. Potencies were often unbalanced; several selected candidates were active on only one intended target, and the authors caution the success rate may not generalize beyond target pairs with compatible ligand chemistry. Lead optimization remains necessary.",
      "ai_role": "The pooled fine-tuning strategy combined known ligands of each target pair to bias de-novo SMILES generation toward both ligand classes. Researchers used model sampling frequency, similarity and docking to select molecules for synthesis; assays then tested receptor agonism and soluble epoxide hydrolase inhibition. The model proposed chemical candidates, while biological activity was measured experimentally.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-52060-8"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-iterative-machine-learning-identifies-potent-inhibitors-of-alpha-synuclein-secondary-nucleation",
    "title": "Iterative machine learning identifies potent inhibitors of alpha-synuclein secondary nucleation",
    "summary": "A structure-based iterative-learning workflow combined prior α-synuclein fibril-binding hits, a junction-tree variational autoencoder, a random-forest regressor and Gaussian-process residual model to rank compounds for repeated testing. Prospective iterations produced multiple molecules inhibiting seeded α-synuclein aggregation in vitro; lead I4.05 had a reported kinetic inhibitory constant (KIC50) of 0.52 μM (95% CI 0.45–0.59 μM), compared with extrapolated values of 18.2 μM for its parent and 36.4 μM for Anle-138b in the authors’ assay. Leads also showed activity in lipid-induced and spontaneous aggregation assays.",
    "field": "chemistry",
    "ai_technology": "Iterative structure-based learning workflow using a junction-tree variational autoencoder, random-forest regressor and Gaussian-process residual model",
    "organization": "University of Cambridge; US National Institutes of Health, National Institute of Allergy and Infectious Diseases; A*STAR; Indiana University School of Medicine",
    "source_url": "https://www.nature.com/articles/s41589-024-01580-x.pdf",
    "source_type": "paper",
    "publication_date": "2024-04-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41589-024-01580-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A structure-based iterative-learning workflow combined prior α-synuclein fibril-binding hits, a junction-tree variational autoencoder, a random-forest regressor and Gaussian-process residual model to rank compounds for repeated testing.",
      "unknown": "Evidence is biochemical/in-vitro inhibition of α-synuclein aggregation, including seeded, lipid-induced and spontaneous assays. The paper does not establish animal efficacy, neuronal protection or clinical benefit. The KIC50 and comparison values are assay-specific; no random-molecule comparator was tested, and model QSAR R² was described as modest. This is related to, but not the same paper/result as, the earlier docking-screen study that supplied starting compounds.",
      "ai_role": "The machine-learning model ranked a ZINC-derived evaluation set for potency using earlier assay measurements, then researchers tested and fed results from successive rounds back into the model. It proposed compounds for experimental testing; aggregation assays measured outcomes. Initial fibril binders came from an earlier docking-and-screening study, so this paper reports iterative optimization rather than an entirely independent start.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41589-024-01580-x.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-clickgen-designs-parp1-inhibitors-with-measured-nanomolar-activity",
    "title": "ClickGen designs PARP1 inhibitors with measured nanomolar activity",
    "summary": "ClickGen combines reaction-based molecular generation, reinforcement learning and inpainting to generate synthesizable candidate structures. In a PARP1 case study, researchers selected and synthesized three generated lead molecules; two showed nanomolar PARP1 inhibition in vitro, with reported IC50 values of 19.24 ± 1.63 nM and 69.73 ± 14.57 nM. The two compounds also showed antiproliferative activity in several tested cancer cell lines and lower toxicity in the reported normal-cell assays than the most toxic lead compounds.",
    "field": "chemistry",
    "ai_technology": "ClickGen deep-learning reaction-based molecular generator with reinforcement learning, Monte Carlo tree search and an inpainting generator",
    "organization": "Zhejiang University; Chengde Medical University; Hunan University; Macau University of Science and Technology; CarbonSilicon AI Technology Co., Ltd.; Macau Polytechnic University",
    "source_url": "https://www.nature.com/articles/s41467-024-54456-y.pdf",
    "source_type": "paper",
    "publication_date": "2024-11-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-54456-y",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a PARP1 case study, researchers selected and synthesized three generated lead molecules; two showed nanomolar PARP1 inhibition in vitro, with reported IC50 values of 19.24 ± 1.63 nM and 69.73 ± 14.57 nM.",
      "unknown": "Three selected molecules were synthesized and tested in vitro; the article reports two with nanomolar PARP1 inhibition. Cell-line antiproliferative and toxicity assays do not show tumor efficacy or safety in animals or people. The reported IC50 values are from triplicate measurements in the study. This record captures the PARP1 experimental result, not the paper’s model benchmark comparisons.",
      "ai_role": "The system generated candidate molecules by assembling reaction-compatible synthons and using reinforcement learning/inpainting to guide novelty and binding-oriented generation. Researchers computationally filtered and selected compounds, synthesized three leads and tested them; PARP1 inhibition and cell effects were experimentally measured, not generated by the model.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-54456-y.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-joint-molecular-modeling-identifies-seven-low-micromolar-kinase-hits-beyond-its-training-chemistry",
    "title": "Joint molecular modeling identifies seven low-micromolar kinase hits beyond its training chemistry",
    "summary": "A joint molecular model combined bioactivity prediction and molecule reconstruction to estimate how unfamiliar a candidate structure was to its training distribution. In a prospective screen of commercial compounds against PIM1 and CDK1, unfamiliarity-informed selection led to seven compounds with low-micromolar potency and limited similarity to training molecules. The authors also evaluated the metric across 33 experimentally labelled bioactivity datasets.",
    "field": "chemistry",
    "ai_technology": "Joint molecular model (JMM): semi-supervised molecular property-prediction and reconstruction model using a SMILES encoder; reconstruction loss defines the unfamiliarity score",
    "organization": "Eindhoven University of Technology; Centre for Living Technologies (TU/e, WUR, UU and UMC Utrecht)",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13109061/",
    "source_type": "paper",
    "publication_date": "2026-04-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42256-026-01216-w",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors also evaluated the metric across 33 experimentally labelled bioactivity datasets. The reported wet-lab candidate assays directly test the molecular-model predictions within the study.",
      "unknown": "The kinase finding is a prospective in-vitro hit-identification result, not evidence of therapeutic efficacy. Seven low-micromolar hits were reported across PIM1 and CDK1; CDK1 was described as independent of the study’s earlier dataset analysis. The study screened only a small selected set per target after computational ranking. It does not establish the unfamiliarity measure as universally superior or demonstrate animal or clinical benefit.",
      "ai_role": "The model used reconstruction-based unfamiliarity alongside predicted bioactivity and prediction uncertainty to rank molecules for a prospective virtual screen. Researchers tested selected commercial compounds against PIM1 and CDK1; seven compounds showed low-micromolar potency. AI supported prioritization and selection, while wet-lab assays established activity.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13109061/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-esm-guided-metagenome-mining-identifies-seven-new-cas12a-subtypes-with-distinct-activities",
    "title": "ESM-guided metagenome mining identifies seven new Cas12a subtypes with distinct activities",
    "summary": "AIL-Scan fine-tuned an evolutionary-scale protein language model (ESM) on Cas proteins to search metagenomic sequences and identify previously undocumented Cas12a lineages. The study classified seven previously unreported Cas12a subtypes with distinct CRISPR-locus organization, then biochemically characterized 16 Cas12a proteins from the candidate clades, documenting different DNA-cleavage and PAM-recognition preferences. The authors also demonstrated a specific assay strategy for distinguishing the KRAS G12C variant using a Cas12a variant with broader PAM recognition.",
    "field": "biology",
    "ai_technology": "AIL-Scan, using fine-tuned ESM-2 protein language-model embeddings and a machine-learning classifier to identify Cas proteins and predict Cas12a trans-cleavage activity",
    "organization": "Zhejiang Lab; Shanghai Jiao Tong University; Nanjing Medical University; Tongji University; National University of Singapore; ShanghaiTech University; Zhejiang University; Chinese Academy of Sciences",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12375029/",
    "source_type": "paper",
    "publication_date": "2025-08-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-63160-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study classified seven previously unreported Cas12a subtypes with distinct CRISPR-locus organization, then biochemically characterized 16 Cas12a proteins from the candidate clades, documenting different DNA-cleavage and PAM-recognition preferences.",
      "unknown": "The paper reports metagenomic classification and biochemical assays, not a clinical genome-editing or diagnostic application. The PAM/SNP detection demonstration is an experimental proof of concept. The trans-cleavage prediction model was trained on a small labeled set (69 proteins), with two unreported candidates used as an explicit prediction check; CRISPR loci and small-gene annotation in metagenomes remain incomplete. Distinct from other Cas-protein discovery papers by target family, candidate set and measured result.",
      "ai_role": "The fine-tuned ESM model screened metagenomic proteins for Cas identity and generated embeddings; supervised models on small experimental labels predicted trans-cleavage activity to help select uncharacterized candidates. Researchers then manually inspected candidate loci and experimentally measured binding, cleavage and PAM preferences. AI supported sequence-family identification and candidate prioritization; laboratory assays established the reported biochemical properties.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12375029/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-11-evo-genomic-language-model-designs-functional-toxin-antitoxin-and-anti-crispr-systems",
    "title": "Evo genomic language model designs functional toxin-antitoxin and anti-CRISPR systems",
    "summary": "Evo used genomic context to generate functional bacterial defence-system components. The paper reports an AI-generated toxin and low-sequence-similarity antitoxins that neutralized it in growth assays, including antitoxins that also acted against multiple natural toxins; it also reports five generated anti-CRISPR proteins that protected bacteria against Cas9 activity in cell and phage assays. Two generated anti-CRISPR proteins lacked significant sequence similarity to known proteins. These experiments demonstrate function-guided genomic design beyond simple copying of natural sequences.",
    "field": "biology",
    "ai_technology": "Evo 1.5, a long-context generative genomic language model trained on prokaryotic DNA sequences",
    "organization": "Arc Institute; Stanford University",
    "source_url": "https://www.nature.com/articles/s41586-025-09749-7.pdf",
    "source_type": "paper",
    "publication_date": "2025-11-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09749-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports an AI-generated toxin and low-sequence-similarity antitoxins that neutralized it in growth assays, including antitoxins that also acted against multiple natural toxins; it also reports five generated anti-CRISPR proteins that protected bacteria against Cas9 activity in cell and phage assays.",
      "unknown": "This is a set of functional in-vitro bacterial assays, not evidence of therapeutic use, broad host range or ecological performance. One generated toxin had 71% sequence identity to a known RelE toxin; novelty is especially supported by low-similarity antitoxins and anti-CRISPRs, two of which lacked significant sequence identity to known proteins. Earlier Evo phage-genome design and Evo 2 regulatory-sequence work are different papers and different functional results; this row is limited to the toxin-antitoxin/anti-CRISPR findings in the cited Nature article.",
      "ai_role": "Researchers prompted Evo 1.5 with genomic contexts that associate functionally related bacterial genes and used it to generate candidate toxin-antitoxin and anti-CRISPR sequences. Candidates were filtered computationally, synthesized/expressed and tested in bacterial growth, Cas9-protection and phage assays. AI generated candidates; the biological activity was established experimentally.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-025-09749-7.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-ai-generated-virtual-spatial-proteomics-reveals-lung-cancer-tumorimmune-niches-associated-with-immunotherapy-response",
    "title": "AI-generated virtual spatial proteomics reveals lung-cancer tumor–immune niches associated with immunotherapy response",
    "summary": "HEX, a pathology-image AI model, predicts spatial expression of 40 protein biomarkers from routine H&E slides. In retrospective NSCLC cohorts, combining HEX-derived virtual spatial proteomics with histology improved prognosis prediction and, in a separate 148-patient immune-checkpoint-inhibitor cohort, improved response prediction over H&E-only and virtual-proteomics-only models. The authors interpreted virtual maps as showing co-localized T-helper and cytotoxic T-cell niches among responders and immunosuppressive macrophage/neutrophil aggregates among non-responders.",
    "field": "medicine",
    "ai_technology": "HEX (H&E to protein expression), using a pathology foundation-model backbone; MICA (multimodal integration via co-attention)",
    "organization": "Stanford University; and collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12823406/",
    "source_type": "paper",
    "publication_date": "2026-01-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-025-04060-4",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In retrospective NSCLC cohorts, combining HEX-derived virtual spatial proteomics with histology improved prognosis prediction and, in a separate 148-patient immune-checkpoint-inhibitor cohort, improved response prediction over H&E-only and virtual-proteomics-only models.",
      "unknown": "Retrospective observational cohorts; predictive associations do not establish causal mechanisms or improved patient outcomes from model-guided treatment. The 148-patient immunotherapy cohort is separate from the prognosis cohorts. The paper itself says extensive clinical validation remains necessary; do not describe the finding as a clinically validated diagnostic or prospective treatment-selection tool.",
      "ai_role": "HEX predicts expression of 40 proteins and produces spatially resolved virtual proteomics from H&E images; MICA fuses the original image with AI-derived maps for outcome prediction. Researchers used predicted marker distributions to interpret therapy-associated tumor–immune spatial patterns. These are retrospective model analyses with cohort-based clinical associations, not prospective treatment evidence.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12823406/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-05-contrastive-learning-biomarker-discovered-from-phase-2-data-retrospectively-generalizes-to-phase-3-immunotherapy-outcomes",
    "title": "Contrastive-learning biomarker discovered from phase 2 data retrospectively generalizes to phase 3 immunotherapy outcomes",
    "summary": "The Predictive Biomarker Modeling Framework (PBMF), an ensemble neural network trained with a treatment-versus-control contrastive objective, identified a composite clinicogenomic biomarker using phase 2 POPLAR data for atezolizumab versus docetaxel. When applied retrospectively to the phase 3 OAK trial, its biomarker-positive group had a treatment-versus-control overall-survival hazard ratio of 0.59 (95% CI 0.47–0.74); an interpretable distilled decision tree produced HR 0.55 (95% CI 0.43–0.70). The tree used baseline ctDNA maximum allele frequency, lesion diameter, and mutation status in KMT2D, TSC1, ATM, PDGFRA and LRP1B.",
    "field": "medicine",
    "ai_technology": "Predictive Biomarker Modeling Framework (PBMF), an ensemble of neural networks using a contrastive treatment/control objective; knowledge-distilled decision tree",
    "organization": "AstraZeneca; Tempus AI",
    "source_url": "https://ellykipkogei.io/assets/pdf/1-s2.0-S1535610825001308-main.pdf",
    "source_type": "paper",
    "publication_date": "2025-05-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.ccell.2025.03.029",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The Predictive Biomarker Modeling Framework (PBMF), an ensemble neural network trained with a treatment-versus-control contrastive objective, identified a composite clinicogenomic biomarker using phase 2 POPLAR data for atezolizumab versus docetaxel.",
      "unknown": "Retrospective reanalysis; the biomarker has not prospectively guided patient selection or treatment. The outcome comparisons do not establish benefit from applying PBMF clinically. The authors call for prospective validation and note potential cohort comparability limitations. The identified feature set includes previously studied biomarkers; the paper’s novel result is the composite PBMF-derived predictive subgroup and retrospective generalization, not discovery of each individual feature.",
      "ai_role": "PBMF searched clinicogenomic features for a treatment-specific predictive subgroup, using phase 2 POPLAR training data and contrasting atezolizumab against docetaxel outcomes. Researchers transferred the derived model/biomarker to phase 3 OAK data and distilled it into an interpretable tree. This was a retrospective analysis of trial data, not prospective trial stratification.",
      "evidence_urls": [
        "https://ellykipkogei.io/assets/pdf/1-s2.0-S1535610825001308-main.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-07-deep-learning-splice-prediction-links-a-parp2-variant-to-altered-protein-isoforms-and-influenza-susceptibility",
    "title": "Deep-learning splice prediction links a PARP2 variant to altered protein isoforms and influenza susceptibility",
    "summary": "By combining single-cell splicing-QTL analysis during influenza A infection with SpliceAI/Pangolin annotation and statistical fine-mapping, researchers prioritized causal splice variants. They experimentally characterized rs2297616 in PARP2: its A allele creates an alternative 5′ splice donor, adding 39 nucleotides and producing a protein isoform 13 amino acids longer. CRISPR-edited A549 cells showed the predicted splice and protein changes; the A-allele edited cells also had higher influenza A infection than G-allele cells. The variant was associated with higher viral burden in a panel of 96 lymphoblastoid cell lines.",
    "field": "biology",
    "ai_technology": "SpliceAI and Pangolin deep-learning splice-effect predictors, integrated with single-cell scHi-HOST sQTL mapping and SuSiE statistical fine-mapping",
    "organization": "Duke University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13663244/",
    "source_type": "paper",
    "publication_date": "2026-07-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s13059-026-04205-9",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "They experimentally characterized rs2297616 in PARP2: its A allele creates an alternative 5′ splice donor, adding 39 nucleotides and producing a protein isoform 13 amino acids longer.",
      "unknown": "The 96-donor LCL association is observational and modest (the paper reports p=0.029 for the genotype association). Infection experiments used edited A549 lung epithelial cells, not patients or an in-vivo infection model. The same locus had been linked to alternative splicing previously, but the paper provides new evidence connecting genotype to PARP2 protein isoform expression and to influenza susceptibility. Broader viral-susceptibility implications remain unproven.",
      "ai_role": "Deep-learning splice models prioritized variants predicted to alter donor/acceptor sites; statistical fine-mapping and sQTL association narrowed causal candidates. The AI-supported analysis nominated rs2297616; the authors then tested its splice and protein-isoform effects using reporter assays, CRISPR-edited A549 cells, sequencing and western blotting, and assessed influenza infection in edited cells.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13663244/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-alphafold2-guided-mining-identifies-context-independent-cytidine-deaminases-for-efficient-base-editing",
    "title": "AlphaFold2-guided mining identifies context-independent cytidine deaminases for efficient base editing",
    "summary": "Researchers used AlphaFold2 to predict structures for 1,483 cytidine deaminases, clustered the predictions and experimentally characterized representatives. The screen identified previously uncharacterized enzymes with high editing efficiency, and CD0208^P52A showed little sequence-context preference across AC/GC/TC/CC motifs, a narrow editing window and relatively low off-target activity. In mammalian cells, its base editor introduced stop codons in single- and multicopy genes without DNA double-strand breaks.",
    "field": "biology",
    "ai_technology": "AlphaFold2 protein-structure prediction, followed by structural clustering of candidate cytidine deaminases",
    "organization": "Agricultural Genomics Institute at Shenzhen, Chinese Academy of Agricultural Sciences",
    "source_url": "https://www.nature.com/articles/s41551-024-01220-8.pdf",
    "source_type": "paper",
    "publication_date": "2024-06-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41551-024-01220-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers used AlphaFold2 to predict structures for 1,483 cytidine deaminases, clustered the predictions and experimentally characterized representatives.",
      "unknown": "The base-editing experiments were preclinical cell experiments, not animal or clinical efficacy tests. AlphaFold2 supplied predicted rather than experimentally solved structures. The paper reports multiple promising deaminases; CD0208^P52A is the specific context-independent editor highlighted here. No therapeutic safety or in-vivo delivery claim is implied.",
      "ai_role": "AlphaFold2-predicted structures at scale were used to mine and cluster 1,483 putative deaminases; representatives were then selected for experimental characterization in cytosine base editors. The model enabled structure-guided candidate selection; editing outcomes were measured experimentally in mammalian cells.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41551-024-01220-8.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-gpt-3-5-reasoning-helps-connect-mre11-to-social-context-sleep-regulation-in-drosophila",
    "title": "GPT-3.5 reasoning helps connect MRE11 to social-context sleep regulation in Drosophila",
    "summary": "A study combined GPT-3.5-generated gene-function context with a genome-wide Drosophila RNAi behavioral screen. It identified MRE11 knockdown as reducing sleep in group-housed flies, a result the authors report as previously unrecognized in this social context. GPT-3.5 reasoning proposed a link from MRE11 to Dop1R1/dopaminergic signaling; treatment with a Dop1R1 antagonist partially rescued sleep, locomotor and social-activity phenotypes in MRE11-RNAi flies.",
    "field": "biology",
    "ai_technology": "GPT-3.5, prompted for genome-wide Drosophila gene-function interpretation and chain-of-thought relationship hypotheses",
    "organization": "Huazhong University of Science and Technology; and collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41467-024-48005-w",
    "source_type": "paper",
    "publication_date": "2024-05-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-48005-w",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "It identified MRE11 knockdown as reducing sleep in group-housed flies, a result the authors report as previously unrecognized in this social context.",
      "unknown": "The experiments are in Drosophila and concern sleep in the presence of conspecifics. Dop1R1 antagonist treatment only partially rescued phenotypes; this does not establish direct MRE11 regulation of the receptor or dopamine pathway. GPT-3.5 had high false-negative rates in the authors’ known-gene evaluation and contributed knowledge summarization/hypothesis generation, while the broad RNAi screen itself was experimental. No mammalian or human result is implied.",
      "ai_role": "GPT-3.5 was queried about fly gene functions and used to construct hypothesized relationships among behavioral-screen candidates; it proposed a MRE11–Dop1R1 link that researchers tested pharmacologically. The RNAi screen and behavioral measurements were experimental, while direct molecular regulation was not demonstrated.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-48005-w"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-get-and-interaction-assays-link-a-leukemia-risk-pax5-variant-to-stronger-nr2c2-binding",
    "title": "GET and interaction assays link a leukemia-risk PAX5 variant to stronger NR2C2 binding",
    "summary": "The General Expression Transformer (GET) model inferred cell-type-specific transcription-factor motif interactions from regulatory data and nominated a lymphocyte-specific interaction between PAX5 and nuclear-receptor-family factors. AlphaFold multimer modeling and PAX5 proximity-labeling in a B-ALL cell line supported interaction with NR2C2; the leukemia-associated PAX5 G183S variant showed increased PAX5–NR2C2 proximity signal. The variant-specific expression profile was enriched for predicted NR/3 and shared PAX5/NR/3 target genes, suggesting a possible regulatory mechanism for the germline risk variant.",
    "field": "biology",
    "ai_technology": "GET (General Expression Transformer) foundation model for cell-type-specific regulatory grammar; AlphaFold2/AlphaFold Multimer for structural hypothesis assessment",
    "organization": "Columbia University; Mohamed bin Zayed University of Artificial Intelligence; collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41586-024-08391-z.pdf",
    "source_type": "paper",
    "publication_date": "2025-01-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08391-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The General Expression Transformer (GET) model inferred cell-type-specific transcription-factor motif interactions from regulatory data and nominated a lymphocyte-specific interaction between PAX5 and nuclear-receptor-family factors.",
      "unknown": "This is a cell-line interaction and expression analysis, not proof that the variant causes leukemia or directly alters disease risk through NR2C2. PAX5–NR2C2 interaction itself was consistent with prior evidence for the paralogue PAX2 and mouse pre-B cells; the specific reported distinction is increased interaction signal with PAX5 G183S and associated transcriptional enrichment. The authors call the disease-driving mechanism possible and identify limitations in cell-state and variant resolution.",
      "ai_role": "GET learned regulatory syntax across 213 cell types and used motif co-regulation to predict TF–TF interactions, which nominated PAX5/nuclear receptor family interactions in B cells. Structural modeling prioritized candidate binding interfaces. Researchers then used proximity labeling and mass spectrometry in a B-ALL cell line to compare wild-type PAX5 with the G183S leukemia-risk variant.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-024-08391-z.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-paper2agent-integrates-perturbation-datasets-to-prioritize-gpr137-at-a-psoriasis-risk-locus",
    "title": "Paper2Agent integrates perturbation datasets to prioritize GPR137 at a psoriasis risk locus",
    "summary": "Paper2Agent-generated agents used AlphaGenome to prioritize GPR137 as the top predicted gene affected by psoriasis-associated variant rs887314 in CD4+ T cells, then integrated that prediction with existing MPRA-coupled scCRISPRi and CD4+ T-cell Perturb-seq data. A cross-screen correlation of gene-expression signatures supported GPR137 as the probable causal gene under stimulated conditions (Spearman rho 0.613 at 8 h and 0.630 at 48 h; FDR < 0.05), but not at rest. The paper presents this as activation-dependent computational evidence.",
    "field": "biology",
    "ai_technology": "Paper2Agent-generated AlphaGenome, MPRA-coupled scCRISPRi and CD4+ T-cell Perturb-seq agents, coordinated through an AI co-scientist",
    "organization": "Stanford University; collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41586-026-11044-y",
    "source_type": "paper",
    "publication_date": "2026-09-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-11044-y",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Paper2Agent-generated agents used AlphaGenome to prioritize GPR137 as the top predicted gene affected by psoriasis-associated variant rs887314 in CD4+ T cells, then integrated that prediction with existing MPRA-coupled scCRISPRi and CD4+ T-cell Perturb-seq data.",
      "unknown": "The analysis reuses existing published datasets and does not report new wet-lab experiments for this psoriasis case. AlphaGenome supplied a model prediction; cross-dataset signature correlation supports, but does not establish, causal biology. The GPR137 concordance was significant only after stimulation, and the authors describe it as probable/activation-dependent. A human researcher selected which AI-proposed strategy to run. Primary paper source checked; no independent replication assessed.",
      "ai_role": "The AlphaGenome paper agent scored and ranked candidate genes at rs887314. Under human-in-the-loop supervision, the AI co-scientist inspected two other paper agents' datasets, proposed ten validation strategies, and the researcher selected signature-correlation analysis. The resulting cross-screen, cross-modality analysis found significant concordance for GPR137 only under stimulated conditions.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-026-11044-y"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-05-machine-learning-links-mycobacterial-protein-disorder-to-targeted-degradation-potential",
    "title": "Machine learning links mycobacterial protein disorder to targeted degradation potential",
    "summary": "Researchers combined a genetic chemical-proximity platform and experimental screening of 72 Mycobacterium smegmatis protein candidates with machine learning. A Lasso model trained on measured degradation data identified N-terminal disorder propensity as associated with degradation potential and predicted candidate targets among 348 essential proteins conserved between M. smegmatis and M. tuberculosis. Experimentally induced degradation of selected essential M. smegmatis proteins inhibited bacterial growth and, for some targets, potentiated antimicrobial compounds.",
    "field": "biology",
    "ai_technology": "Lasso regression model trained on sequence and structural protein descriptors, including AlphaFold2 confidence and predicted disorder propensity",
    "organization": "Harvard Medical School; Massachusetts Institute of Technology; collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41467-024-48506-8",
    "source_type": "paper",
    "publication_date": "2024-05-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-48506-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers combined a genetic chemical-proximity platform and experimental screening of 72 Mycobacterium smegmatis protein candidates with machine learning.",
      "unknown": "Most mechanistic and growth evidence was generated in M. smegmatis, a non-pathogenic model, not in clinical treatment or an animal efficacy study. The machine-learning target predictions for the larger set of 348 essential proteins require orthogonal experimental validation. The study reports that targeted degradation of some proteins inhibits growth and potentiates antimicrobials; this does not establish a deployable drug or clinical benefit. Primary paper source checked; no independent replication assessed.",
      "ai_role": "Machine learning connected measured degradation constants from experimentally screened protein substrates to protein sequence/structure features, identifying N-terminal disorder propensity as a predictor of degradation potential; the trained model then ranked candidate targets among essential proteins conserved between M. smegmatis and M. tuberculosis. Experiments—not the model alone—showed degradation and antibacterial phenotypes for selected targets.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-48506-8"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-12-machine-learning-guided-closed-loop-synthesis-experimentally-realizes-unreported-dye-molecules",
    "title": "Machine-learning-guided closed-loop synthesis experimentally realizes unreported dye molecules",
    "summary": "A closed-loop molecular discovery platform combined machine-learning property prediction, automated synthesis and characterization, and iterative model updates to explore dye-like molecules. In the published Science abstract, the team reports experimentally realizing 294 previously unreported molecules across three design–make–test–analyze cycles on four rarely reported scaffolds, while also identifying six top-performing molecules in a second case study using pre-existing chemical-space examples.",
    "field": "chemistry",
    "ai_technology": "Autonomous molecular-discovery platform integrating generative and property-prediction machine-learning models, Bayesian optimization, and synthesis-planning software with automated chemical synthesis and characterization",
    "organization": "Massachusetts Institute of Technology; collaborating institutions",
    "source_url": "https://doi.org/10.1126/science.adi1407",
    "source_type": "paper",
    "publication_date": "2023-12-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In the published Science abstract, the team reports experimentally realizing 294 previously unreported molecules across three design–make–test–analyze cycles on four rarely reported scaffolds, while also identifying six top-performing molecules in a second case study using pre-existing chemical-space examples.",
      "unknown": "This is molecular discovery in dye-like chemical systems, not a validated drug or biological effect. The 294 number is from the published Science abstract; an earlier ChemRxiv preprint describes 312 unreported molecules, so version-specific counts differ and the published figure is used here. The measurements addressed absorption wavelength, lipophilicity and photo-oxidative stability; no environmental, toxicological or commercial benefit is claimed. Primary-source abstract and author-lab description checked; independent replication not assessed.",
      "ai_role": "Generative ML proposed candidate dye-like structures; property-prediction models ranked molecules against absorption, lipophilicity, and photo-oxidative-stability targets; automated multi-step synthesis and measurements generated training data for model retraining and the next cycles. The closed-loop process converted model-guided suggestions into synthesized and characterized compounds.",
      "evidence_urls": [
        "https://doi.org/10.1126/science.adi1407"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-sequential-model-based-optimization-discovers-two-previously-unreported-chemical-reactions",
    "title": "Sequential model-based optimization discovers two previously unreported chemical reactions",
    "summary": "A programmable robotic synthesis platform used inline HPLC, Raman or NMR feedback and closed-loop optimization to explore selected reaction spaces. In a tosMIC campaign, sequential model-based optimization with a Gaussian-process surrogate identified a reaction between toluenesulphonylmethyl isocyanide and benzylidenemalononitrile; subsequent optimization increased yield from the 27% initial condition to 49% (22 percentage points) over 50 reactions. The article also reports a second previously unreported reaction and an exploratory trifluoromethylation campaign that discovered new molecules.",
    "field": "chemistry",
    "ai_technology": "ChemputationOptimizer, a sensor-integrated reaction engine with sequential model-based optimization (Gaussian-process surrogate); the platform also exposes random search, design-of-experiments, Bayesian-optimization and genetic-algorithm routines",
    "organization": "University of Glasgow",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10858227/",
    "source_type": "paper",
    "publication_date": "2024-02-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-45444-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article also reports a second previously unreported reaction and an exploratory trifluoromethylation campaign that discovered new molecules.",
      "unknown": "The platform's reported yield gains are specific to these selected reaction spaces and objectives, not general chemistry performance. The tosMIC and phloroglucinol campaigns are reported as previously unreported reactions; product characterization and complete procedure details are in the article/supplement. The system includes several optimization algorithms and the authors note random search can be a strong baseline; the reported tosMIC optimization used SMBO with a Gaussian-process surrogate. No claim is made that the reactions are industrially useful or broadly generalizable. Primary source checked; no independent replication assessed.",
      "ai_role": "For the tosMIC and phloroglucinol explorations, machine-guided combinatorial search proposed reaction parameters and inline spectroscopic feedback supplied product-yield measurements; a Gaussian-process sequential-model-based optimizer guided the reaction-space exploration and follow-up optimization. Human chemists defined reaction space and experimental constraints, and performed/characterized the experiments.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10858227/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-chai-image-only-histology-biomarker-stratifies-metastatic-hormone-sensitive-prostate-cancer-outcomes-in-independent-trial-data",
    "title": "CHAI image-only histology biomarker stratifies metastatic hormone-sensitive prostate cancer outcomes in independent trial data",
    "summary": "A computational histology AI platform built a prognostic signature from diagnostic whole-slide images in 507 CHAARTED participants and evaluated the locked signature in 684 ENZAMET participants. In the independent ENZAMET cohort, the unfavorable-risk group had worse overall survival (HR 2.6, 95% CI 2.0–3.4) and progression-free survival (HR 2.2, 95% CI 1.7–2.7); associations remained after adjustment for clinical variables. The article also reports treatment-interaction analyses, but this retrospective secondary analysis does not establish prospective treatment-selection benefit.",
    "field": "medicine",
    "ai_technology": "Computational Histology Artificial Intelligence (CHAI) platform",
    "organization": "Valar Labs; and the CHAARTED/ENZAMET trial collaborators",
    "source_url": "https://ascopubs.org/doi/10.1200/PO-26-00321",
    "source_type": "paper",
    "publication_date": "2026-07-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1200/PO-26-00321",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In the independent ENZAMET cohort, the unfavorable-risk group had worse overall survival (HR 2.6, 95% CI 2.0–3.4) and progression-free survival (HR 2.2, 95% CI 1.7–2.7); associations remained after adjustment for clinical variables.",
      "unknown": "The locked image-only biomarker was independently evaluated across two prospective randomized-trial datasets, but this article is a post hoc secondary analysis, not a prospective biomarker-directed treatment trial. The treatment-interaction result is suggestive and does not prove that choosing triplet therapy by this score improves outcomes. No clinical deployment or benefit claim is made. Primary journal abstract checked; no independent replication assessed.",
      "ai_role": "CHAI extracted quantitative histomorphologic features from H&E diagnostic whole-slide images and learned an image-only risk signature on CHAARTED data; the locked signature was tested in ENZAMET. Humans specified cohorts/endpoints and analyzed the trial outcomes.",
      "evidence_urls": [
        "https://ascopubs.org/doi/10.1200/PO-26-00321"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-03-prospective-multicentre-study-combines-ev-protein-biomarkers-and-machine-learning-for-preclinical-oesophageal-cancer-detection",
    "title": "Prospective multicentre study combines EV protein biomarkers and machine learning for preclinical oesophageal cancer detection",
    "summary": "The BarFlare assay and an interpretable multi-criteria decision-making classification-fusion machine-learning framework combined extracellular-vesicle SCC and MMP13 biomarkers with clinical factors. The model was evaluated in prospective multicentre diagnostic cohorts and two population cohorts; the longitudinal cohort contained baseline samples from 30 people who later developed oesophageal squamous cell carcinoma, with a reported median lead time of 34.9 months and AUC 0.864. The paper also reports external diagnostic validation in two cohorts.",
    "field": "medicine",
    "ai_technology": "Multi-criteria decision-making Classification Fusion (MCF) machine-learning framework; 289 candidate feature-selector/classifier pairs were compared and three base models fused",
    "organization": "Sun Yat-sen University Cancer Center; and collaborating hospitals/cohorts",
    "source_url": "https://isevjournals.onlinelibrary.wiley.com/doi/10.1002/jev2.70246",
    "source_type": "paper",
    "publication_date": "2026-03-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/jev2.70246",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The BarFlare assay and an interpretable multi-criteria decision-making classification-fusion machine-learning framework combined extracellular-vesicle SCC and MMP13 biomarkers with clinical factors.",
      "unknown": "This is a prospective observational diagnostic study, not an intervention trial or proof that screening improves outcomes. The longitudinal preclinical cohort had 30 incident cases and matched controls; performance estimates require external replication in larger populations. The source reports an association and lead time, not guaranteed individual prediction. Primary publisher full text and abstract checked; no independent replication assessed.",
      "ai_role": "The MCF framework integrated newly identified EV biomarkers with clinical variables, selected and fused machine-learning classifiers, and produced the risk score evaluated in diagnostic and prospective population cohorts. BarFlare is the high-sensitivity EV assay platform, not the AI model.",
      "evidence_urls": [
        "https://isevjournals.onlinelibrary.wiley.com/doi/10.1002/jev2.70246"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-04-machine-learning-selection-of-a-six-protein-panel-for-atherosclerosis-severity-and-patient-stratification",
    "title": "Machine-learning selection of a six-protein panel for atherosclerosis severity and patient stratification",
    "summary": "Five machine-learning classifiers integrated clinical and serum-proteomic data from carotid atherosclerotic stenosis (60 participants), dyslipidemia (55) and healthy controls (66). A consensus approach identified a six-protein panel (B2M, GPV, MMP9, PLF4, TSP1 and fibrinogen/FB isoforms) that distinguished the atherosclerosis group from the comparison groups, with reported ROC-AUC above 0.9 across the tested classifiers. Protein levels were evaluated by flow cytometry and ELISA in an independent external cohort that included patients with acute atherothrombotic stroke.",
    "field": "medicine",
    "ai_technology": "Five machine-learning classifiers: Naive Bayes, linear SVM, random forest, Extra Trees and AdaBoost",
    "organization": "Biomedical Research and Innovation Institute of Cadiz, University of Cadiz and collaborating clinical institutions",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41961364/",
    "source_type": "paper",
    "publication_date": "2026-04-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s43556-026-00438-z",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A consensus approach identified a six-protein panel (B2M, GPV, MMP9, PLF4, TSP1 and fibrinogen/FB isoforms) that distinguished the atherosclerosis group from the comparison groups, with reported ROC-AUC above 0.9 across the tested classifiers.",
      "unknown": "The discovery cohorts were small (60 atherosclerosis, 55 dyslipidemia and 66 healthy controls); external validation supports the measured panel but does not establish population-wide performance, prospective clinical utility, or improved patient outcomes. The proteins have prior biological associations, so the finding is the specific integrated panel and cohort discrimination rather than discovery of six new proteins or a new mechanism.",
      "ai_role": "Five classifiers integrated clinical and serum-proteomic features and selected a consensus six-protein panel for patient-group discrimination. The authors then evaluated the selected proteins in an independent clinical cohort using flow cytometry and ELISA.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41961364/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-esm-fep-combines-protein-language-model-fitness-screening-and-free-energy-calculations-to-engineer-herbicide-detoxifying-maize-dioxygenase",
    "title": "ESM-FEP combines protein-language-model fitness screening and free-energy calculations to engineer herbicide-detoxifying maize dioxygenase",
    "summary": "ESM-FEP was applied to Zea mays dioxygenase ZmHSL1B and nominated a quadruple mutant (Q140H/Y205F/L332R/K336F) with approximately sevenfold higher catalytic efficiency for mesotrione detoxification than wild type, measured by in-vitro assays and kinetic analysis. Arabidopsis thaliana expressing the engineered variant showed enhanced herbicide tolerance in the reported transgenic tests.",
    "field": "biology",
    "ai_technology": "ESM-FEP (fine-tuned ESM-2 protein language model plus alchemical free-energy perturbation)",
    "organization": "Central China Normal University, State Key Laboratory of Green Pesticide",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42675858/",
    "source_type": "paper",
    "publication_date": "2026-08-31",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.xplc.2026.102088",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "ESM-FEP was applied to Zea mays dioxygenase ZmHSL1B and nominated a quadruple mutant (Q140H/Y205F/L332R/K336F) with approximately sevenfold higher catalytic efficiency for mesotrione detoxification than wild type, measured by in-vitro assays and kinetic analysis.",
      "unknown": "Reported results concern one maize dioxygenase and mesotrione in specified in-vitro and Arabidopsis tests. They do not establish crop-field performance, commercial efficacy, or generality across herbicides/enzymes. Primary PubMed abstract and publisher citation metadata checked; no independent replication assessed.",
      "ai_role": "Fine-tuned ESM-2 screened saturation-mutagenesis variants for retained global protein fitness; free-energy perturbation then estimated substrate-binding effects and prioritized mutants. The selected M5 mutant was assayed in vitro and in transgenic Arabidopsis by the researchers.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42675858/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-04-esm-ezy-mining-identifies-multicopper-oxidases-and-asparaginases-with-experimentally-improved-catalytic-properties",
    "title": "ESM-Ezy mining identifies multicopper oxidases and asparaginases with experimentally improved catalytic properties",
    "summary": "The ESM-1b semantic-embedding mining strategy nominated low-sequence-similarity enzyme candidates. In the reported experimental assessment, 44% of tested multicopper oxidases outperformed their query enzymes in at least one measured property (including catalytic efficiency, heat/solvent tolerance or pH stability); 51% were reported to excel in environmental-remediation applications. Among mined L-asparaginases, 40% had higher specific activity and catalytic efficiency than query enzymes.",
    "field": "biology",
    "ai_technology": "ESM-Ezy, using ESM-1b protein-language-model embeddings and semantic-space similarity",
    "organization": "Westlake University; and collaborating institutions",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40188191/",
    "source_type": "paper",
    "publication_date": "2025-04-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-58521-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In the reported experimental assessment, 44% of tested multicopper oxidases outperformed their query enzymes in at least one measured property (including catalytic efficiency, heat/solvent tolerance or pH stability); 51% were reported to excel in environmental-remediation applications.",
      "unknown": "Percentages describe the selected candidates and tested property comparisons in this study, not the fraction of all database enzymes that will be superior. The reported remediation suitability is assay-specific; environmental or industrial deployment is not established. Primary abstract/full-text search result checked; no independent replication assessed.",
      "ai_role": "ESM-1b sequence embeddings and semantic-space similarity ranked low-similarity candidate enzymes from sequence databases for experimental follow-up; assays assessed catalytic properties of selected multicopper oxidases and L-asparaginases.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40188191/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-pepmlm-designs-experimentally-binding-target-conditioned-peptide-ligands-including-peptides-used-for-targeted-protein-degradation",
    "title": "PepMLM designs experimentally binding target-conditioned peptide ligands, including peptides used for targeted protein degradation",
    "summary": "PepMLM fine-tunes ESM-2 using masked-language modeling conditioned on a target protein sequence to generate de novo linear peptide binders without structural input. In the reported experiments, selected peptides showed sequence-specific binding to targets including NCAM1 and AMHR2; incorporation into ubiquitin-based antibody constructs enabled targeted degradation of disease-relevant proteins, including Huntington’s disease drivers and viral phosphoproteins.",
    "field": "biology",
    "ai_technology": "PepMLM, ESM-2 fine-tuned with target-sequence-conditioned masked language modeling",
    "organization": "Cornell University; and collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41587-025-02761-2",
    "source_type": "paper",
    "publication_date": "2025-08-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-025-02761-2",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In the reported experiments, selected peptides showed sequence-specific binding to targets including NCAM1 and AMHR2; incorporation into ubiquitin-based antibody constructs enabled targeted degradation of disease-relevant proteins, including Huntington’s disease drivers and viral phosphoproteins.",
      "unknown": "The reported evidence is preclinical binding/degradation assays, not therapeutic efficacy, pharmacokinetics or safety. The article notes that experimental optimization/stabilization of its selected peptides was not performed and further validation is needed for viral-target applications. Primary publisher full text/abstract checked; no independent replication assessed.",
      "ai_role": "Given a target protein sequence, PepMLM generates candidate linear binder peptide sequences by reconstructing masked peptide regions. The authors tested selected designs in binding and degradation assays and reported target-specific activity in the stated systems.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41587-025-02761-2"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-07-machine-learning-analysis-of-plasma-metabolomic-fingerprints-yields-a-21-metabolite-gastric-cancer-panel-validated-across-multicentre-cohorts",
    "title": "Machine-learning analysis of plasma metabolomic fingerprints yields a 21-metabolite gastric-cancer panel validated across multicentre cohorts",
    "summary": "In a multicentre study, machine-learning analysis of plasma metabolic fingerprints from 1,944 retrospective participants selected a 21-metabolite panel that differentiated gastric-cancer from non-cancer cohorts; five model types achieved AUCs of 0.921–0.971 in discovery and 0.907–0.940 in external verification. Prospective testing in 264 participants using both NPELDI-MS and UPLC-MS reported diagnostic AUCs of 0.855–0.918 and 0.856–0.916, respectively. The paper also reports a four-metabolite prognostic score based on retrospective survival data.",
    "field": "medicine",
    "ai_technology": "Neural network, lasso regression, ridge regression, support-vector machine and random-forest models applied to plasma metabolic fingerprints",
    "organization": "Zhejiang Cancer Hospital; Shanghai Jiao Tong University and six collaborating clinical centers",
    "source_url": "https://gut.bmj.com/content/72/11/2051",
    "source_type": "paper",
    "publication_date": "2023-07-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1136/gutjnl-2023-330045",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a multicentre study, machine-learning analysis of plasma metabolic fingerprints from 1,944 retrospective participants selected a 21-metabolite panel that differentiated gastric-cancer from non-cancer cohorts; five model types achieved AUCs of 0.921–0.971 in discovery and 0.907–0.940 in external verification.",
      "unknown": "The panel was evaluated in selected clinical cohorts and is not established for routine screening or clinical use. Prospective validation was at one lead center, and only six panel metabolites were robustly detected by confirmatory UPLC-MS under the reported testing. Model performance does not show that the panel is causal or improves outcomes. Primary full text/abstract checked; no independent replication assessed.",
      "ai_role": "The study used machine-learning feature selection/modeling (including logistic-regression coefficient filtering) to identify and evaluate a 21-metabolite diagnostic panel from high-throughput mass-spectrometry fingerprints; classifiers were then validated on independent retrospective and prospective cohorts.",
      "evidence_urls": [
        "https://gut.bmj.com/content/72/11/2051"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-08-machine-learning-guided-csra-homolog-screens-reveal-divergent-bacterial-regulatory-activities",
    "title": "Machine-learning-guided CsrA homolog screens reveal divergent bacterial regulatory activities",
    "summary": "A Swarm-seq assay tested more than 500 bacterial CsrA homologs for regulation of Bacillus subtilis swarming and found two functional classes: several homologs strongly inhibited swarming while others failed despite conserved sequence and canonical GGA motifs in the hag transcript. A random-forest model trained on these assays predicted homolog activity across thousands of sequences; experimental testing supported predicted Class I regulatory activity for Bdellovibrio bacteriovorus CsrA and Pseudomonas putida RsmA. The findings indicate sequence conservation and the canonical RNA motif do not alone determine these tested regulatory effects.",
    "field": "biology",
    "ai_technology": "Random-forest machine-learning classifier trained on Swarm-seq functional assay data",
    "organization": "University of Chicago",
    "source_url": "https://academic.oup.com/nar/article/54/14/gkag756/8753004",
    "source_type": "paper",
    "publication_date": "2026-08-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/nar/gkag756",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Swarm-seq assay tested more than 500 bacterial CsrA homologs for regulation of Bacillus subtilis swarming and found two functional classes: several homologs strongly inhibited swarming while others failed despite conserved sequence and canonical GGA motifs in the hag transcript.",
      "unknown": "Activity classifications are based on the reported B. subtilis swarming assay and tested proteins; they do not establish all native targets or regulatory functions in each source organism. The predicted function landscape across thousands of homologs is not experimentally validated wholesale. Primary publisher abstract checked; no independent replication assessed.",
      "ai_role": "High-throughput swarming assays generated a functional dataset; machine learning was then trained to classify homolog functional activity across the bacterial domain and nominate homologs for validation. Selected CsrABb and RsmAPp predictions were experimentally validated.",
      "evidence_urls": [
        "https://academic.oup.com/nar/article/54/14/gkag756/8753004"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-grace-designs-two-carbonic-anhydrase-like-enzymes-with-measured-catalytic-activity",
    "title": "GRACE designs two carbonic-anhydrase-like enzymes with measured catalytic activity",
    "summary": "The GRACE workflow combined RFdiffusion structure generation, ProteinMPNN sequence design, CLEAN enzyme classification and solubility/structure filtering to select two carbonic-anhydrase-like sequences from 10,000 candidates. The authors report experimental expression and carbonic anhydrase activity: dCA12_2 at 400 WAU/mL and dCA23_1 at 260 WAU/mL in the reported assay.",
    "field": "biology",
    "ai_technology": "GRACE (RFdiffusion, ProteinMPNN and CLEAN, with computational solubility/structural filters)",
    "organization": "National Cheng Kung University; and collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11669173/",
    "source_type": "paper",
    "publication_date": "2024-11-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acssynbio.4c00624",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors report experimental expression and carbonic anhydrase activity: dCA12_2 at 400 WAU/mL and dCA23_1 at 260 WAU/mL in the reported assay.",
      "unknown": "The reported activity values are from the study's carbonic-anhydrase assay and do not establish purified-enzyme kinetics, physiological CO2-fixation performance, or industrial utility. The designs and validation are a two-protein proof of concept; no independent replication assessed.",
      "ai_role": "The generative-design workflow produced structures/sequences and machine-learning function classification prioritized two candidates from a 10,000-sequence pool; researchers expressed both and measured carbonic-anhydrase activity.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11669173/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-self-iterative-learning-finds-rna-targeting-prokaryotic-argonautes-and-an-rna-modification-sensing-activity",
    "title": "Self-iterative learning finds RNA-targeting prokaryotic Argonautes and an RNA-modification-sensing activity",
    "summary": "The self-iterative hierarchical ensemble model (SIM) nominated six high-confidence prokaryotic Argonaute candidates; five were experimentally validated as RNA-targeting proteins. The study identified three uncharacterized Agos with a novel N-terminal domain, defining a previously unrecognized subclass. Biochemical and in-vivo experiments on Haloferax profundi Ago (HpAgo) established RNA cleavage and a distinctive RNA-modification-sensing capability, which the authors used to build a method for quantifying modified RNA.",
    "field": "biology",
    "ai_technology": "Self-Iterative hierarchical ensemble Model (SIM), an ensemble-learning and self-training pipeline",
    "organization": "Shanghai Jiao Tong University; ShanghaiTech University and collaborating institutions",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42039271/",
    "source_type": "paper",
    "publication_date": "2026-01-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.apsb.2026.01.009",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The self-iterative hierarchical ensemble model (SIM) nominated six high-confidence prokaryotic Argonaute candidates; five were experimentally validated as RNA-targeting proteins.",
      "unknown": "Five of six selected candidates were validated; the broader predicted sequence space was not experimentally tested. The specific RNA modification-sensing mechanism and assay performance are supported for HpAgo and tested conditions, not all Argonautes. No therapeutic or universal detection claim is made. Primary abstract/citation record checked; no independent replication assessed.",
      "ai_role": "SIM iteratively expanded its training labels to rank prokaryotic Argonautes for RNA targeting from sparse initial data; it prioritized candidates and identified a previously unrecognized domain-containing subclass. The researchers validated RNA cleavage and RNA-modification sensing biochemically and in vivo.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42039271/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-01-machine-learning-models-identify-carbon-source-dependent-metabolic-reactions-that-control-escherichia-coli-growth",
    "title": "Machine-learning models identify carbon-source-dependent metabolic reactions that control Escherichia coli growth",
    "summary": "Elastic-net and multilayer-perceptron models integrated genome-wide gene-deletion data and simulated fluxes to predict beneficial and detrimental reactions for E. coli growth on 30 carbon sources. Beyond training-set examples, the authors experimentally validated model predictions about the glyoxylate shunt, pyruvate dehydrogenase, and redundant purine-biosynthesis reactions. The findings showed that effects of energy-generating pathways depend on carbon source, while biosynthetic pathways were more generally growth-promoting across the tested conditions.",
    "field": "biology",
    "ai_technology": "Elastic-net regression and multilayer-perceptron models",
    "organization": "Konkuk University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10912204/",
    "source_type": "paper",
    "publication_date": "2024-01-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44320-024-00017-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Beyond training-set examples, the authors experimentally validated model predictions about the glyoxylate shunt, pyruvate dehydrogenase, and redundant purine-biosynthesis reactions.",
      "unknown": "Reported reaction effects are for E. coli under the tested carbon-source and genetic-perturbation conditions; they do not establish universal metabolic-control rules. The results should not be interpreted as direct clinical or industrial outcomes. Primary PubMed abstract and PMC article record checked; no independent replication assessed.",
      "ai_role": "The models combined gene-deletion phenotypes with simulated metabolic fluxes to prioritize reactions whose perturbation would benefit or harm growth across carbon conditions; selected out-of-training-set predictions were experimentally tested.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10912204/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-04-deep-learning-guided-mining-of-nrps-scaffolds-yields-active-naturally-inspired-antimicrobial-peptides",
    "title": "Deep-learning-guided mining of NRPS scaffolds yields active naturally inspired antimicrobial peptides",
    "summary": "The team mined 216,408 bacterial genomes, identifying 335,024 NRPS clusters and 328 dereplicated peptide scaffolds. Deep-learning scoring selected five synthesized peptides; P2 showed MIC50 values of 1–2 μM against Vibrio parahaemolyticus and V. alginolyticus. Deep-learning-guided sequence optimization produced P2.2, which showed increased antibacterial activity, membrane effects and reduced hemolysis in assays.",
    "field": "medicine",
    "ai_technology": "Deep-learning peptide activity scoring and sequence-optimization algorithms",
    "organization": "Zhejiang University of Technology; and collaborating institutions",
    "source_url": "https://homepage.zjut.edu.cn/_upload/article/files/51/c3/6b78dffc40f7932148ad26cbf65b/e1d2421e-c600-4dac-82a5-bc29fa545929.pdf",
    "source_type": "paper",
    "publication_date": "2025-04-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.bioorg.2025.108444",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Deep-learning-guided sequence optimization produced P2.2, which showed increased antibacterial activity, membrane effects and reduced hemolysis in assays.",
      "unknown": "The study calls the peptides naturally inspired; it did not establish that the synthesized sequences are the native products of the mined clusters. P2's MIC50 of 1–2 μM did not mean 80% inhibition: its MIC80 exceeded 128 μM under the reported 24-hour conditions. P2.2's measured MIC80 values were 32 μM for V. parahaemolyticus and 64 μM for V. alginolyticus. These are in-vitro findings, not clinical efficacy.",
      "ai_role": "AI-ranked peptide candidates inspired by nonribosomal peptide synthetase cluster scaffolds from bacterial genomes and guided amino-acid optimization of P2 into P2.2; the selected sequences were synthesized and tested in vitro.",
      "evidence_urls": [
        "https://homepage.zjut.edu.cn/_upload/article/files/51/c3/6b78dffc40f7932148ad26cbf65b/e1d2421e-c600-4dac-82a5-bc29fa545929.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-kppepgen-generates-experimentally-active-pathogen-specific-antimicrobial-peptides",
    "title": "KPPepGen generates experimentally active pathogen-specific antimicrobial peptides",
    "summary": "KPPepGen combines Gene Ontology and pathogen-knowledge-graph prompts with a diffusion model to generate pathogen-conditioned peptide sequences. The paper reports valid peptide generation for 56 pathogens and wet-lab evaluation of nine novel peptides for Escherichia coli and Staphylococcus aureus, reporting antimicrobial activity and low cytotoxicity. Its Magainin 2 optimization result is a separate model-comparison metric, not a clinical outcome.",
    "field": "medicine",
    "ai_technology": "KPPepGen knowledge-aware prompt diffusion model",
    "organization": "Huazhong Agricultural University",
    "source_url": "https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.202507457",
    "source_type": "paper",
    "publication_date": "2025-09-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202507457",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports valid peptide generation for 56 pathogens and wet-lab evaluation of nine novel peptides for Escherichia coli and Staphylococcus aureus, reporting antimicrobial activity and low cytotoxicity.",
      "unknown": "The source reports nine experimentally evaluated peptides for E. coli and S. aureus, not experimental validation for all 56 pathogens. Antimicrobial activity and cytotoxicity are in-vitro outcomes; therapeutic efficacy, safety in animals or people, and resistance outcomes are not established by this report.",
      "ai_role": "The model used pathogen-specific knowledge prompts to generate and optimize antimicrobial peptide sequences; researchers synthesized candidates and evaluated the nine reported novel peptides in wet-lab antimicrobial and cytotoxicity assays.",
      "evidence_urls": [
        "https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.202507457"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-02-latent-diffusion-designs-antimicrobial-peptides-active-against-fungi-and-drug-resistant-bacteria",
    "title": "Latent diffusion designs antimicrobial peptides active against fungi and drug-resistant bacteria",
    "summary": "A latent-diffusion and molecular-dynamics pipeline generated and filtered peptide sequences; 40 synthesized candidates were tested, and 25 met the study's MIC threshold for antibacterial or antifungal activity. AMP-29 showed activity against Candida glabrata and efficacy in a murine skin-infection model, while AMP-24 showed in-vitro Gram-negative activity and efficacy in murine Acinetobacter baumannii skin and lung infection models.",
    "field": "medicine",
    "ai_technology": "Latent diffusion model coupled to molecular dynamics and peptide filtering/ranking",
    "organization": "Shandong University; and collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11797553/",
    "source_type": "paper",
    "publication_date": "2025-02-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.adp7171",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A latent-diffusion and molecular-dynamics pipeline generated and filtered peptide sequences; 40 synthesized candidates were tested, and 25 met the study's MIC threshold for antibacterial or antifungal activity.",
      "unknown": "Twenty-five of 40 tested peptides met the authors' stated MIC ≤200 μM activity threshold, which includes a broad range of potency. The murine efficacy reports concern selected peptides and infection models, not human treatment. Some fungal peptides showed hemolysis; favorable safety cannot be generalized to the whole set.",
      "ai_role": "The diffusion model generated a diverse peptide sequence set, and computational filtering selected 40 candidates for chemical synthesis and experimental antimicrobial testing; the selected peptides included leads taken into mouse infection models.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11797553/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-10-caiman-validates-ai-prescreening-of-typical-and-atypical-large-bowel-biopsies-across-uk-and-portuguese-cohorts",
    "title": "CAIMAN validates AI prescreening of typical and atypical large-bowel biopsies across UK and Portuguese cohorts",
    "summary": "CAIMAN is a weakly supervised deep-learning model trained on pathologist-labeled whole-slide images to classify typical, atypical non-neoplastic and atypical neoplastic bowel biopsies. In three external cohorts comprising 1,536 slides from the UK and Portugal, it detected atypical biopsies with AUROC values of 0.9431, 0.9576 and 0.9636. At 99% sensitivity, it outperformed the prior IDaRS pipeline on two external cohorts, but not the third.",
    "field": "medicine",
    "ai_technology": "CAIMAN (Colorectal AI Model for Abnormality Detection), a weakly supervised deep-learning whole-slide image classifier",
    "organization": "University of Warwick; University Hospitals Coventry and Warwickshire NHS Foundation Trust, collaborating UK hospitals and IMP Diagnostics (Portugal)",
    "source_url": "https://wrap.warwick.ac.uk/id/eprint/180883/",
    "source_type": "paper",
    "publication_date": "2023-10-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/S2589-7500(23)00148-6",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In three external cohorts comprising 1,536 slides from the UK and Portugal, it detected atypical biopsies with AUROC values of 0.9431, 0.9576 and 0.9636.",
      "unknown": "Retrospective validation on archived biopsies does not establish clinical deployment or improved patient outcomes. CAIMAN did not outperform IDaRS at 99% sensitivity in one of the three external validation cohorts; the authors call for larger multicentre validation and a clinical trial. The reported 99% sensitivity is paired with cohort-specific specificity, not a universal performance guarantee.",
      "ai_role": "AI learned biopsy-level histological patterns from slide-level diagnostic labels, then classified whole-slide images into typical, atypical non-neoplastic or atypical neoplastic classes. The trained frozen model was evaluated on three external cohorts; the reported finding is a specific retrospective clinical diagnostic-validation result.",
      "evidence_urls": [
        "https://wrap.warwick.ac.uk/id/eprint/180883/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-10-multi-centre-validation-tests-ai-empowered-oncoseek-across-15-122-participants-and-four-assay-platforms",
    "title": "Multi-centre validation tests AI-empowered OncoSeek across 15,122 participants and four assay platforms",
    "summary": "A 2025 study added four cohorts to evaluate the seven-protein-marker OncoSeek test, which combines blood protein tumor markers and clinical variables with AI. Across seven cohorts from seven centres in China, Brazil and the United States, totaling 15,122 participants, the reported overall sensitivity was 58.4% at 92.0% specificity; the symptomatic cohort had 73.1% sensitivity at 90.6% specificity. The study also tested performance across four platforms and two sample types.",
    "field": "medicine",
    "ai_technology": "OncoSeek AI-empowered blood test combining seven protein tumor markers with individual clinical data",
    "organization": "SeekIn; Henan Cancer Hospital, Peking University Shenzhen Hospital, First Saúde, Yonsei University and collaborating clinical centres",
    "source_url": "https://www.nature.com/articles/s41698-025-01105-2.pdf",
    "source_type": "paper",
    "publication_date": "2025-10-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41698-025-01105-2",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A 2025 study added four cohorts to evaluate the seven-protein-marker OncoSeek test, which combines blood protein tumor markers and clinical variables with AI.",
      "unknown": "This is a retrospective/observational cohort validation of a blood test, not evidence that screening improves outcomes or reduces mortality. Aggregate sensitivity was 58.4% and performance varied by cancer type; several cancer-specific groups were small. The article adds validation to a previously published OncoSeek test, so this record is limited to the new expanded-cohort/platform result, not a claim of first test discovery.",
      "ai_role": "The AI-assisted test integrated measured protein-marker values and clinical variables for multicancer classification. This paper evaluated its performance in four additional validation cohorts, including a prospective blinded study and symptomatic case-control cohort, and aggregated results across seven cohorts.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41698-025-01105-2.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-hidden-finds-malignant-like-cell-signals-in-early-multiple-myeloma-precursor-samples",
    "title": "HiDDEN finds malignant-like cell signals in early multiple-myeloma precursor samples",
    "summary": "HiDDEN refines sample-level case labels at single-cell resolution. In human bone-marrow single-cell RNA-seq data, its labels tracked manual neoplastic-cell annotations in mixed precursor samples and inferred malignant-like cells in three low-purity MGUS samples that the original manual analysis had labeled as containing only healthy cells. The inferred cells were enriched for previously described malignancy-associated gene signatures. In a separate mouse demyelination dataset, HiDDEN identified an early endothelial subpopulation with expression changes associated with angiogenesis, blood–brain-barrier breakdown and extracellular-matrix responses.",
    "field": "biology",
    "ai_technology": "HiDDEN, a machine-learning framework for refining disease/condition labels at single-cell resolution",
    "organization": "Broad Institute of MIT and Harvard; collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41467-024-53666-8.pdf",
    "source_type": "paper",
    "publication_date": "2024-11-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-53666-8",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "HiDDEN refines sample-level case labels at single-cell resolution.",
      "unknown": "The MGUS finding is a transcriptomic inference in previously collected samples, computationally supported by known gene signatures and sample-purity estimates; the paper does not report independent pathology confirmation, clinical follow-up, or validation as a diagnostic test. The endothelial subpopulation was found in a mouse lesion model; its association with blood–brain-barrier dysfunction does not establish causality.",
      "ai_role": "The model inferred which individual cells in case samples were affected rather than assigning the sample-level case label to all cells. This enabled the paper’s reanalysis of human myeloma precursor single-cell data and mouse demyelination single-nucleus data, yielding the specified malignancy and endothelial-population findings.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-53666-8.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-09-milton-augmented-disease-cohorts-surface-putative-new-genedisease-associations-in-uk-biobank-data",
    "title": "MILTON-augmented disease cohorts surface putative new gene–disease associations in UK Biobank data",
    "summary": "MILTON uses machine learning on routine clinical biomarkers, quantitative traits and plasma proteomics to predict disease phenotypes and augment case labels in UK Biobank. In a phenome-wide analysis of 484,230 genome-sequenced participants, the authors report 182 putative novel gene–disease associations that were not genome-wide significant in baseline cohorts or explained by biomarker associations; 38.9% of the 167 mappable signals were nominally supported with the same direction in FinnGen.",
    "field": "biology",
    "ai_technology": "MILTON (Machine Learning with Phenotype Associations), an ensemble machine-learning framework for disease prediction from clinical and molecular biomarkers",
    "organization": "AstraZeneca Centre for Genomics Research; and collaborating UK Biobank/FinnGen research institutions",
    "source_url": "https://www.nature.com/articles/s41588-024-01898-1.pdf",
    "source_type": "paper",
    "publication_date": "2024-09-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41588-024-01898-1",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a phenome-wide analysis of 484,230 genome-sequenced participants, the authors report 182 putative novel gene–disease associations that were not genome-wide significant in baseline cohorts or explained by biomarker associations; 38.9% of the 167 mappable signals were nominally supported with the same direction in FinnGen.",
      "unknown": "The authors classify the 182 associations as putative novel; they are computational associations, not experimentally established causal genes. FinnGen nominal same-direction support applied to 38.92% (65/167) of the mappable putative novel PheWAS associations, while the paper also reports other aggregate exome-wide results. A separate example, APOB–thyrotoxicosis, showed FinnGen support. These signals require further biological validation.",
      "ai_role": "MILTON predicted cases from clinical biomarkers and other quantitative traits, augmenting phenotype labels for genetic association analyses. The AI-enabled cohort enrichment yielded putative novel gene–disease signals; the study compared them with baseline analyses, orthogonal prioritization tools and FinnGen.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41588-024-01898-1.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-scimilarity-links-fibrosis-associated-macrophage-states-across-diseases-and-validates-a-matching-in-vitro-cell-state",
    "title": "SCimilarity links fibrosis-associated macrophage states across diseases and validates a matching in-vitro cell state",
    "summary": "SCimilarity searched a 23.4-million-cell reference atlas spanning 412 single-cell RNA-sequencing studies using query profiles from interstitial-lung-disease macrophages and fibroblasts. It identified fibrosis-associated macrophage-like states in other fibrotic diseases, COVID-19 and selected tumour datasets, and a 3D hydrogel culture as a high-ranking in-vitro match; the authors report experimentally demonstrating that this system reproduces the queried macrophage state.",
    "field": "biology",
    "ai_technology": "SCimilarity, a deep metric-learning model for querying cell-state similarity across single-cell atlases",
    "organization": "Genentech; and collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41586-024-08411-y",
    "source_type": "paper",
    "publication_date": "2024-11-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08411-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "It identified fibrosis-associated macrophage-like states in other fibrotic diseases, COVID-19 and selected tumour datasets, and a 3D hydrogel culture as a high-ranking in-vitro match; the authors report experimentally demonstrating that this system reproduces the queried macrophage state.",
      "unknown": "Most cross-disease and tumour-context links are computational similarity observations in previously published single-cell datasets; they do not establish causal roles in those diseases. Some rare tissue/cancer matches are from individual datasets or samples. The in-vitro 3D hydrogel assay is the experimentally tested part; broad disease associations need further biological validation.",
      "ai_role": "The model ranked cell profiles in a 23.4-million-cell atlas by similarity to ILD-derived macrophage/fibroblast query profiles, linking these states to disease/tissue contexts and nominating the 3D hydrogel culture for experimental follow-up.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-024-08411-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-machine-learning-analysis-identifies-a-plasma-protein-signature-associated-with-als-years-before-symptoms",
    "title": "Machine-learning analysis identifies a plasma protein signature associated with ALS years before symptoms",
    "summary": "In a plasma-proteomics study, machine learning selected a 17-protein panel from 33 proteins that differed between 183 people with ALS and 309 controls. The panel distinguished ALS from healthy participants and other neurological diseases in a held-out test set (AUC 96.2%, balanced accuracy 89.3%) and was assessed in independent cohorts, including UK Biobank samples. Applying the model to presymptomatic samples from people who later developed ALS showed risk-score shifts up to about 10 years before symptom onset.",
    "field": "medicine",
    "ai_technology": "Machine-learning classification model trained on Olink plasma proteomic features",
    "organization": "National Institute of Neurological Disorders and Stroke; and collaborating ALS registry and research institutions",
    "source_url": "https://www.nature.com/articles/s41591-025-03890-6",
    "source_type": "paper",
    "publication_date": "2025-08-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-025-03890-6",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The panel distinguished ALS from healthy participants and other neurological diseases in a held-out test set (AUC 96.2%, balanced accuracy 89.3%) and was assessed in independent cohorts, including UK Biobank samples.",
      "unknown": "This is a candidate biomarker result, not a clinically available diagnostic or screening test. The discovery and neurological-disease test cohorts are modest; the presymptomatic analysis is retrospective, and broader ancestry representation and prospective longitudinal validation remain needed. The paper reports multiple cohort-specific metrics; the summary uses the detailed held-out test-set AUC and balanced accuracy rather than the abstract’s 98.3% model figure.",
      "ai_role": "The model selected a multiplex protein signature and classified ALS versus healthy and neurological-disease controls; the researchers then applied its risk score to presymptomatic samples to estimate disease-associated changes before onset.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41591-025-03890-6"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-01-a-seven-gene-machine-learning-score-shows-an-alzheimer-associated-signal-in-an-independent-plasma-cohort",
    "title": "A seven-gene machine-learning score shows an Alzheimer-associated signal in an independent plasma cohort",
    "summary": "A machine-learning pipeline selected a seven-gene lactylation-related transcriptomic score (LDS) from nine Alzheimer disease cohorts. The study evaluated the signature in seven independent brain transcriptomic datasets and a single-center clinical plasma cohort of 540 participants (180 controls, 90 with amnestic mild cognitive impairment and 270 with Alzheimer disease). Plasma LDS had AUC 0.772 for Alzheimer disease; combining LDS with p-tau181 and p-tau217 had AUC 0.859. LDS scores increased across control, aMCI and Alzheimer groups and correlated with MMSE and Braak stage.",
    "field": "medicine",
    "ai_technology": "Lactylation-Derived Score (LDS), a random-forest feature-screening and partial least-squares logistic-regression model",
    "organization": "Southeast University Zhongda Hospital; and collaborating clinical and research institutions",
    "source_url": "https://www.nature.com/articles/s41746-025-02305-7",
    "source_type": "paper",
    "publication_date": "2026-01-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41746-025-02305-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning pipeline selected a seven-gene lactylation-related transcriptomic score (LDS) from nine Alzheimer disease cohorts.",
      "unknown": "The plasma validation cohort was recruited at one center and the analyses are retrospective and observational. The score does not establish lactylation as a causal mechanism, clinical benefit, or readiness for routine diagnosis; broader and prospective validation is needed. This is a transcriptomic signature and a different result from previously reported Alzheimer spectral-protein/chemical biomarker studies.",
      "ai_role": "Machine learning selected a seven-gene expression signature from lactylation-related genes across multi-cohort brain transcriptomic data; the resulting score was measured by RT-qPCR in a separate clinical plasma cohort and evaluated alone and with plasma p-tau markers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41746-025-02305-7"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-06-a-machine-learning-plasma-panel-classifies-some-rem-sleep-behavior-disorder-participants-years-before-motor-parkinson-diagnosis",
    "title": "A machine-learning plasma panel classifies some REM sleep behavior disorder participants years before motor Parkinson diagnosis",
    "summary": "Researchers developed a targeted multiplex mass-spectrometry assay for eight plasma proteins and applied machine learning to distinguish Parkinson disease from healthy controls and classify people with isolated REM sleep behavior disorder (iRBD). In the reported longitudinal iRBD cohorts, the model classified 79% of participants up to seven years before motor Parkinson disease or dementia with Lewy bodies onset. The work connects a targeted protein panel to a pre-motor disease signal in measured blood samples.",
    "field": "medicine",
    "ai_technology": "Machine-learning classifier using recursive feature elimination on targeted plasma proteomic measurements",
    "organization": "University of Oxford; University Medical Center Göttingen and collaborating European research institutions",
    "source_url": "https://www.nature.com/articles/s41467-024-48961-3",
    "source_type": "paper",
    "publication_date": "2024-06-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-48961-3",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers developed a targeted multiplex mass-spectrometry assay for eight plasma proteins and applied machine learning to distinguish Parkinson disease from healthy controls and classify people with isolated REM sleep behavior disorder (iRBD).",
      "unknown": "The discovery phase used only 10 Parkinson cases and 10 controls. Although the study includes larger targeted-assay cohorts, the longitudinal iRBD groups were small (18 and 54 participants), and the finding is not a validated screening test or proof that intervention changes outcomes. The model’s apparent early classification needs larger prospective cohorts.",
      "ai_role": "The classifier selected an eight-protein panel from targeted mass-spectrometry measurements, then classified motor Parkinson cases and iRBD participants; the longitudinal analysis linked panel scores with later clinical conversion.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-48961-3"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-01-machine-learning-selects-csf-and-serum-protein-panels-associated-with-early-alzheimer-disease",
    "title": "Machine learning selects CSF and serum protein panels associated with early Alzheimer disease",
    "summary": "In a proteomic study, researchers measured paired CSF and serum samples in a discovery cohort and used random-forest machine learning to select a 19-protein CSF panel and an eight-protein serum panel to distinguish mild cognitive impairment due to Alzheimer disease from cognitively normal participants. Targeted proteomics verified candidate proteins in an independent multicenter cohort. In small held-out test groups, the CSF panel reached AUC 0.984 and the serum panel AUC 0.881; the CSF panel also distinguished Alzheimer-related MCI from frontotemporal dementia, ALS and Huntington disease in the reported analyses.",
    "field": "medicine",
    "ai_technology": "Random-forest machine-learning classifiers on CSF and serum proteomic profiles",
    "organization": "Capital Medical University; Westlake University, Zhejiang University and collaborating Chinese clinical institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10794110/",
    "source_type": "paper",
    "publication_date": "2024-01-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.xinn.2023.100544",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a proteomic study, researchers measured paired CSF and serum samples in a discovery cohort and used random-forest machine learning to select a 19-protein CSF panel and an eight-protein serum panel to distinguish mild cognitive impairment due to Alzheimer disease from cognitively normal participants.",
      "unknown": "The AUC test sets were small (CSF: 9 Alzheimer-related MCI and 21 controls; serum: 8 and 21). The study used separate independent multicenter targeted-proteomics validation for candidate proteins; that does not make the small classifier test cohorts large or establish clinical utility. The serum panel was less effective at distinguishing Alzheimer-related MCI from other neurodegenerative diseases than the CSF panel. These are research panels, not routine tests.",
      "ai_role": "Random forests prioritized 19 CSF proteins and eight serum proteins to form disease-specific MCI classifiers after proteomic discovery and targeted validation; the selected panels were evaluated in held-out samples and against other neurodegenerative diseases.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10794110/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-lightgbm-links-a-three-protein-blood-panel-to-biopsy-staged-masld-fibrosis",
    "title": "LightGBM links a three-protein blood panel to biopsy-staged MASLD fibrosis",
    "summary": "A translational study combined a disease-related mouse gene-expression signature, human liver-biopsy transcriptomes and serum protein measurements to select IGFBP7, SSc5D and Sema4D as a three-protein panel for metabolic dysfunction-associated steatotic liver disease (MASLD). A LightGBM model classified biopsy-staged fibrosis, with AUCs 0.82 for F0/F1, 0.89 for F2 and 0.87 for F3/F4 in the development/testing data. In an independent Danish cohort of 156 patients, AUC was 0.84 for F0/F1 and F3/F4; performance for F2 was lower (AUC 0.62).",
    "field": "medicine",
    "ai_technology": "LightGBM classifier (TLM3) using the three serum proteins IGFBP7, SSc5D and Sema4D",
    "organization": "TNO; Amsterdam UMC, Leiden University Medical Center and collaborating Dutch and Danish clinical institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11137090/",
    "source_type": "paper",
    "publication_date": "2024-05-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-48956-0",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A translational study combined a disease-related mouse gene-expression signature, human liver-biopsy transcriptomes and serum protein measurements to select IGFBP7, SSc5D and Sema4D as a three-protein panel for metabolic dysfunction-associated steatotic liver disease (MASLD).",
      "unknown": "The model is an early translational biomarker result. Its independent cohort replicated better classification at F0/F1 and F3/F4 (AUC 0.84) but was modest for F2 (AUC 0.62). The authors say mechanistic claims could not be tested in the single-biopsy cohorts and call for prospective and intervention studies before clinical utility is established. TNO disclosed a patent application related to the biomarkers.",
      "ai_role": "The LightGBM model combined the three protein measurements selected from a translational mouse/human fibrosis signature to classify liver-biopsy fibrosis stages. The authors compared the model with Fib-4, APRI and FibroScan in held-out and independent cohorts.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11137090/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-12-a-machine-learning-selected-p4-serum-panel-detects-hepatocellular-carcinoma-and-signals-cirrhosis-to-cancer-progression",
    "title": "A machine-learning-selected P4 serum panel detects hepatocellular carcinoma and signals cirrhosis-to-cancer progression",
    "summary": "A staged serum-proteomics study across 1,002 individuals used random-forest machine learning to select a four-marker panel, P4 (HABP2, CD163, AFP and PIVKA-II). In an independent validation set, the panel distinguished hepatocellular carcinoma from cirrhosis (AUC 0.979; sensitivity 0.925; specificity 0.915) and from healthy controls (AUC 0.992). In a prospective external cohort of cirrhosis patients followed for HCC, it predicted conversion with AUC 0.890 and median lead time of 11.4 months before imaging diagnosis.",
    "field": "medicine",
    "ai_technology": "Random-forest machine-learning model for serum-protein panel selection and HCC classification",
    "organization": "Mengchao Hepatobiliary Hospital of Fujian Medical University; Fujian Cancer Hospital, Zhejiang University and collaborating Chinese clinical institutions",
    "source_url": "https://www.nature.com/articles/s41467-023-44255-2.pdf",
    "source_type": "paper",
    "publication_date": "2023-12-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-44255-2",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In an independent validation set, the panel distinguished hepatocellular carcinoma from cirrhosis (AUC 0.979; sensitivity 0.925; specificity 0.915) and from healthy controls (AUC 0.992). The prospectively collected cirrhosis cohort was split into 177 training and 76 validation participants for predictive-model construction; collection timing alone does not establish a prospective locked-model evaluation.",
      "unknown": "The prospective cohort included 253 cirrhosis patients, with 36 developing HCC; 8 of 65 participants were false positives for conversion by the reported endpoint. This is source-reported research performance, not proof of improved survival or a clinically implemented screening test. The lead time is measured relative to imaging diagnosis in the study cohorts.",
      "ai_role": "The random-forest model evaluated serum protein panels and selected the four-marker P4 panel; researchers then tested classification in an independent validation set and assessed prediction of cirrhosis-to-HCC conversion in prospective follow-up data.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-023-44255-2.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-02-mal-id-maps-disease-associated-immune-receptor-patterns-across-infections-and-autoimmune-conditions",
    "title": "Mal-ID maps disease-associated immune-receptor patterns across infections and autoimmune conditions",
    "summary": "Mal-ID analyzed B-cell and T-cell receptor repertoires from 593 people, including patients with COVID-19, HIV, systemic lupus erythematosus and type-1 diabetes, influenza-vaccination recipients and healthy controls. Its combined models classified held-out immune states and highlighted disease-associated receptor patterns; interpretation identified distinct autoreactivity features in lupus and type-1 diabetes and antigen-associated receptor sequences in COVID-19. The authors also evaluated classification on independent COVID-19 cohorts.",
    "field": "medicine",
    "ai_technology": "Mal-ID (MAchine Learning for Immunological Diagnosis), an ensemble using repertoire feature models and ESM-2 protein-language-model embeddings",
    "organization": "Stanford University; Oklahoma Medical Research Foundation, Duke University, University of Pennsylvania and collaborating institutions",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39977494/",
    "source_type": "paper",
    "publication_date": "2025-02-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adp2407",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors also evaluated classification on independent COVID-19 cohorts.",
      "unknown": "This is a patient-sample diagnostic-validation and immune-biology study, not evidence of clinical deployment or benefit. Validation cohorts and disease-specific sample sizes are limited; external COVID-19 classification involved small cohorts and thresholds were tuned in some analyses. Additional disease-specific validation, threshold calibration, sequencing-depth assessment and interpretation alongside standard clinical assessment remain necessary. Its 2024 bioRxiv preprint is the same work, not a separate finding.",
      "ai_role": "Three machine-learning representations of B-cell and T-cell receptor data were combined to classify immune states and prioritize disease-associated receptor sequences, gene segments and isotypes for biological interpretation.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39977494/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-05-active-learning-identifies-a-higher-activity-streptavidin-variant-for-gold-catalyzed-hydroamination",
    "title": "Active learning identifies a higher-activity streptavidin variant for gold-catalyzed hydroamination",
    "summary": "An iterative active-learning campaign engineered a streptavidin artificial metalloenzyme for the known-to-the-authors gold-catalyzed hydroamination of 2-ethynylaniline to indole. Gaussian-process models guided screening across five residues in a 3.2-million-variant sequence space. The final-round library contained 38% variants more active than the prior reference, compared with 3% in the initial library. The leading variant, Sav CTNLV, showed 18-fold higher cell-specific activity than wild-type Sav and threefold higher activity than the previous reference variant in replicate testing.",
    "field": "biology",
    "ai_technology": "Active-learning Gaussian-process regression model with optimized sequence descriptors and experimental-noise estimates",
    "organization": "ETH Zurich; University of Basel and University of Regensburg",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11273458/",
    "source_type": "paper",
    "publication_date": "2024-05-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acscentsci.4c00258",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The leading variant, Sav CTNLV, showed 18-fold higher cell-specific activity than wild-type Sav and threefold higher activity than the previous reference variant in replicate testing.",
      "unknown": "The hydroamination reaction and streptavidin–gold artificial metalloenzyme were established in earlier work by the same group; this record is specifically the later AI-guided identification of a higher-activity five-residue variant, not discovery of a new reaction. Activity was measured in an E. coli periplasmic whole-cell assay; purified variants did not retain the same rank order in vitro. Industrial utility or catalytic scale-up was not demonstrated.",
      "ai_role": "The model used measured sequence-activity data from thousands of variants to select exploratory and exploitation libraries across five residues, then prioritized variants for experimental testing. It enabled identification of a high-activity sequence in a large combinatorial space; the paper reports considering experimental uncertainty in its rankings.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11273458/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-modify-guided-library-screening-identifies-a-cytochrome-c-variant-active-in-both-carbonboron-and-carbonsilicon-bond-formation",
    "title": "MODIFY-guided library screening identifies a cytochrome c variant active in both carbon–boron and carbon–silicon bond formation",
    "summary": "Researchers applied MODIFY, an ensemble model combining protein language models and sequence-density models, to design a diverse high-fitness library for Rhodothermus marinus cytochrome c. Experimental screening identified variant MLYPPT, which catalyzed both non-natural borylation and silylation: reported total turnover numbers were 2,880 and 2,740, with enantiomeric ratios of 94:6 and 97:3, respectively. The same library yielded distinct high-performing variants for each reaction.",
    "field": "biology",
    "ai_technology": "MODIFY (ML-optimized library design with improved fitness and diversity)",
    "organization": "Georgia Institute of Technology; University of Pittsburgh; University of California, Santa Barbara",
    "source_url": "https://par.nsf.gov/servlets/purl/10529357",
    "source_type": "paper",
    "publication_date": "2024-07-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-50698-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimental screening identified variant MLYPPT, which catalyzed both non-natural borylation and silylation: reported total turnover numbers were 2,880 and 2,740, with enantiomeric ratios of 94:6 and 97:3, respectively.",
      "unknown": "The C–B and C–Si reactions had previously been demonstrated with separately evolved cytochrome c variants; this finding is the discovery of additional, structurally distinct variants and a generalist MLYPPT catalyst. The reported yields and enantiomeric ratios are from the study's assays; industrial scale-up was not shown.",
      "ai_role": "MODIFY used evolutionary sequence models and protein language models for zero-shot fitness estimates, then co-optimized predicted fitness and residue-level diversity to propose a 1,000-variant library. Researchers synthesized and assayed sampled library members; the model-guided library contained experimentally measured new cytochrome c variants, including MLYPPT with activity in both reactions.",
      "evidence_urls": [
        "https://par.nsf.gov/servlets/purl/10529357"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-active-learning-identifies-a-protoglobin-variant-with-99-yield-and-cis-selective-cyclopropanation",
    "title": "Active learning identifies a protoglobin variant with 99% yield and cis-selective cyclopropanation",
    "summary": "An iterative machine-learning campaign engineered the protoglobin from Pyrobaculum arsenaticum (ParPgb) for non-native cyclopropanation of 4-vinylanisole. After two model-guided rounds following an initial library screen, variant MPFDY gave 99% total cyclopropane yield and a 14:1 cis:trans ratio in the reported assay, compared with a parent variant that produced about 40% yield and favored trans product. The result demonstrates a specific sequence combination with improved yield and stereoselectivity in this enzyme scaffold.",
    "field": "biology",
    "ai_technology": "ALDE (Active Learning-assisted Directed Evolution), using a bootstrapped deep-neural-network ensemble and Thompson sampling",
    "organization": "California Institute of Technology; University of California, Berkeley; Elegen Corp",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11739421/",
    "source_type": "paper",
    "publication_date": "2025-01-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-55987-8",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "After two model-guided rounds following an initial library screen, variant MPFDY gave 99% total cyclopropane yield and a 14:1 cis:trans ratio in the reported assay, compared with a parent variant that produced about 40% yield and favored trans product.",
      "unknown": "Cyclopropanation is an established non-native enzyme reaction class; the study's distinct result is the ALDE-guided ParPgb sequence and its measured product yield/selectivity. The 99% yield and 14:1 ratio are study assay results, not evidence of industrial process readiness. The study is a single enzyme-engineering campaign.",
      "ai_role": "ALDE trained on measured sequence–yield data, quantified model uncertainty, and selected variants across two iterative rounds in a five-residue, 3.2-million-sequence design space. Experimental screening measured product yield and stereoselectivity; the model-guided variants included MPFDY, whose combined mutations improved the targeted cis-product objective where tested single mutations and their naive recombinations did not.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11739421/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-an-ml-generated-mine-homolog-restores-bacterial-cell-division-patterning-in-vivo",
    "title": "An ML-generated MinE homolog restores bacterial cell division patterning in vivo",
    "summary": "A machine-learning-generated MinE homolog, synMinEv25, restored Min oscillations and wild-type-like cell growth and morphology when substituted for the native minE gene in an E. coli strain lacking minDE. Researchers generated variants with a multiple-sequence-alignment variational autoencoder, used AlphaFold2-based structural scoring to select candidates, and screened them in synthetic-cell mimics before in-vivo testing. The validated variant is a distant homolog with less than 50% sequence identity to E. coli MinE.",
    "field": "biology",
    "ai_technology": "Multiple-sequence-alignment variational autoencoder (MSA-VAE), AlphaFold2 Multimer, and a structure-based divide-and-conquer scoring pipeline",
    "organization": "Max Planck Institute of Biochemistry",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10914801/",
    "source_type": "paper",
    "publication_date": "2024-03-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-46203-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers generated variants with a multiple-sequence-alignment variational autoencoder, used AlphaFold2-based structural scoring to select candidates, and screened them in synthetic-cell mimics before in-vivo testing.",
      "unknown": "This is a proof-of-principle result in one engineered E. coli MinDE system, not a general demonstration that generated proteins can replace natural genes. The paper notes weak correlation between several in-silico scores and individual measured protein properties; it reports one best-performing variant and does not establish applications outside this assay.",
      "ai_role": "The MSA-VAE generated MinE homolog sequences; structure-based scoring estimated membrane binding, MinD complex formation and dimerization to prioritize 48 candidates from 4,000 generated sequences. Cell-free synthetic-cell assays identified 14 variants with Min oscillations, and researchers tested these in E. coli; synMinEv25 restored oscillations, growth and morphology. AI generated and prioritized candidates, while wet-lab experiments established function.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10914801/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-08-closed-loop-learning-links-donoracceptor-molecules-photostability-to-higher-triplet-states",
    "title": "Closed-loop learning links donor–acceptor molecules’ photostability to higher triplet states",
    "summary": "A closed-loop synthesis and learning campaign found that solution photostability across a donor–bridge–acceptor molecular series correlated with the high-energy portion of the triplet excited-state manifold, beyond the commonly considered lowest triplet state. Researchers tested the model-derived hypothesis on an experimental set and changed solvent triplet energy to test the proposed Dexter triplet-energy-transfer degradation pathway. The study used Bayesian optimization and interpretable supervised learning to extract the chemical hypothesis from measured photostability data.",
    "field": "chemistry",
    "ai_technology": "Closed-loop transfer (CLT), combining Gryffin Bayesian optimization with physics-feature supervised learning",
    "organization": "University of Illinois Urbana-Champaign; University of Toronto; Vector Institute; University of Chicago; University of California, Berkeley",
    "source_url": "https://schroeder.cbe.princeton.edu/sites/g/files/toruqf7951/files/documents/Nature-2024.pdf",
    "source_type": "paper",
    "publication_date": "2024-08-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07892-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers tested the model-derived hypothesis on an experimental set and changed solvent triplet energy to test the proposed Dexter triplet-energy-transfer degradation pathway.",
      "unknown": "The finding concerns solution photostability in the studied donor–bridge–acceptor chemical space; it does not by itself establish a universal law for all organic molecules or performance in processed films/devices. The work used human expertise in the loop and experimentally tested the model-generated hypothesis. Device deployment was not demonstrated.",
      "ai_role": "Bayesian optimization recommended molecules for automated modular synthesis and photostability assays; an interpretable supervised model using physics-based features identified a triplet-state hypothesis, which researchers tested with experimental test sets and solvent-energy perturbation. AI guided experimental selection and surfaced a specific chemical relationship; the follow-up measurements tested it.",
      "evidence_urls": [
        "https://schroeder.cbe.princeton.edu/sites/g/files/toruqf7951/files/documents/Nature-2024.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-03-machine-learning-search-of-archived-spectra-reveals-vinylnhc-coupling-products-in-palladium-catalysis",
    "title": "Machine-learning search of archived spectra reveals vinyl–NHC coupling products in palladium catalysis",
    "summary": "MEDUSA Search mined archived high-resolution mass spectra and uncovered previously unreported vinyl–N-heterocyclic-carbene (NHC) coupling products in Pd/NHC-catalyzed Mizoroki–Heck reactions. Researchers retrieved the corresponding reaction records and independently tested the proposed transformation with multiple palladium–NHC complexes; ultrahigh-resolution mass spectrometry and MS/MS supported the product formulas and structures. The work also found an ethyl–NHC product across tested reaction mixtures, suggesting additional catalyst-transformation pathways.",
    "field": "chemistry",
    "ai_technology": "MEDUSA Search, an isotope-distribution-driven HRMS search engine with machine-learning ion-threshold regression and false-positive classification",
    "organization": "Zelinsky Institute of Organic Chemistry, Russian Academy of Sciences; Skolkovo Institute of Science and Technology",
    "source_url": "https://www.nature.com/articles/s41467-025-56905-8.pdf",
    "source_type": "paper",
    "publication_date": "2025-03-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-56905-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "MEDUSA Search mined archived high-resolution mass spectra and uncovered previously unreported vinyl–N-heterocyclic-carbene (NHC) coupling products in Pd/NHC-catalyzed Mizoroki–Heck reactions.",
      "unknown": "The claimed novelty is specific to vinyl–NHC coupling products in the studied Pd/NHC catalytic context. HRMS formulas alone cannot distinguish structural isomers, so the paper's MS/MS, ultrahigh-resolution measurements, reaction notebooks and independent experiments are essential evidence. The authors state that most other detected ions could not be associated confidently with reaction conditions because of incomplete metadata; those are not counted here.",
      "ai_role": "Researchers generated candidate product-ion formulas from reaction fragments and catalysts; MEDUSA Search screened archived spectra using isotope-pattern matching, with two ML models estimating ion-presence thresholds and filtering likely false positives. It surfaced previously overlooked ions in old spectra; follow-up experiments with different NHC ligands and MS/MS tests checked the proposed vinyl–NHC products.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-56905-8.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-deep-learning-on-routine-pathology-slides-prescreens-fgfr-altered-bladder-cancers-in-a-global-study",
    "title": "Deep learning on routine pathology slides prescreens FGFR-altered bladder cancers in a global study",
    "summary": "A deep-learning device analyzed routine H&E whole-slide images to estimate whether advanced urothelial tumors carried selected FGFR2/FGFR3 alterations used for erdafitinib eligibility. In a separate 350-slide retrospective validation set, the device achieved 88.7% sensitivity and 31.8% specificity; it was also prospectively integrated into a non-interventional study spanning 89 sites in nine countries. The paper projected a 28.7% reduction in molecular tests assuming 15% biomarker prevalence; that estimate was simulated, not a measured population-level reduction.",
    "field": "medicine",
    "ai_technology": "FGFR Device, a multi-instance-learning attention model with a convolutional neural-network backbone",
    "organization": "Janssen Research & Development, LLC (Johnson & Johnson)",
    "source_url": "https://www.nature.com/articles/s41467-024-49153-9.pdf",
    "source_type": "paper",
    "publication_date": "2024-06-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-49153-9",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a separate 350-slide retrospective validation set, the device achieved 88.7% sensitivity and 31.8% specificity; it was also prospectively integrated into a non-interventional study spanning 89 sites in nine countries.",
      "unknown": "The 350-slide retrospective validation set was deliberately enriched (150 FGFR-positive and 200 negative; 43% prevalence), and the device fell slightly short of the prespecified 90% sensitivity target. Prospective real-time workflow validation covered only 17 samples in one month (15 predictions; two quality-control failures); production deployment reported 24 samples, with only 20 receiving molecular testing. The 28.7% test reduction is a simulation assuming 15% prevalence, not an observed clinical outcome. This is prescreening support, not a stand-alone diagnosis or evidence of patient benefit.",
      "ai_role": "Trained on more than 3,000 urothelial-cancer slides with molecular test labels, the deep-learning model inferred selected FGFR alteration status from H&E images and returned a likelihood for prescreening before confirmatory molecular testing. The device was prospectively integrated into a non-interventional study workflow; molecular tests remained the reference standard.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-49153-9.pdf"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2024-10-machine-guided-synthetic-regulatory-dna-drives-cell-type-specific-expression-across-cell-lines-and-vertebrates",
    "title": "Machine-guided synthetic regulatory DNA drives cell-type-specific expression across cell lines and vertebrates",
    "summary": "The CODA platform used the Malinois deep-learning model and sequence-search algorithms to design synthetic cis-regulatory elements (CREs). In massively parallel reporter assays, 94.1% of the synthetic sequences met the authors’ cell-type-specificity criterion, versus 73.6% of model-selected natural sequences and 40.6% of chromatin-mark-selected natural sequences; synthetic sequences also had higher measured target/off-target separation across K562, HepG2 and SK-N-SH cells. In transgenic zebrafish, two of three liver-directed and two of three neuron-directed CREs showed the intended activity. One neuron-directed sequence, synN1, retained cortical and neuronal expression in mice, including postnatal neocortical layer 6 and subplate neurons.",
    "field": "biology",
    "ai_technology": "CODA (Computational Optimization of DNA Activity) with Malinois deep neural network and AdaLead, simulated annealing, and Fast SeqProp sequence search/design algorithms",
    "organization": "Broad Institute of MIT and Harvard; The Jackson Laboratory; Yale University",
    "source_url": "https://www.nature.com/articles/s41586-024-08070-z",
    "source_type": "paper",
    "publication_date": "2024-10-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-08070-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In massively parallel reporter assays, 94.1% of the synthetic sequences met the authors’ cell-type-specificity criterion, versus 73.6% of model-selected natural sequences and 40.6% of chromatin-mark-selected natural sequences; synthetic sequences also had higher measured target/off-target separation across K562, HepG2 and SK-N-SH cells.",
      "unknown": "The in-vitro work used three transformed human cell lines and reporter assays. In-vivo testing covered a small selected set: two of three liver-directed and two of three neuron-directed elements were active in zebrafish; mouse validation focused on neuronal synN1, with n=3 synN1 mice for postnatal cell-type analysis and n=6 for the 5-week cortical reporter assay. These are transgenic reporter results, not therapeutic delivery or evidence of clinical utility. Earlier synthetic enhancer studies existed; this paper’s counted result is its CODA/Malinois comparison against natural sequences plus selected vertebrate validation, not the general idea of AI-designed enhancers.",
      "ai_role": "Malinois predicted MPRA activity of candidate DNA sequences in K562, HepG2 and SK-N-SH cells; CODA iteratively optimized sequences for specified cell-type specificity. Researchers synthesized and experimentally measured the designs by MPRA, then tested selected CREs with reporter constructs in zebrafish and mice. The paper attributes the novel sequences and their measured expression specificity to this model-guided design loop.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-024-08070-z"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-ai-derived-histology-biomarker-stratifies-first-line-chemotherapy-outcomes-in-advanced-pancreatic-cancer",
    "title": "AI-derived histology biomarker stratifies first-line chemotherapy outcomes in advanced pancreatic cancer",
    "summary": "The CHAI computational-histology platform generated a locked H&E-based biomarker intended to distinguish advanced pancreatic ductal adenocarcinoma patients with differential outcomes on fluoropyrimidine-based versus gemcitabine-based chemotherapy. In an independent validation cohort of 299 patients from the prospectively collected COMPASS and Know Your Tumor cohorts, the 173 patients classified F-pref had longer time to next treatment or death (8.6 vs 7.5 months, P=.035) and overall survival (14.4 vs 11.7 months, P=.003) on F- versus G-based therapy. Among 126 G-pref patients, G-based therapy was associated with longer time to next treatment or death (9.6 vs 7.2 months, P=.038), but not significantly different overall survival.",
    "field": "medicine",
    "ai_technology": "Computational Histology Artificial Intelligence (CHAI) platform",
    "organization": "Cedars-Sinai Medical Center; Valar Labs; Johns Hopkins University; University of Pittsburgh Medical Center; Pancreatic Cancer Action Network; University Health Network / Princess Margaret Cancer Centre",
    "source_url": "https://ascopubs.org/doi/pdfdirect/10.1200/JCO-25-02199",
    "source_type": "paper",
    "publication_date": "2026-02-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1200/JCO-25-02199",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In an independent validation cohort of 299 patients from the prospectively collected COMPASS and Know Your Tumor cohorts, the 173 patients classified F-pref had longer time to next treatment or death (8.6 vs 7.5 months, P=.035) and overall survival (14.4 vs 11.7 months, P=.003) on F- versus G-based therapy. The locked biomarker was applied to assembled data from prospectively collected cohorts; prospective validation of treatment selection is still needed.",
      "unknown": "Validation used 299 patients in two prospectively collected but nonrandomized cohorts; treatment assignment was not randomized, and the authors also used propensity weighting and target-trial emulation. The primary endpoint was time to next treatment or death, not progression-free survival. In the biomarker’s G-pref subgroup, the OS comparison was not significant. The authors call for further prospective validation before clinical adoption. An earlier CHAI study concerned adjuvant gemcitabine outcomes in resected disease; this entry is limited to the distinct locked first-line fluoropyrimidine-versus-gemcitabine biomarker and external-cohort results in advanced disease.",
      "ai_role": "Deep-learning models segmented H&E biopsy slides and quantified more than 30,000 histomorphologic features. Using the development cohort, the investigators statistically selected and weighted features into a treatment-preference biomarker and locked its threshold before applying it to an independent validation cohort. The platform supplied the image-derived feature measurements used in the biomarker; this is a biomarker-development and validation result, not autonomous treatment assignment.",
      "evidence_urls": [
        "https://ascopubs.org/doi/pdfdirect/10.1200/JCO-25-02199"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-plantrna-fm-identifies-translation-associated-rna-motifs-and-structure-motifs-in-plants",
    "title": "PlantRNA-FM identifies translation-associated RNA motifs and structure motifs in plants",
    "summary": "The PlantRNA-FM model, pretrained on sequence and predicted RNA structure from 1,124 plant transcriptomes, identified 5′-UTR motifs associated with translation efficiency across plant transcriptomes. Reporter assays in plants tested selected predictions: mutating a high-translation-associated RNA secondary-structure motif reduced measured translation efficiency by up to 5.3-fold, while disrupting a model-prioritized low-translation-associated RNA G-quadruplex motif increased it by up to 5.8-fold. The paper reports the rG4 motifs as translation-repressive, consistent with prior evidence on individual rG4s; the transcriptome-wide model prioritization and tested instances are the scoped findings here.",
    "field": "biology",
    "ai_technology": "PlantRNA-FM, an interpretable RNA foundation model integrating RNA sequence and predicted structure",
    "organization": "Northeast Normal University; Chinese Academy of Sciences; University of Exeter; John Innes Centre; Alan Turing Institute",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11652376/",
    "source_type": "paper",
    "publication_date": "2024-12-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42256-024-00946-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Reporter assays in plants tested selected predictions: mutating a high-translation-associated RNA secondary-structure motif reduced measured translation efficiency by up to 5.3-fold, while disrupting a model-prioritized low-translation-associated RNA G-quadruplex motif increased it by up to 5.8-fold.",
      "unknown": "Experimental reporter validation was limited to selected motifs and had n=3, as reported in the figure caption. The model’s transcriptome-wide associations are not individually causal validations. The repressive role of rG4 motifs was previously known for individual cases; the paper’s contribution includes model-prioritized plant 5′-UTR motifs and experimental effects for selected constructs, not first discovery of rG4 biology. Reporter assays do not establish whole-plant phenotype or crop benefit.",
      "ai_role": "PlantRNA-FM was pretrained with plant RNA sequence and structure data, then fine-tuned on translation-efficiency data. An attention-contrast interpretation framework prioritized 5′-UTR sequence/structure motifs associated with translation; researchers tested selected motif predictions by mutating reporter constructs and measuring translation efficiency in plants.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11652376/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-deep-learning-mines-venom-proteomes-for-antimicrobial-peptides-active-in-vitro-and-in-a-mouse-infection-model",
    "title": "Deep learning mines venom proteomes for antimicrobial peptides active in vitro and in a mouse infection model",
    "summary": "The APEX deep-learning model screened 16,123 venom proteins by generating and scoring 40,626,260 venom-encrypted peptides. After activity prediction and filtering for low similarity to known antimicrobial peptides, the authors selected 58 for experiments; 53 showed antimicrobial activity against at least one tested pathogen. Three selected peptides reduced bacterial counts after topical treatment in a mouse Acinetobacter baumannii skin-abscess model; the wolf-spider-derived Arachnoserver-5 peptide produced a reported three-log reduction versus untreated controls through day four, without significant body-weight change under the study conditions.",
    "field": "medicine",
    "ai_technology": "APEX deep-learning model for bacterial strain-specific minimum inhibitory concentration prediction",
    "organization": "University of Pennsylvania",
    "source_url": "https://www.nature.com/articles/s41467-025-60051-6.pdf",
    "source_type": "paper",
    "publication_date": "2025-07-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-60051-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "After activity prediction and filtering for low similarity to known antimicrobial peptides, the authors selected 58 for experiments; 53 showed antimicrobial activity against at least one tested pathogen.",
      "unknown": "Only 58 of the 386 low-similarity candidates were experimentally tested, and activity was measured against a defined panel of 11 strains; some tested peptides were cytotoxic. The mouse study was a topical single-dose skin-abscess model with six animals per group, not systemic treatment or human efficacy. The paper notes possible ion-channel activity for many candidates and calls for follow-up safety testing. A related APEX study mined archaeal proteomes (10.1038/s41564-025-02061-0); this entry is limited to the separate venom-protein source library and its distinct peptide set and experiments.",
      "ai_role": "APEX scored 40.6 million candidate peptide fragments generated from venom proteins for predicted antimicrobial activity. The authors filtered predicted hits for sequence dissimilarity to known antimicrobial peptides, synthesized a selected subset, and experimentally tested antimicrobial activity, cytotoxicity and selected animal-model efficacy. The AI model narrowed an otherwise very large peptide search space; assays established which selected candidates were active.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-60051-6.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-02-chance-uses-tumor-genomes-to-prioritize-non-cancer-drugs-with-activity-in-patient-derived-cancer-cells",
    "title": "CHANCE uses tumor genomes to prioritize non-cancer drugs with activity in patient-derived cancer cells",
    "summary": "CHANCE, a supervised machine-learning model integrating patient-specific coding and non-coding mutations with drug and protein-interaction information, screened 5,197 TCGA/ICGC tumors against 923 FDA-approved non-oncology drugs. For ex-vivo validation, the investigators sequenced tumor cells from seven patients with pancreatic or esophageal cancer and tested CHANCE-prioritized drugs; at least one predicted non-oncology drug showed anticancer activity in tumor cells from five of the seven patients. This is patient-specific cell-assay validation, not a treatment-response study in patients.",
    "field": "medicine",
    "ai_technology": "CHANCE, a supervised gradient-boosted-tree model integrating coding/non-coding variants, protein–protein interaction networks, drug targets and drug-response data",
    "organization": "Tianjin Medical University; collaborating Chinese cancer research and clinical institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11794852/",
    "source_type": "paper",
    "publication_date": "2025-02-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41698-025-00813-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "For ex-vivo validation, the investigators sequenced tumor cells from seven patients with pancreatic or esophageal cancer and tested CHANCE-prioritized drugs; at least one predicted non-oncology drug showed anticancer activity in tumor cells from five of the seven patients.",
      "unknown": "Only seven patient-derived samples were tested, from pancreatic or esophageal cancers; at least one prioritized non-oncology drug showed activity in five. This was an ex-vivo drug-sensitivity assay, not clinical treatment, and the study does not establish efficacy, safety, dosing or patient benefit. The 5,197-tumor/923-drug screen is computational prediction; only the small patient-derived subset had experimental validation. Drug-response predictions may include drugs represented in training, as the article notes 149 of 923 screened drugs were in the training data.",
      "ai_role": "Trained on drug-sensitivity measurements from 318 cancer cell lines across 647 drugs, CHANCE maps each patient tumor’s whole-genome mutations to drug-target/pathway features and ranks non-oncology drugs by predicted patient-specific anticancer activity. Investigators used these predictions to select drugs for ex-vivo testing in tumor cells from seven patients; the reported five-patient activity result tests the model’s personalized prioritization.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11794852/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-alphacd-maps-experimentally-measured-catalytic-properties-across-cytidine-deaminases-and-supports-a-high-fidelity-editor",
    "title": "AlphaCD maps experimentally measured catalytic properties across cytidine deaminases and supports a high-fidelity editor",
    "summary": "The researchers measured catalytic efficiency, editing windows, motif preferences and off-target activity for 1,100 APOBEC-like cytidine deaminases fused to nCas9 in HEK293T cells, then trained AlphaCD to predict these properties. They applied it to 21,335 UniProt deaminases and experimentally checked a 28-protein subsample. As an application, the team used AlphaCD to identify a high-efficiency deaminase and alanine scanning to produce the E100A variant, which retained editing activity while reducing off-target activity; cell experiments tested editing at endogenous loci and a PCSK9 splice site.",
    "field": "biology",
    "ai_technology": "AlphaCD, a machine-learning model trained on sequence, predicted structure and related protein features",
    "organization": "Agricultural Genomics Institute at Shenzhen, Chinese Academy of Agricultural Sciences",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40825882/",
    "source_type": "paper",
    "publication_date": "2025-08-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41422-025-01164-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The researchers measured catalytic efficiency, editing windows, motif preferences and off-target activity for 1,100 APOBEC-like cytidine deaminases fused to nCas9 in HEK293T cells, then trained AlphaCD to predict these properties.",
      "unknown": "The large functional dataset was measured in HEK293T reporter assays. The 21,335-protein landscape is model-predicted, with a 28-protein validation subsample. The engineered E100A editor was tested in cells; these experiments do not establish therapeutic safety, in-vivo efficacy or clinical benefit. This is distinct from the earlier AlphaFold2 structural-clustering papers: the named discovery and assay here concern the 1,100-enzyme dataset, AlphaCD predictions and A0A2R2Z4E4 E100A variant, not CD0208^P52A or the 2023 DddA-like clade findings.",
      "ai_role": "AlphaCD learned experimentally measured activity, editing-window, motif and off-target properties across 1,100 deaminases, predicted properties for 21,335 database sequences, and was applied to select a high-efficiency candidate for subsequent alanine-scanning engineering. The authors report the model's predictions validated on 28 sampled deaminases; the E100A mutation itself was experimentally produced by alanine scanning, not generated by the model.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40825882/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-deep-learning-design-produced-soluble-analogues-of-membrane-protein-folds-with-preserved-binding-functions",
    "title": "Deep-learning design produced soluble analogues of membrane-protein folds with preserved binding functions",
    "summary": "A deep-learning pipeline based on inversion of AlphaFold2 and ProteinMPNN designed soluble proteins with claudin, rhomboid-protease and GPCR-like folds. Experiments reported stable, folded soluble designs and structural agreement with target folds. The researchers grafted functional motifs, including a ghrelin-receptor epitope that bound its antibody and active/inactive A2A-receptor-like designs that differed in mini-Gs binding; a binding-site knockout served as a control.",
    "field": "biology",
    "ai_technology": "AF2seq deep-learning backbone design with ProteinMPNN sequence design",
    "organization": "École Polytechnique Fédérale de Lausanne; University at Buffalo; University of Washington; Massachusetts Institute of Technology",
    "source_url": "https://www.nature.com/articles/s41586-024-07601-y",
    "source_type": "paper",
    "publication_date": "2024-06-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-024-07601-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experiments reported stable, folded soluble designs and structural agreement with target folds.",
      "unknown": "This is a protein-design proof of concept with in-vitro biophysical and binding assays; the analogues are not native membrane receptors and do not demonstrate signaling, drug activity or therapeutic efficacy. The GPCR analogues tested binding to mini-Gs in conformation-designed constructs. The separate live-catalog record on designed proteins modulating dopamine D1 receptor signaling reports a different binder/target and signaling result; this paper reports soluble receptor-fold analogues and binding assays, not D1 signaling modulation.",
      "ai_role": "The deep-learning design pipeline generated soluble analogues of complex membrane-protein folds and the scaffolds into which native functional motifs were grafted. The study tested solubility, thermal stability, structural accuracy and selected binding functions experimentally, including conformation-specific A2A-receptor-like constructs.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-024-07601-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-09-mimal-predicts-and-experimentally-validates-five-yeast-gene-functions-from-proteinmetabolite-profiles",
    "title": "MIMaL predicts and experimentally validates five yeast gene functions from protein–metabolite profiles",
    "summary": "MIMaL used tree-based regression and SHAP interpretation on yeast proteome/metabolome measurements, then clustered protein-control profiles across single-gene knockouts. It predicted functions for five genes, with experimental validation reported for all five. Two previously uncharacterized genes, YJR120W and YDL157C, were predicted to modulate mitochondrial translation; functions were also predicted and validated for SDH9, ISC1 and FMP52. The study additionally validated previously unrecognized protein links to citrate.",
    "field": "biology",
    "ai_technology": "MIMaL: ExtraTrees regression of metabolite abundance from proteomic profiles, SHAP interpretation, and UMAP/OPTICS clustering of protein-control profiles",
    "organization": "Cedars-Sinai Medical Center; Medical College of Wisconsin; Stockholm University; University of Gothenburg",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9801967/",
    "source_type": "paper",
    "publication_date": "2022-09-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/bioinformatics/btac631",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "MIMaL used tree-based regression and SHAP interpretation on yeast proteome/metabolome measurements, then clustered protein-control profiles across single-gene knockouts.",
      "unknown": "This was a yeast study using a prior 174-single-gene-knockout multi-omics dataset, with 348 condition profiles; it does not establish corresponding human gene functions. Five gene functions are reported validated, and the article specifically states YJR120W and YDL157C modulate mitochondrial translation. Some experimental validation used existing knockout strains and measurements; no independent replication is implied.",
      "ai_role": "A tree-based model predicted metabolite profiles from proteomic measurements; SHAP-derived protein-control values and clustering across knockout conditions were used to infer new protein–metabolite connections and gene functions. The researchers experimentally validated predicted citrate-control relationships and five gene-function predictions.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9801967/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-06-simmer-guided-experiments-identify-gut-bacterial-strains-that-deplete-methotrexate",
    "title": "SIMMER-guided experiments identify gut bacterial strains that deplete methotrexate",
    "summary": "SIMMER combines reaction-chemistry and protein-sequence similarity to predict gut microbial species and enzymes responsible for drug biotransformations. For methotrexate (MTX), the model predicted candidate strains and enzyme sequences; researchers tested 42 gut bacterial isolates by HPLC. SIMMER predicted 13 isolates as MTX-metabolizing, and 10 of those tested isolates showed MTX depletion; the association was significant in the study (Fisher exact p=0.046). Predicted MTX-hydrolysis enzyme signatures were also enriched in stool samples of MTX nonresponders in an analyzed rheumatoid-arthritis cohort.",
    "field": "biology",
    "ai_technology": "SIMMER, a computational reaction-similarity and protein-homology pipeline using chemical fingerprints, MetaCyc reactions and gut metagenomic sequences",
    "organization": "Gladstone Institutes; University of California, San Francisco; Chan Zuckerberg Biohub San Francisco",
    "source_url": "https://elifesciences.org/articles/82401",
    "source_type": "paper",
    "publication_date": "2023-06-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.7554/eLife.82401",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "For methotrexate (MTX), the model predicted candidate strains and enzyme sequences; researchers tested 42 gut bacterial isolates by HPLC.",
      "unknown": "The study reports species-level evidence for bacterial MTX depletion, but the enzyme predictions themselves were not individually purified or functionally validated. Only 13 of 42 tested isolates were predicted positive and 10 depleted MTX; predictions may have false positives, as the authors note. The RA association is observational and does not establish that microbial metabolism caused treatment nonresponse. The underlying fact that gut bacteria can metabolize MTX was already known; this result concerns species-level prediction and validation, not discovery of MTX metabolism itself.",
      "ai_role": "SIMMER proposed specific bacterial species and enzyme sequences for known drug transformations. The team tested its MTX predictions against an independent panel of 42 gut bacterial isolates and HPLC-measured drug depletion, and compared predicted enzyme presence with baseline stool metagenomes and later MTX response in an RA dataset.",
      "evidence_urls": [
        "https://elifesciences.org/articles/82401"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-04-alphafold-structure-analysis-helps-reveal-phgdh-transcriptional-activity-linked-to-alzheimers-amyloid-pathology",
    "title": "AlphaFold structure analysis helps reveal PHGDH transcriptional activity linked to Alzheimer’s amyloid pathology",
    "summary": "AlphaFold-based structural analysis highlighted a helix-helix-turn-helix subdomain in PHGDH that resembles a TALE homeodomain, prompting tests of a previously unrecognized DNA-binding and transcriptional function. Experiments supported PHGDH binding to promoters and activation of IKKα and HMGB1 in astrocytes, independently of its metabolic enzyme activity. In mouse models and human brain organoids, PHGDH expression promoted amyloid pathology; a blood-brain-barrier-permeable inhibitor of its transcriptional function reduced amyloid pathology and improved measured behavioral deficits in mice.",
    "field": "biology",
    "ai_technology": "AlphaFold structure prediction, followed by structural comparison of the predicted PHGDH fold",
    "organization": "University of California San Diego; collaborating biomedical research institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12204802/",
    "source_type": "paper",
    "publication_date": "2025-04-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cell.2025.03.045",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "AlphaFold-based structural analysis highlighted a helix-helix-turn-helix subdomain in PHGDH that resembles a TALE homeodomain, prompting tests of a previously unrecognized DNA-binding and transcriptional function.",
      "unknown": "The AlphaFold-derived structural resemblance generated the mechanistic hypothesis; experiments establish the reported PHGDH transcriptional and pathology effects in the tested models. The inhibitor and behavioral effects were tested in mice, not people. Human brain organoid findings are in vitro and do not establish clinical disease causation or treatment efficacy. PHGDH overexpression alone did not reproduce all Alzheimer’s disease features, and no human therapeutic outcome is reported.",
      "ai_role": "The researchers inspected AlphaFold-predicted PHGDH structure and identified a helix-helix-turn-helix subdomain resembling a TALE homeodomain, which motivated the hypothesis and targeted mutation tests for DNA binding. AlphaFold did not establish PHGDH function or disease causality; the transcriptional activity and pathology effects were tested through biochemical, cellular, organoid and mouse experiments.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12204802/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-07-itargex-analysis-identifies-elg1-as-a-regulator-of-yeast-transcription-buffering-and-four-protein-turnover-regulators",
    "title": "iTARGEX analysis identifies ELG1 as a regulator of yeast transcription buffering and four protein-turnover regulators",
    "summary": "iTARGEX applied a mixture-regression model to gene-expression profiles from 1,484 single-gene deletion yeast mutants, associating transcriptomic changes with S-phase transcriptional buffering and global protein turnover. Experiments supported ELG1 as a previously unrecognized regulator of transcriptional buffering during S phase; the phenotype was reproduced in a second genetic background and rescued by a PCNA-D150E mutation. Follow-up protein-turnover measurements showed altered turnover in MRN1, UPF3, RAM1 and SSN3 deletion strains.",
    "field": "biology",
    "ai_technology": "iTARGEX, a finite-mixture regression framework fitted by expectation maximization to single-gene-deletion transcriptomic data",
    "organization": "Academia Sinica, Taiwan",
    "source_url": "https://academic.oup.com/nar/article/49/13/7318/6312755",
    "source_type": "paper",
    "publication_date": "2021-07-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/nar/gkab555",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experiments supported ELG1 as a previously unrecognized regulator of transcriptional buffering during S phase; the phenotype was reproduced in a second genetic background and rescued by a PCNA-D150E mutation.",
      "unknown": "The study used budding yeast and pre-existing deletion-mutant transcriptome profiles, followed by experiments on selected candidates. ELG1's role in S-phase transcription buffering was reproduced in a second yeast genetic background and rescued by PCNA-D150E; the protein-turnover effects concern yeast deletion strains. These results do not establish corresponding human functions.",
      "ai_role": "iTARGEX associated gene-expression profiles from deletion strains with biological traits without requiring prior gene annotations and nominated candidate regulators. The researchers then performed S-phase transcription and protein-turnover experiments to test the model-prioritized candidates, supporting ELG1's role in transcriptional buffering and altered turnover for four deletion strains.",
      "evidence_urls": [
        "https://academic.oup.com/nar/article/49/13/7318/6312755"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-camil-regression-predicts-continuous-homologous-recombination-deficiency-and-tumor-immune-biomarkers-from-histology",
    "title": "CAMIL regression predicts continuous homologous-recombination deficiency and tumor immune biomarkers from histology",
    "summary": "A self-supervised, attention-based deep-learning regression model predicted continuous homologous-recombination deficiency (HRD) scores and tumor-microenvironment biomarkers from H&E pathology slides. On TCGA internal validation, HRD classification AUROC exceeded 0.70 in five of seven cancer types; external CPTAC validation reached AUROC 0.81 for lung adenocarcinoma and 0.96 for endometrial cancer. In 2,297 colorectal-cancer patients in the DACHS cohort, continuous regression scores for leukocyte fraction and lymphocyte-infiltration-related signatures were independently prognostic in multivariable survival analysis.",
    "field": "medicine",
    "ai_technology": "CAMIL regression, a self-supervised contrastive-clustering and attention-based multiple-instance deep-learning model",
    "organization": "Technische Universität Dresden; University Hospital Dresden; collaborating clinical and pathology institutions",
    "source_url": "https://www.nature.com/articles/s41467-024-45589-1.pdf",
    "source_type": "paper",
    "publication_date": "2024-02-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-45589-1",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "On TCGA internal validation, HRD classification AUROC exceeded 0.70 in five of seven cancer types; external CPTAC validation reached AUROC 0.81 for lung adenocarcinoma and 0.96 for endometrial cancer.",
      "unknown": "Retrospective computational pathology analysis; the model predicts molecular and immune-related biomarkers from slides and does not replace confirmatory molecular testing or establish treatment selection benefit. HRD performance varied by cancer type, with weak internal performance in some cohorts. The reported survival associations in DACHS are prognostic associations, not evidence that using model predictions improves patient outcomes. Scores and endpoints were tested in the named datasets.",
      "ai_role": "The model analyzed whole-slide H&E images to estimate continuous HRD and tumor-microenvironment biomarker scores. The team compared regression with classification methods across cohorts, externally validated HRD predictions in CPTAC, and tested prognostic associations of predicted scores in a separate retrospective colorectal-cancer cohort.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-45589-1.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-04-an-integrated-arabidopsis-regulatory-network-predicts-13-previously-unlinked-reactive-oxygen-stress-regulators",
    "title": "An integrated Arabidopsis regulatory network predicts 13 previously unlinked reactive-oxygen-stress regulators",
    "summary": "Researchers integrated DNA-motif, chromatin-accessibility, TF-binding and expression evidence with supervised learning to build an Arabidopsis gene-regulatory network of 1,491 transcription factors and 1.7 million TF–target interactions. The network predicted candidate reactive-oxygen-species (ROS) regulators; loss- and gain-of-function phenotyping supported 13 regulators not previously linked to ROS or stress responses. Newly tested examples included WRKY55, WRKY47, AT2G33710 and AT5G57150. The paper also reports that 75% of its ROS-predicted TFs had ROS or physiological stress functions supported by either this study or prior evidence.",
    "field": "biology",
    "ai_technology": "Supervised-learning integrated gene-regulatory network (iGRN) built from motif, open-chromatin, transcription-factor-binding and expression-based networks",
    "organization": "Ghent University; VIB Center for Plant Systems Biology; Bioinformatics Institute Ghent",
    "source_url": "https://www.nature.com/articles/s41477-021-00894-1.pdf",
    "source_type": "paper",
    "publication_date": "2021-04-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41477-021-00894-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Newly tested examples included WRKY55, WRKY47, AT2G33710 and AT5G57150.",
      "unknown": "The iGRN integrated existing experimental data and the ROS findings were tested in Arabidopsis stress assays, not crop-yield or field-resilience trials. The 75% figure combines prior literature with new phenotypic tests; it is not a rate of independently replicated new discoveries. The new TF functions are limited to tested oxidative-stress conditions and mutant lines.",
      "ai_role": "The supervised integration model inferred TF–target regulatory links across multiple experimental input networks and predicted candidate TF functions, including ROS-response regulators. The investigators tested predicted TF candidates using Arabidopsis loss- or gain-of-function lines and ROS-specific phenotyping; these experiments supported 13 TFs without prior ROS/stress associations.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41477-021-00894-1.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-04-longitudinal-ehr-machine-learning-identifies-patients-who-may-need-autoimmune-rheumatic-disease-evaluation-years-earlier",
    "title": "Longitudinal EHR machine learning identifies patients who may need autoimmune rheumatic-disease evaluation years earlier",
    "summary": "A random-forest model trained on longitudinal electronic health records from 161,584 people across two institutions identified records associated with high-specificity autoantibody testing for systemic autoimmune rheumatic diseases (SARDs). The model achieved AUROC 0.93 in internal validation and 0.87 on the external All of Us cohort. In a separate BioMe cohort, model probabilities tracked SARD and other autoimmune diagnoses; records could predict later autoantibody testing and rheumatology encounters up to five years before those events. The study also found higher autoantibody yield among model-prioritized individuals than under the comparison clinical-testing rate.",
    "field": "medicine",
    "ai_technology": "Random-forest machine-learning model using longitudinal electronic-health-record diagnoses, medications, laboratory results and vital signs",
    "organization": "Icahn School of Medicine at Mount Sinai; University of Maryland School of Medicine; Hospital for Special Surgery",
    "source_url": "https://www.nature.com/articles/s41467-023-37996-7.pdf",
    "source_type": "paper",
    "publication_date": "2023-04-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-37996-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A random-forest model trained on longitudinal electronic health records from 161,584 people across two institutions identified records associated with high-specificity autoantibody testing for systemic autoimmune rheumatic diseases (SARDs).",
      "unknown": "The model predicts clinical suspicion and need for rheumatological testing, not a SARD diagnosis. The study was retrospective and opportunistic; autoantibody testing and rheumatology encounters are imperfect proxies for clinical need, and some diseases were not included in the training target. The authors state prospective studies are needed to determine whether model-guided testing improves care or outcomes. The five-year lead is retrospective prediction of later evaluation events, not proven earlier diagnosis or patient benefit.",
      "ai_role": "The model ranked patients by probability of receiving SARD-specific autoantibody testing, using clinical data preceding the test or rheumatology encounter. Investigators evaluated performance in internal validation, the external All of Us cohort and an independent BioMe cohort, and tested whether higher scores tracked diagnosed autoimmune disease and identified later testing or encounters earlier in retrospective records.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-023-37996-7.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-07-deep-learning-detects-elevated-tissue-derived-cfdna-across-cancer-and-cirrhosis-cohorts",
    "title": "Deep learning detects elevated tissue-derived cfDNA across cancer and cirrhosis cohorts",
    "summary": "A study built a fragment-level methylation atlas from 521 noncancer tissue samples spanning 29 human tissue types and used it with cfSort, a supervised deep neural network, to estimate tissue composition in plasma cell-free DNA. Applying cfSort to patient data, the investigators found higher cfDNA fractions from the affected tissue in liver, lung, colorectal and stomach cancer and cirrhosis cohorts than in healthy controls; the estimated fractions also increased with cancer stage. The tissue-origin signal was reproduced on an independent whole-genome-bisulfite-sequencing dataset for liver cancer. In serial samples from four non-small-cell lung cancer patients receiving anti-PD-1 therapy, estimated liver or kidney tissue fractions tracked measured biochemical markers of organ injury.",
    "field": "medicine",
    "ai_technology": "cfSort, an ensemble of two dense deep neural networks trained on simulated mixtures using a fragment-level tissue methylation atlas",
    "organization": "University of California, Los Angeles; Stanford University; EarlyDiagnostics; University of Southern California; University of California, San Diego",
    "source_url": "https://par.nsf.gov/servlets/purl/10428542",
    "source_type": "paper",
    "publication_date": "2023-07-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2305236120",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Applying cfSort to patient data, the investigators found higher cfDNA fractions from the affected tissue in liver, lung, colorectal and stomach cancer and cirrhosis cohorts than in healthy controls; the estimated fractions also increased with cancer stage.",
      "unknown": "The disease cohorts were retrospective and the clinical analyses demonstrate association and measurement feasibility, not prospective diagnostic benefit or improved outcomes. The clinical groups were 21 cirrhosis patients and 201 cancer patients, with 100 healthy controls; cross-platform liver-cancer validation used 24 cases and 32 controls. Anti-PD-1 injury monitoring involved only four NSCLC patients at three timepoints. The authors note that disease-specific methylation markers could improve cancer detection beyond tissue-of-origin fractions.",
      "ai_role": "The team used the supervised cfSort model to deconvolve methylation profiles in plasma cfDNA and estimate contributions from 29 tissue types. This enabled the reported comparison of affected-tissue cfDNA fractions in cancer and cirrhosis patient cohorts versus controls and longitudinal tracking of tissue-injury signals during immunotherapy. The atlas and model were also evaluated against simulated mixtures and independent methylation data.",
      "evidence_urls": [
        "https://par.nsf.gov/servlets/purl/10428542"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-ehr-deep-learning-and-sequencing-identify-two-previously-undiagnosed-noonan-syndrome-cases",
    "title": "EHR deep learning and sequencing identify two previously undiagnosed Noonan syndrome cases",
    "summary": "Investigators applied a deep-learning model to diagnosis text in the electronic health records of 92,428 pediatric patients. It flagged 171 patients above a prespecified Noonan syndrome risk threshold for chart review; genetic sequencing and specialist assessment identified two previously undiagnosed Noonan syndrome cases among evaluated patients, in addition to three patients diagnosed during the study period. The confirmed cases had pathogenic variants in Noonan-associated genes, including PTPN11 and LZTR1. This supplies real-world evidence that EHR-based model screening can surface rare-disease cases for clinical genetic evaluation.",
    "field": "medicine",
    "ai_technology": "Deep convolutional neural network trained on structured electronic-health-record diagnosis text to estimate Noonan syndrome risk",
    "organization": "Cincinnati Children's Hospital Medical Center; University of Cincinnati College of Medicine; University of Wisconsin–Madison",
    "source_url": "https://www.nature.com/articles/s41525-025-00512-5",
    "source_type": "paper",
    "publication_date": "2025-07-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41525-025-00512-5",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The confirmed cases had pathogenic variants in Noonan-associated genes, including PTPN11 and LZTR1.",
      "unknown": "This was a real-world retrospective screening evaluation at one pediatric biobank, not a prospective screening trial. Sequencing identified two cases among 83 patients evaluated, and the overall precision among the 171 high-risk patients was 2.92%; the model therefore required substantial chart review and genetic confirmation. Retrospective risk trajectories suggest possible earlier recognition but do not establish earlier care or improved outcomes.",
      "ai_role": "The previously developed deep-learning model scored 92,428 patient records and ranked 171 individuals above a prespecified risk threshold. The study team reviewed high-risk records, then used clinical genetic assessment and sequencing to evaluate patients without a prior diagnosis; these follow-up tests confirmed two previously undiagnosed cases. Model risk trajectories were also reconstructed from cumulative EHR diagnoses to assess whether the signal appeared before final diagnosis.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41525-025-00512-5"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-07-machine-learning-identifies-plasma-protein-signatures-that-predict-onset-of-common-and-rare-diseases",
    "title": "Machine learning identifies plasma protein signatures that predict onset of common and rare diseases",
    "summary": "Researchers used regularized machine-learning models to select sparse plasma-protein signatures for 10-year incidence across 218 diseases in 41,931 UK Biobank Pharma Proteomics Project participants. Five to twenty proteins improved held-out risk prediction over basic clinical information for 67 diseases and over clinical information plus routine assays for 52. The study identified TNFRSF17 as a specific predictor for future multiple myeloma and TNFRSF13B for monoclonal gammopathy of undetermined significance. In bone-marrow single-cell data from 11 newly diagnosed myeloma patients and three controls, four of five myeloma predictor proteins were most highly expressed by plasma cells; FCRLB and QPCT elevation was attributable to malignant plasma cells. Proteomic signatures also replicated improved prediction for five of six diseases tested in EPIC-Norfolk.",
    "field": "medicine",
    "ai_technology": "LASSO feature selection and regularized Cox machine-learning models integrating plasma proteomics and clinical covariates",
    "organization": "MRC Epidemiology Unit, University of Cambridge; Queen Mary University of London; University College London; GSK Research and Development; Berlin Institute of Health at Charité; EPIC-Norfolk",
    "source_url": "https://www.nature.com/articles/s41591-024-03142-z",
    "source_type": "paper",
    "publication_date": "2024-07-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-024-03142-z",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Five to twenty proteins improved held-out risk prediction over basic clinical information for 67 diseases and over clinical information plus routine assays for 52.",
      "unknown": "The disease predictions are observational risk associations, not causal mechanisms or clinical screening outcomes. Most protein signatures were not externally replicated; EPIC-Norfolk tested six diseases, with improved prediction replicated for five. The myeloma single-cell analysis involved 11 patients and three healthy controls and supports cellular expression attribution, not prospective validation of an assay. Absolute quantification with clinical-grade assays and further validation would be needed before clinical use.",
      "ai_role": "The authors applied a three-stage machine-learning framework—feature selection, hyperparameter tuning and held-out validation—to ~2,923 proteins in UK Biobank. It selected parsimonious protein signatures for 218 future disease outcomes and identified disease-specific predictors and shared predictors. For multiple myeloma, the model-selected five-protein signature and follow-up single-cell expression analysis linked four predictors to marrow plasma cells and showed malignant-cell-associated FCRLB and QPCT expression.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41591-024-03142-z"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-12-astroreg-maps-functional-enhancergene-links-in-human-astrocytes",
    "title": "AstroREG maps functional enhancer–gene links in human astrocytes",
    "summary": "A CRISPRi and single-cell RNA-sequencing screen tested 979 candidate enhancers in cultured human primary astrocytes and identified 158 functional enhancer–gene pairs, including enhancers regulating genes implicated in Alzheimer disease. The study trained the EGrf random-forest model on the measured screen and used it to predict additional enhancer–gene regulatory interactions genome-wide; selected model-predicted interactions were then functionally assessed by CRISPRi and qRT–PCR.",
    "field": "biology",
    "ai_technology": "EGrf random-forest enhancer–gene interaction model, trained on the AstroREG CRISPRi/single-cell RNA-sequencing screen",
    "organization": "University of New South Wales; University of Western Australia; Macquarie University; Monash University; University of Michigan",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41413662/",
    "source_type": "paper",
    "publication_date": "2025-12-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41593-025-02154-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A CRISPRi and single-cell RNA-sequencing screen tested 979 candidate enhancers in cultured human primary astrocytes and identified 158 functional enhancer–gene pairs, including enhancers regulating genes implicated in Alzheimer disease.",
      "unknown": "The primary CRISPRi screen and genome-wide model outputs are distinct evidence classes. The 158 reported enhancer–gene pairs came from the screen; genome-wide EGrf predictions are computational until tested. Figure descriptions report qRT–PCR follow-up for selected model predictions with independent duplicate transductions (n=4 per enhancer). The experiments used cultured astrocytes and do not establish enhancer causality in human Alzheimer disease or clinical relevance.",
      "ai_role": "The authors used experimentally measured enhancer–gene pairs to train EGrf, then applied the model to predict regulatory interactions genome-wide. Selected predictions were followed up by individual enhancer CRISPRi and qRT–PCR. Machine learning contributed the genome-wide extension of a functional astrocyte enhancer map; the initial 979-enhancer screen itself was experimentally measured.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41413662/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-04-single-cell-imaging-and-machine-learning-reveal-a-cdk2cop9-interaction-that-induces-a-senescence-like-immune-state-in-drosophila-cells",
    "title": "Single-cell imaging and machine learning reveal a Cdk2–COP9 interaction that induces a senescence-like immune state in Drosophila cells",
    "summary": "A synthetic genetic-interaction and cell-morphology map covering more than 6,800 Drosophila genes supported machine-learning-based gene-function assignments across 47 modules. In follow-up, Cytoclass single-cell classification revealed a state-specific interaction between Cdk2 and COP9 signalosome knockdown: co-depletion arrested cell-cycle progression and produced a senescence-like phenotype with increased antimicrobial-peptide expression in cultured cells and hemocytes, including without bacterial infection.",
    "field": "biology",
    "ai_technology": "Single-cell morphology classifier ensemble and machine-learning analysis of genetic-interaction profiles (Cytoclass; support-vector machines and random forest)",
    "organization": "German Cancer Research Center (DKFZ); Heidelberg University; University of Applied Sciences Bingen",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/37116498/",
    "source_type": "paper",
    "publication_date": "2023-04-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cels.2023.03.003",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A synthetic genetic-interaction and cell-morphology map covering more than 6,800 Drosophila genes supported machine-learning-based gene-function assignments across 47 modules.",
      "unknown": "The findings are from cultured Drosophila cells and hemocytes; they do not establish mammalian or human disease mechanisms. The senescence-like phenotype and antimicrobial-peptide induction were measured after experimental RNAi co-depletion. The global module assignments and many network interactions are computational inferences; only selected interactions received targeted follow-up.",
      "ai_role": "The authors generated single-cell morphology and synthetic genetic-interaction profiles, used machine-learning-based function discovery to assign functions to genes in 47 modules, and applied Cytoclass state classification to reveal interactions hidden in bulk-average profiles. The state-specific Cdk2–COP9 interaction was followed up by double-RNAi assays, single-cell transcriptomics, beta-galactosidase and antimicrobial-peptide measurements.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/37116498/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-deep-learning-of-salivary-gland-biopsies-identifies-acinar-cd8-t-cell-infiltration-associated-with-sjogren-disease",
    "title": "Deep learning of salivary-gland biopsies identifies acinar CD8 T-cell infiltration associated with Sjögren disease",
    "summary": "A deep-learning model trained on digitized minor salivary-gland biopsy slides from five expert centres and externally validated at a sixth classified focus score and Sjögren disease in a retrospective cohort of 545 participants. Explainable-model analysis identified a histological pattern of CD8-positive T cells around acinar epithelial cells associated with Sjögren disease. External-validation AUROC was 0.89 for disease classification and 0.92 among patients negative for anti-SSA antibodies.",
    "field": "medicine",
    "ai_technology": "Deep-learning histopathology classifier with Shapley-value interpretation",
    "organization": "Scienta Lab; William Harvey Research Institute, Queen Mary University of London; University of Birmingham; National and Kapodistrian University of Athens; Newcastle University; Université Paris-Saclay; participating NECESSITY consortium centres",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41038216/",
    "source_type": "paper",
    "publication_date": "2025-09-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/S2665-9913(25)00181-X",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A deep-learning model trained on digitized minor salivary-gland biopsy slides from five expert centres and externally validated at a sixth classified focus score and Sjögren disease in a retrospective cohort of 545 participants.",
      "unknown": "Retrospective multicentre cohort; external validation used one held-out expert centre. The authors state further validation is needed. The CD8 acinar-infiltration pattern was associated with disease classification and is not established as causal, pathognomonic, or clinically actionable. The study does not establish improved patient outcomes or replace expert pathology.",
      "ai_role": "The authors trained the deep-learning model on digitalized biopsy slides to classify focus score and Sjögren disease, tested generalization at a held-out centre, and used Shapley-value explanations to identify histological patterns associated with model decisions. The study reports acinar CD8-positive T-cell infiltration as a disease-associated pattern.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41038216/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-deep-learning-ct-response-score-stratifies-outcomes-in-lung-cancer-immunotherapy-cohorts",
    "title": "Deep-learning CT response score stratifies outcomes in lung-cancer immunotherapy cohorts",
    "summary": "A deep-learning radiomic pipeline generated a pretreatment CT response score (CTRS) for advanced non-small-cell lung cancer treated with PD-(L)1 checkpoint inhibitors. In an independent real-world test cohort, high-response-score first-line ICI-monotherapy patients had longer progression-free and overall survival after adjustment for baseline covariates including PD-L1 (adjusted HR 0.46 and 0.50). In a secondary analysis of 54 participants from a prospective sasanlimab trial, the score was associated with overall survival (adjusted HR 0.33) but not progression-free survival.",
    "field": "medicine",
    "ai_technology": "Three-dimensional hierarchical convolutional neural network feature extractor combined with a Cox survival model (CT response score, CTRS)",
    "organization": "Onc.AI; Pfizer; University of Pennsylvania; Moffitt Cancer Center",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11658027/",
    "source_type": "paper",
    "publication_date": "2024-12-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1200/CCI.24.00133",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In an independent real-world test cohort, high-response-score first-line ICI-monotherapy patients had longer progression-free and overall survival after adjustment for baseline covariates including PD-L1 (adjusted HR 0.46 and 0.50).",
      "unknown": "The real-world data were retrospective. Association of the score with survival among treated patients does not establish that the score predicts differential treatment benefit versus an untreated or alternative-treatment control. Clinical-trial analysis was secondary and small (n=54): PFS association was not significant, while OS association was significant. Authors say larger prospective studies are needed; no improved-care or deployment outcome is shown.",
      "ai_role": "The model learned image features from pretreatment CT/PET-CT scans and combined them with a Cox survival model to produce an imaging-only response score. Researchers evaluated the score in a held-out real-world cohort and in a secondary analysis of a prospective clinical-trial cohort; it stratified PFS/OS in the first and OS in the latter.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11658027/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-09-machine-learning-identifies-a-high-risk-diabetic-cardiomyopathy-phenotype-linked-to-future-heart-failure",
    "title": "Machine learning identifies a high-risk diabetic cardiomyopathy phenotype linked to future heart failure",
    "summary": "Unsupervised clustering of 25 echocardiographic and biomarker variables in 1,199 ARIC participants with diabetes identified three phenogroups; the group with the greatest cardiac remodelling and neurohormonal-stress abnormalities had 12.1% five-year heart-failure incidence, versus 4.6% and 3.1% in the other groups. A deep-neural-network classifier trained to recognize this phenotype identified corresponding high-risk groups in the external CHS cohort (n=802) and UT Southwestern EHR cohort (n=5,071); phenotype classification was associated with incident heart failure in both.",
    "field": "medicine",
    "ai_technology": "Unsupervised hierarchical clustering and a deep neural network (DeepNN) classifier, with random forest variable-importance analysis",
    "organization": "University of Texas Southwestern Medical Center; Texas Heart Institute; Houston Methodist DeBakey Heart & Vascular Center; Duke University; National Heart Centre Singapore; University of Toronto; University of Colorado; Massachusetts General Hospital",
    "source_url": "https://onlinelibrary.wiley.com/doi/10.1002/ejhf.3443",
    "source_type": "paper",
    "publication_date": "2024-09-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/ejhf.3443",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Unsupervised clustering of 25 echocardiographic and biomarker variables in 1,199 ARIC participants with diabetes identified three phenogroups; the group with the greatest cardiac remodelling and neurohormonal-stress abnormalities had 12.1% five-year heart-failure incidence, versus 4.6% and 3.1% in the other groups.",
      "unknown": "The cohorts were observational; association with future heart failure does not establish that model-guided intervention improves outcomes. The phenotype is an operational study definition, not a universally accepted diagnostic threshold. The authors describe potential preventive use, but it was not tested. The cohorts were external validations rather than prospective deployment trials.",
      "ai_role": "The authors used unsupervised hierarchical clustering of echocardiography and cardiac biomarkers to define outcome-associated diabetic-cardiomyopathy phenogroups, then trained and externally validated a DeepNN classifier to recognize the high-risk phenotype. Random forest permutation importance prioritized NT-proBNP, left-ventricular mass, Doppler measurements, left-atrial dimension and LV end-diastolic volume.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/10.1002/ejhf.3443"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-deep-learning-identifies-externally-replicated-fast-and-slow-cognitive-progressor-groups-in-early-alzheimer-dementia",
    "title": "Deep learning identifies externally replicated fast and slow cognitive-progressor groups in early Alzheimer dementia",
    "summary": "Deep learning clustered cognitive and functional trajectories in 283 people with early dementia probably due to Alzheimer disease into fast- and slow-progressing groups. The trajectory groups were externally validated and independently replicated in a 2,779-person dementia cohort. A classifier using cross-sectional data at diagnosis predicted progression group in external validation with AUC 0.70 ± 0.01.",
    "field": "medicine",
    "ai_technology": "Deep-learning clustering of longitudinal cognitive/functional trajectories, followed by machine-learning classification from cross-sectional diagnosis data",
    "organization": "Fraunhofer Institute for Algorithms and Scientific Computing (SCAI); University of Bonn; Massachusetts General Hospital and Harvard Medical School; Boehringer Ingelheim",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39713242/",
    "source_type": "paper",
    "publication_date": "2024-12-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/braincomms/fcae445",
    "relevance": {
      "impact": 3,
      "validation": 5,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The trajectory groups were externally validated and independently replicated in a 2,779-person dementia cohort.",
      "unknown": "These are prognostic progression groups, not established biological disease subtypes or treatment-response predictors. The classifier had modest external discrimination (AUC 0.70 ± 0.01). Trial-size reduction and cost savings were estimated by emulating hypothetical enrichment and were not demonstrated in an actual clinical trial. The study does not show improved patient outcomes.",
      "ai_role": "Deep learning grouped patients using multivariate disease trajectories; the authors then trained a classifier to predict those progression subgroups from measurements available at diagnosis. The model-supported fast/slow progression groups were externally validated and independently replicated in a separate dementia cohort.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39713242/"
      ],
      "test_type": "replication or operational use"
    }
  },
  {
    "id": "2026-01-deep-learning-histopathology-identifies-tertiary-lymphoid-structure-features-linked-to-stage-ii-colorectal-cancer-relapse",
    "title": "Deep-learning histopathology identifies tertiary lymphoid structure features linked to stage II colorectal cancer relapse",
    "summary": "SurvFinder analyzed 6,950 H&E slides from 1,604 stage II colorectal cancer patients across four cohorts and identified tertiary lymphoid structures (TLSs) as prognostic tissue features. Its multi-view model integrated TLS morphology and spatial location; explainable analysis highlighted TLS maturity and position near the tumor periphery as features associated with relapse-free survival and adjuvant-chemotherapy outcome patterns across the retrospective datasets.",
    "field": "medicine",
    "ai_technology": "SurvFinder, a multi-component deep-learning pipeline: WSINet for prognosis-associated tissue features, SegNet for TLS segmentation/subtyping, and MVNet for multi-view risk prediction",
    "organization": "Sun Yat-sen University Cancer Center; Sun Yat-sen University; Tianjin Medical University Cancer Institute and Hospital; Johns Hopkins University; Fudan University Shanghai Cancer Center",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41532433/",
    "source_type": "paper",
    "publication_date": "2026-01-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pmed.1004614",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "SurvFinder analyzed 6,950 H&E slides from 1,604 stage II colorectal cancer patients across four cohorts and identified tertiary lymphoid structures (TLSs) as prognostic tissue features.",
      "unknown": "This was a retrospective analysis of patients enrolled in 2012–2018; results do not establish prospective clinical utility. Associations between predicted risk, TLS features and adjuvant-chemotherapy outcomes are observational and do not prove treatment benefit or that TLS features causally determine relapse. The authors identify absence of prospective validation and real-world deployment as limitations.",
      "ai_role": "Deep learning autonomously identified prognosis-associated tissue features in whole-slide images, segmented and subtyped TLSs, and integrated morphological and spatial features for risk prediction. Explainable-AI attribution identified TLS location near the tumor margin and TLS maturity state as important prognostic features.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41532433/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-01-machine-learning-identifies-new-cyp2c9-inhibiting-drugs-and-drug-metabolites",
    "title": "Machine learning identifies new CYP2C9-inhibiting drugs and drug metabolites",
    "summary": "A structure-, dynamics- and physicochemical-feature machine-learning approach predicted CYP2C9 inhibition and experimentally identified four drugs—vatalanib, piriqualone, ticagrelor and cloperidone—as strong inhibitors, plus sertindole, asapiprant, duvelisib and dasatinib as moderate inhibitors in reported assays. Vatalanib had a reported CYP2C9 inhibition IC50 of 0.067 μM. Separate metabolism assays characterized CYP2C9-produced metabolites for abemaciclib, cloperidone, vatalanib and tarafenacin.",
    "field": "chemistry",
    "ai_technology": "Machine-learning inhibitor predictors combining CYP2C9 structural/dynamics information with inhibitor physicochemical properties",
    "organization": "INSERM; CNRS and Université de Paris; Conservatoire National des Arts et Métiers; Université de Lorraine; Nantes Université",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/35081108/",
    "source_type": "paper",
    "publication_date": "2022-01-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pcbi.1009820",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A structure-, dynamics- and physicochemical-feature machine-learning approach predicted CYP2C9 inhibition and experimentally identified four drugs—vatalanib, piriqualone, ticagrelor and cloperidone—as strong inhibitors, plus sertindole, asapiprant, duvelisib and dasatinib as moderate inhibitors in reported assays.",
      "unknown": "Inhibition and metabolite findings are from the reported in vitro systems and do not by themselves establish clinically relevant drug–drug interactions, exposure changes, or patient harm. Reported inhibition strengths are assay-specific; clinical relevance depends on dose, exposure and context and requires pharmacokinetic/clinical assessment.",
      "ai_role": "The authors used machine-learning models to prioritize drug-like compounds for CYP2C9 inhibition testing. Follow-up in vitro assays identified the listed strong and moderate inhibitors; CYP2C9 metabolism assays separately characterized metabolites of four drugs.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/35081108/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-deep-learning-enables-multi-site-bone-marrow-adiposity-genetics-in-uk-biobank",
    "title": "Deep learning enables multi-site bone marrow adiposity genetics in UK Biobank",
    "summary": "Deep learning measured bone marrow fat fraction at four skeletal sites in UK Biobank MRI data; after quality control, 48,608 participants contributed to genome-wide analyses. In the white-participant meta-GWAS, the study reported 67, 147, 134 and 174 independent significant variants for femoral head, total hip, femoral diaphysis and spine, mapping to 54, 90, 43 and 100 genes, respectively. Analyses also identified site- and sex-specific genetic associations and added signals in multi-ancestry analyses.",
    "field": "biology",
    "ai_technology": "Previously validated deep-learning models for MRI segmentation and bone marrow fat-fraction measurement",
    "organization": "University of Edinburgh; University of Lille; Heriot-Watt University; University of Southern Denmark; University of Dundee; University of Southern Denmark and collaborating UK Biobank institutions",
    "source_url": "https://www.nature.com/articles/s41467-024-55422-4",
    "source_type": "paper",
    "publication_date": "2025-01-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-55422-4",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Deep learning measured bone marrow fat fraction at four skeletal sites in UK Biobank MRI data; after quality control, 48,608 participants contributed to genome-wide analyses.",
      "unknown": "The GWAS, transcriptome-wide and colocalization results are statistical associations and candidate gene mappings, not proof of causal mechanisms. Ancestry subgroup sizes differ, and results do not establish clinical utility or disease prevention. The deep-learning measurement method was established in an earlier study; this record concerns the distinct population-scale phenotype and genetic findings. A 2025 author correction fixed a misspelling of AKNA as ANKA; it did not alter the reported analysis or findings.",
      "ai_role": "The study applied deep-learning models to segment four skeletal regions and derive bone marrow fat-fraction phenotypes at UK Biobank scale. Those AI-derived measurements were the phenotypes used in the study's subsequent genetic association analyses; the paper's genetic findings are not themselves outputs of a generative model.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-55422-4"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-05-deep-learning-maps-previously-unknown-disease-signals-in-ecgs-across-two-health-systems",
    "title": "Deep learning maps previously unknown disease signals in ECGs across two health systems",
    "summary": "Researchers independently trained PheWASNet on ECG and electronic health-record data from Stanford Health Care and Cedars-Sinai to assess 1,243 disease phenotypes. In held-out patient splits at both centers, the study reported previously unrecognized ECG associations for disorders including respiratory failure, neutropenia and menstrual disorders (AUCs 0.86, 0.83 and 0.84 at Stanford, respectively). Cross-condition model-output analysis found that 35 of 37 strongly detectable non-cardiac phenotypes shared signals with only four other disease outputs, and simple ECG measurement models explained some findings.",
    "field": "medicine",
    "ai_technology": "PheWASNet, a multi-task deep-learning model based on a one-dimensional EfficientNet architecture",
    "organization": "Stanford University; Cedars-Sinai Medical Center; University of California, Los Angeles",
    "source_url": "https://academic.oup.com/ehjdh/article/6/4/595/8126967",
    "source_type": "paper",
    "publication_date": "2025-05-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/ehjdh/ztaf047",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In held-out patient splits at both centers, the study reported previously unrecognized ECG associations for disorders including respiratory failure, neutropenia and menstrual disorders (AUCs 0.86, 0.83 and 0.84 at Stanford, respectively).",
      "unknown": "This retrospective EHR-linked study establishes model-detectable associations, not causal physiology, screening benefit or clinical utility. The disease labels are diagnostic codes associated with ECGs within 30 days; results may reflect comorbidity or care context. The authors caution that generalization to other health systems and populations is not assured. Some detectable signals were explained by simple conventional ECG measurements, and the paper does not claim that deep learning is required for every association.",
      "ai_role": "The researchers trained separate deep-learning models on raw 12-lead ECG waveforms linked to EHR diagnoses at two medical centers, then used held-out patient test sets to identify disease phenotypes whose signals were detectable from ECGs. Cross-phenotype prediction comparisons and simple-model analyses were used to characterize shared and interpretable signals.",
      "evidence_urls": [
        "https://academic.oup.com/ehjdh/article/6/4/595/8126967"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-08-machine-learning-finds-plasma-protein-signals-up-to-14-years-before-parkinson-diagnosis",
    "title": "Machine learning finds plasma protein signals up to 14 years before Parkinson diagnosis",
    "summary": "Using UK Biobank measurements of 2,937 plasma proteins, researchers found 446 dysregulated proteins in incident Parkinson cases and used machine learning to select a 23-protein panel for prediction. The panel achieved test AUCs of 0.78 in incident and 0.795 in prevalent cases; nine selected proteins were already dysregulated 10–14 years before diagnosis. In a separate Parkinson’s Progression Markers Initiative cohort, a 16-protein subset shared between platforms achieved validation AUC up to 0.76.",
    "field": "medicine",
    "ai_technology": "Support vector machine and other machine-learning models, with recursive feature elimination for panel selection",
    "organization": "Johnson & Johnson; UK Biobank; Parkinson’s Progression Markers Initiative collaborators",
    "source_url": "https://academic.oup.com/brain/article/149/4/1254/8234010",
    "source_type": "paper",
    "publication_date": "2025-08-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/brain/awaf303",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using UK Biobank measurements of 2,937 plasma proteins, researchers found 446 dysregulated proteins in incident Parkinson cases and used machine learning to select a 23-protein panel for prediction.",
      "unknown": "The result is a retrospective biomarker/prediction study, not a validated screening test or evidence that intervention based on the panel improves outcomes. The external PPMI analysis retrained and tested models in that cohort rather than evaluating a frozen UK Biobank model; validation AUCs varied by visit definition, and the reported maximum was 0.76. Protein associations and co-expression modules do not establish causal mechanisms. It is distinct from the earlier 2024 iRBD study in this ledger (DOI 10.1038/s41467-024-48961-3): separate cohorts and a different 23-protein/14-year result versus an eight-protein targeted panel and up-to-seven-year conversion finding; this record does not claim first discovery of presymptomatic plasma signals in Parkinson disease.",
      "ai_role": "Machine-learning classifiers were trained on UK Biobank plasma proteomic measurements to predict incident Parkinson disease; recursive feature elimination selected a 23-protein subset. The authors then tested the shared protein subset in the separate PPMI cohort using its own train/test split.",
      "evidence_urls": [
        "https://academic.oup.com/brain/article/149/4/1254/8234010"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-09-machine-learning-and-tear-fluid-proteomics-identify-a-six-protein-als-signature",
    "title": "Machine learning and tear-fluid proteomics identify a six-protein ALS signature",
    "summary": "In discovery samples from 49 people with amyotrophic lateral sclerosis (ALS) and 54 controls, mass spectrometry detected 876 tear-fluid proteins, of which 106 differed in abundance. Biostatistical and machine-learning analyses produced a six-protein signature (CRYM, PFKL, CAPZA2, ALDH16A1, SERPINC1 and HP). In an independent cohort of 51 ALS cases and 52 controls, Western blots replicated significant group differences for SERPINC1 and HP; their combination had the best reported in-silico discrimination.",
    "field": "medicine",
    "ai_technology": "Machine-learning models applied to data-independent acquisition mass-spectrometry proteomics and biomarker-panel discrimination",
    "organization": "Technical University of Munich; Helmholtz Munich; Ludwig-Maximilians University of Munich; University of Melbourne; University Medical Center Göttingen; Max Planck Institute for Multidisciplinary Sciences; German Center for Neurodegenerative Diseases",
    "source_url": "https://actaneurocomms.biomedcentral.com/articles/10.1186/s40478-025-02109-6",
    "source_type": "paper",
    "publication_date": "2025-09-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s40478-025-02109-6",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In an independent cohort of 51 ALS cases and 52 controls, Western blots replicated significant group differences for SERPINC1 and HP; their combination had the best reported in-silico discrimination.",
      "unknown": "This is an exploratory biomarker study with modest discovery and validation cohorts, not a clinically validated diagnostic assay. The article reports that only SERPINC1 and HP differences replicated significantly across cohorts; the six-protein signature is a candidate and its best discriminatory performance was evaluated in silico. It does not establish ALS-specific causality or clinical utility.",
      "ai_role": "Machine learning contributed to identification of ALS-associated differential proteins and pathways and evaluated discrimination by the derived signature. The selected tear-fluid proteins were then assayed in an independent patient/control cohort by Western blot; only SERPINC1 and HP differences are explicitly reported as replicated across cohorts.",
      "evidence_urls": [
        "https://actaneurocomms.biomedcentral.com/articles/10.1186/s40478-025-02109-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-machine-learning-identifies-serum-protein-signatures-for-progression-from-asymptomatic-atherosclerosis-to-ischemic-stroke",
    "title": "Machine learning identifies serum protein signatures for progression from asymptomatic atherosclerosis to ischemic stroke",
    "summary": "In a discovery cohort of 43 patients with asymptomatic large-artery atherosclerosis and a validation cohort of 39, investigators combined serum proteomics with machine learning to distinguish acute ischemic stroke from large-artery atherosclerosis. XGBoost selected a three-protein panel (RNASE4, HBA1 and ATF6B; reported AUC 0.917, specificity 80%) and a six-protein panel distinguishing patients from healthy controls (reported AUC 0.971, specificity 86%). ELISA assays confirmed higher GUCA2A and FLNA in patient groups.",
    "field": "medicine",
    "ai_technology": "XGBoost machine-learning models applied to serum proteomics; DeepPRM-Stellar targeted mass spectrometry supported peptide validation and panel optimization",
    "organization": "Huashan Hospital, Fudan University; Zhongshan Hospital, Fudan University; Xiamen University; Xiang An Biomedicine Laboratory",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40692393/",
    "source_type": "paper",
    "publication_date": "2025-08-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/d5an00396b",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a discovery cohort of 43 patients with asymptomatic large-artery atherosclerosis and a validation cohort of 39, investigators combined serum proteomics with machine learning to distinguish acute ischemic stroke from large-artery atherosclerosis.",
      "unknown": "The abstract reports 43 discovery and 39 validation participants, but does not establish prospective prediction or external clinical deployment. The signatures are candidate biomarkers, not a validated clinical test. Reported performance comes from this study and should not be interpreted as patient benefit or causal evidence. ELISA confirmation is reported for GUCA2A and FLNA, not every protein in the ML panels.",
      "ai_role": "The study reports that machine-learning techniques were used to identify circulating biomarkers of progression and that XGBoost selected the three- and six-protein signatures from targeted proteomic measurements. ELISA follow-up confirmed GUCA2A and FLNA group differences. The models are retrospective cohort analyses; the study did not prospectively predict future stroke events.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40692393/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-mp4-text-to-protein-design-yields-atp-active-proteins-including-one-with-a-novel-fold",
    "title": "MP4 text-to-protein design yields ATP-active proteins, including one with a novel fold",
    "summary": "The transformer MP4 maps plain-language prompts to protein sequences. The study reports two experimentally validated de novo designs that were expressible and thermostable and showed ATP binding and hydrolysis in vitro. X-ray structures at 1.30 Å and 1.77 Å showed that one design adopts a fold not previously observed in the authors’ structural comparisons.",
    "field": "biology",
    "ai_technology": "MP4, a transformer-based text-to-protein model",
    "organization": "310 AI; University of California, Berkeley; McGill University; Arctoris",
    "source_url": "https://pubs.acs.org/asbcd6/article/doi/10.1021/acssynbio.6c00246/5254330/An-Artificial-Intelligence-Model-for-Translating",
    "source_type": "paper",
    "publication_date": "2026-08-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acssynbio.6c00246",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reports two experimentally validated de novo designs that were expressible and thermostable and showed ATP binding and hydrolysis in vitro.",
      "unknown": "Only two designed proteins are described as experimentally validated in the abstract; this is an early proof of concept, not a demonstrated general success rate. ATP binding and hydrolysis are in-vitro results, not a novel physiological function or therapeutic effect. The source describes one fold as novel; no claim is made that both structures are novel.",
      "ai_role": "MP4 converted natural-language protein specifications into candidate sequences. Researchers evaluated 96 prompts in silico, selected candidates, expressed and tested designs, and solved two crystal structures. The source attributes the designed sequences to MP4; experimental teams established expression, thermostability, ATP binding/hydrolysis and structural novelty.",
      "evidence_urls": [
        "https://pubs.acs.org/asbcd6/article/doi/10.1021/acssynbio.6c00246/5254330/An-Artificial-Intelligence-Model-for-Translating"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-a-graph-neural-network-nominates-osmosmr-for-melanoma-immunotherapy-with-ex-vivo-explant-evidence",
    "title": "A graph neural network nominates OSM–OSMR for melanoma immunotherapy, with ex-vivo explant evidence",
    "summary": "MIDAS, a multimodal graph-neural-network model, ranked candidate immuno-oncology targets from gene-interaction, patient multi-omic and immune-biology data. In eight stage III/IV melanoma patient-derived explants, researchers blocked OSM–OSMR signalling with an anti-OSM antibody. The paper reports reduced dysfunctional CD8+ T-cell populations and lower CCL4 levels; PTPN22 inhibition did not produce comparable changes in this assay.",
    "field": "medicine",
    "ai_technology": "MIDAS, a multimodal graph neural network for immuno-oncology target ranking",
    "organization": "University College London; The Francis Crick Institute; Royal Marsden NHS Foundation Trust; University College London Hospitals; Royal Holloway, University of London",
    "source_url": "https://www.nature.com/articles/s42256-026-01201-3",
    "source_type": "paper",
    "publication_date": "2026-05-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42256-026-01201-3",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports reduced dysfunctional CD8+ T-cell populations and lower CCL4 levels; PTPN22 inhibition did not produce comparable changes in this assay.",
      "unknown": "Functional follow-up used only eight stage III/IV melanoma patient-derived explants. This is an ex-vivo preclinical result, not treatment of patients or evidence of clinical benefit. OSM–OSMR biology has prior literature in other tumour settings; the new paper-specific contribution is MIDAS target nomination and measured perturbation effects in melanoma explants. Do not generalize to other cancers or claim a validated therapy.",
      "ai_role": "MIDAS integrated gene-interaction networks with patient multi-omics, immune-cell biology and genetic-perturbation evidence to rank candidate immunotherapy targets. Researchers selected OSM–OSMR for functional follow-up and tested pathway blockade in patient-derived tumour explants; the model nominated the target, while the ex-vivo experiments measured the T-cell and cytokine outcomes.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42256-026-01201-3"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-11-machine-learning-metabolomics-identifies-dheas-and-citrate-signatures-in-anti-mda5-dermatomyositis",
    "title": "Machine-learning metabolomics identifies DHEAS and citrate signatures in anti-MDA5 dermatomyositis",
    "summary": "In a single-centre cohort of 183 patients with idiopathic inflammatory myopathy and 104 healthy controls, seven machine-learning classifiers and LASSO feature selection were used with plasma and PBMC metabolomics to distinguish anti-MDA5-positive dermatomyositis. DHEAS and citrate were selected across the integrated analyses; plasma assays showed lower levels in anti-MDA5-positive cases, with concordant patterns reported in a separate 33-patient/33-control validation cohort. Within cases, DHEAS correlated inversely with rapidly progressive interstitial lung disease and citrate with rash.",
    "field": "medicine",
    "ai_technology": "Seven machine-learning classifiers (including random forest and XGBoost) with LASSO feature selection",
    "organization": "The First Affiliated Hospital of Zhengzhou University",
    "source_url": "https://arthritis-research.biomedcentral.com/articles/10.1186/s13075-025-03676-8",
    "source_type": "paper",
    "publication_date": "2025-11-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s13075-025-03676-8",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a single-centre cohort of 183 patients with idiopathic inflammatory myopathy and 104 healthy controls, seven machine-learning classifiers and LASSO feature selection were used with plasma and PBMC metabolomics to distinguish anti-MDA5-positive dermatomyositis.",
      "unknown": "Single-centre study with a relatively small validation group (33 anti-MDA5-positive cases and 33 controls). The metabolite associations with RP-ILD and rash are observational and do not establish causality. Reported discrimination is exploratory and does not establish a clinical diagnostic test or patient benefit; no treatment response was studied. DHEAS and citrate are known metabolites; the paper-specific result is their measured profile and candidate biomarker association in this disease cohort.",
      "ai_role": "The models ranked discriminatory metabolites and selected features for disease classification; the team integrated these with pathway results to prioritize candidate biomarkers. DHEAS and citrate were then measured using ELISA and a biochemical assay, with patterns assessed in the validation cohort. Machine learning aided biomarker selection and classification; assays and clinical data supplied validation.",
      "evidence_urls": [
        "https://arthritis-research.biomedcentral.com/articles/10.1186/s13075-025-03676-8"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-12-machine-learning-metabolomics-prioritizes-succinate-in-sepsis-associated-encephalopathy",
    "title": "Machine-learning metabolomics prioritizes succinate in sepsis-associated encephalopathy",
    "summary": "In plasma samples from 29 controls, 32 sepsis patients and 27 patients with sepsis-associated encephalopathy (SAE), LC–MS/MS plus machine-learning feature selection identified a 12-metabolite profile. Succinate rose stepwise from controls to sepsis to SAE, ranked among the five leading XGBoost features, and correlated with APACHE II, SOFA and lactate. In a cecal-ligation-and-puncture mouse model, dietary succinate supplementation before sepsis worsened cognitive-test performance, neuronal injury and microglial activation; succinate also increased inflammatory gene expression in cultured microglia.",
    "field": "biology",
    "ai_technology": "Random forest and support-vector-machine feature ranking with XGBoost classification",
    "organization": "Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12809278/",
    "source_type": "paper",
    "publication_date": "2025-12-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.isci.2025.114520",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a cecal-ligation-and-puncture mouse model, dietary succinate supplementation before sepsis worsened cognitive-test performance, neuronal injury and microglial activation; succinate also increased inflammatory gene expression in cultured microglia.",
      "unknown": "The human discovery cohort comprised 88 participants and the held-out model test set had 18 participants, so diagnostic performance and metabolite ranking need larger external cohorts. Succinate supplementation was given before CLP in mice, and cell experiments used a microglial cell line; these are preclinical perturbations and do not prove that succinate causes human SAE or that targeting it improves patient outcomes. Associations with severity scores are observational.",
      "ai_role": "Machine-learning feature selection and classifier comparison were applied to the measured plasma metabolomics to identify discriminatory features; succinate ranked among the top five XGBoost features. Researchers then tested its functional relevance through supplementation in a mouse sepsis model and in cultured microglia. The animal/cell experiments were human-directed follow-up, not model-generated predictions.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12809278/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-10-machine-learning-and-mendelian-randomization-prioritize-ca2-as-a-diabetic-nephropathy-biomarker",
    "title": "Machine learning and Mendelian randomization prioritize CA2 as a diabetic nephropathy biomarker",
    "summary": "Across six kidney-expression datasets, researchers identified 314 differentially expressed genes in diabetic nephropathy and used LASSO, SVM-RFE and random forest to prioritize seven candidate genes. Mendelian-randomization integration further prioritized carbonic anhydrase II (CA2); its ROC AUC exceeded 0.878 across the analyzed datasets. The authors also reported qPCR validation of CA2 in clinical patient and control samples and an association with inferred immune-cell infiltration.",
    "field": "medicine",
    "ai_technology": "LASSO, support-vector-machine recursive feature elimination and random-forest feature selection",
    "organization": "Shanghai Tenth People’s Hospital, Tongji University School of Medicine; Yangpu Hospital, Tongji University School of Medicine",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39370621/",
    "source_type": "paper",
    "publication_date": "2024-10-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1111/dom.15933",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors also reported qPCR validation of CA2 in clinical patient and control samples and an association with inferred immune-cell infiltration.",
      "unknown": "This is a candidate biomarker study, not evidence of a clinically deployed test or patient benefit. Reported AUC values are retrospective dataset performance; qPCR validation supports differential expression but does not establish CA2 as causal or disease-specific. The accessible abstract does not report the clinical validation sample size, so none is inferred.",
      "ai_role": "Three machine-learning approaches integrated candidate-gene selection from kidney-expression datasets; Mendelian randomization with expression-QTL data then prioritized CA2, which was evaluated as a diagnostic biomarker and measured by qPCR in clinical samples. The models supported gene prioritization and discrimination, while qPCR supplied the wet-lab follow-up.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39370621/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-machine-learning-prioritizes-1-methylnicotinamide-as-a-serum-signal-across-hbv-cirrhosis-to-hcc-progression",
    "title": "Machine learning prioritizes 1-methylnicotinamide as a serum signal across HBV cirrhosis-to-HCC progression",
    "summary": "Untargeted serum metabolomics from 69 participants, including eight patients followed from HBV-related cirrhosis to hepatocellular carcinoma (HCC), was analyzed with LASSO and random forest to prioritize metabolites. 1-methylnicotinamide (1-MNAM) declined across healthy, cirrhosis and HCC groups and was lower at the HCC than cirrhosis stage in the eight progressors. Targeted mass spectrometry in a separate cohort of 25 healthy controls and 25 early-stage HCC patients confirmed lower 1-MNAM in HCC (AUC 0.954 for HCC versus healthy controls).",
    "field": "medicine",
    "ai_technology": "LASSO regression and random forest classifier applied to serum metabolomics",
    "organization": "Centre for Medical Research, Ningbo No. 2 Hospital; Liver Disease Department of Integrative Medicine, Ningbo No. 2 Hospital",
    "source_url": "https://link.springer.com/article/10.1007/s00432-024-05848-6",
    "source_type": "paper",
    "publication_date": "2024-06-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1007/s00432-024-05848-6",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Targeted mass spectrometry in a separate cohort of 25 healthy controls and 25 early-stage HCC patients confirmed lower 1-MNAM in HCC (AUC 0.954 for HCC versus healthy controls).",
      "unknown": "The longitudinal transition signal comes from eight HBV-cirrhosis patients who progressed during follow-up. The independent targeted cohort compared 25 early-HCC patients with 25 healthy controls; it did not validate prediction of progression among cirrhosis patients. The authors describe the overall sample as small and call for larger prospective validation. 1-MNAM is a known endogenous metabolite; this record concerns the study-specific association and selection result, not discovery of a new compound, causal mechanism, or clinical test.",
      "ai_role": "The authors used LASSO and random forest to select ten candidate metabolites distinguishing cirrhosis from HCC, then identified 1-MNAM as the sole candidate with a consistent decline across healthy, cirrhosis and HCC groups. The algorithms prioritized the measured metabolite; targeted mass spectrometry in a separate cohort provided assay follow-up.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1007/s00432-024-05848-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-multi-omics-machine-learning-identifies-lysophosphatidic-acid-in-prrsv-infection-and-links-it-to-immune-suppression",
    "title": "Multi-omics machine learning identifies lysophosphatidic acid in PRRSV infection and links it to immune suppression",
    "summary": "Machine-learning analysis of untargeted serum metabolomic and lipidomic profiles from clinical pig cohorts identified lysophosphatidic acid (LPA) as a PRRSV-associated biomarker. Follow-up infection experiments reported increased LPA with autotaxin upregulation; in porcine alveolar macrophages, LPA manipulation altered RIG-I/type-I-interferon responses and viral replication. In infected pigs, targeting LPA was reported to protect against infection and associated disease outcomes.",
    "field": "biology",
    "ai_technology": "Random-forest analysis of serum metabolomic and lipidomic profiles",
    "organization": "Sun Yat-sen University; Huazhong Agricultural University; Wen’s Foodstuff Group Co. Ltd.; Stanford University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11425916/",
    "source_type": "paper",
    "publication_date": "2024-07-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202402025",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Machine-learning analysis of untargeted serum metabolomic and lipidomic profiles from clinical pig cohorts identified lysophosphatidic acid (LPA) as a PRRSV-associated biomarker.",
      "unknown": "The clinical classifier and biomarker work concerns pigs, not human diagnosis. The infection challenge experiments were small (figure descriptions report five piglets per group), and cell assays were performed in vitro. The authors report that the mechanism by which LPA suppresses RIG-I remains unclear; the findings do not establish a treatment for people or broad field effectiveness in livestock.",
      "ai_role": "Random-forest analysis ranked metabolite and lipid features from clinical PRRSV-positive and negative pig cohorts and contributed to identification of LPA as the candidate biomarker. The authors then tested LPA in infection models, including cultured porcine alveolar macrophages and infected pigs, to examine host-response and disease outcomes.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11425916/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-machine-learning-metabolomics-distinguishes-sars-cov-2-influenza-a-and-rsv-from-respiratory-swabs",
    "title": "Machine-learning metabolomics distinguishes SARS-CoV-2, influenza A and RSV from respiratory swabs",
    "summary": "Untargeted metabolomics and machine learning identified upper-respiratory-swab signatures for SARS-CoV-2 positivity, followed by targeted LC-MS/MS profiling. Across 1,226 samples (521 SARS-CoV-2-positive, 97 influenza A, 96 RSV, 211 other respiratory viruses and 301 virus-negative), the best SARS-CoV-2 LightGBM model had AUC 0.99, sensitivity 0.96 and specificity 0.95. Separate models reported AUC 0.97 for influenza A and 0.99 for RSV. The top-ranked SARS-CoV-2 features included 3-oxo-heneicosanoic acid and 2-(4-hydroxyphenyl) ethanol.",
    "field": "medicine",
    "ai_technology": "Light Gradient Boosting Machine and other machine-learning classifiers applied to mass-spectrometry metabolomics",
    "organization": "British Columbia Centre for Disease Control Public Health Laboratory; University of British Columbia; Stanford University School of Medicine; Stanford Health Care; Chan Zuckerberg Biohub",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41065362/",
    "source_type": "paper",
    "publication_date": "2025-10-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1128/jcm.02042-24",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Separate models reported AUC 0.97 for influenza A and 0.99 for RSV.",
      "unknown": "This is an assay-validation study using stored respiratory specimens, not a prospective clinical-impact trial. Results establish classification in the studied cohorts and do not show improved patient outcomes or routine deployment. The source notes unresolved oxo-position/isomer assignment for several features. The metabolite signals are candidate host-response markers, not proof of virus-specific causal biology.",
      "ai_role": "Machine-learning feature ranking identified metabolite signatures in untargeted profiles; the team adapted candidate signatures to targeted LC-MS/MS, and evaluated classification in respiratory-virus sample cohorts. The models supplied the sample-level classification and feature ranking, while mass spectrometry measured the metabolites.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41065362/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-multi-omics-machine-learning-prioritizes-a-plasma-lipid-signature-for-active-tuberculosis-differentiation",
    "title": "Multi-omics machine learning prioritizes a plasma lipid signature for active tuberculosis differentiation",
    "summary": "Machine-learning integration of plasma functional-omics data in two discovery cohorts and a validation cohort identified biosignatures that distinguished active tuberculosis from nontuberculous mycobacterial infection, latent TB infection and other lung diseases. Cohorts included 76 people in the TB–NTM discovery set, 72 in the TB–LTBI–other-lung-disease set, and 68 TB plus 30 LTBI participants in the validation set. PC(14:0_22:6) was the most important predictor and was lower in active TB; additional external datasets yielded reported AUCs of 0.77–1.00.",
    "field": "medicine",
    "ai_technology": "Multi-omics integration and machine-learning predictive models for feature selection and classification",
    "organization": "Inje University College of Medicine; Ton Duc Thang University; Hanoi University of Pharmacy; University of Copenhagen; Korea University Guro Hospital",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40598791/",
    "source_type": "paper",
    "publication_date": "2025-07-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jproteome.4c00989",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Machine-learning integration of plasma functional-omics data in two discovery cohorts and a validation cohort identified biosignatures that distinguished active tuberculosis from nontuberculous mycobacterial infection, latent TB infection and other lung diseases.",
      "unknown": "The article calls PC(14:0_22:6) a candidate differential-diagnostic biomarker. Reported AUCs vary by comparator group and dataset; they do not establish prospective performance, clinical utility, or replacement of microbiological diagnosis. The PubMed abstract does not provide all cohort-level assay details, so this summary remains bounded to abstract-reported information.",
      "ai_role": "The authors used machine-learning predictive modeling across integrated functional-omics measurements to identify and prioritize plasma signatures that discriminate active TB from clinically relevant non-TB groups. PC(14:0_22:6) emerged as the leading lipid feature; the analysis was evaluated across discovery, validation and two external datasets.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40598791/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-triscope-links-cell-type-specific-alzheimers-expression-programs-to-inherited-risk-loci",
    "title": "TriSCOPE links cell-type-specific Alzheimer’s expression programs to inherited risk loci",
    "summary": "TriSCOPE combined cell-level predictive modeling, donor-aware differential expression and pseudotime analysis across six single-nucleus RNA-sequencing datasets (about 4.2 million nuclei from 627 donors). It identified cell-type-specific predictively expressed genes that were often missed by univariate differential-expression analysis; conditional GWAS–eQTL colocalization supported 27 gene–cell-type pairs at Alzheimer’s risk loci, including pairs without significant differential expression. Selected gene signals and analytical branches were compared across independent cohorts.",
    "field": "biology",
    "ai_technology": "TriSCOPE, using LightGBM-based multivariate predictive modeling integrated with differential expression and pseudotime analysis",
    "organization": "Harvard Medical School; Harvard College; New York Genome Center; Rush Alzheimer’s Disease Center, Rush University",
    "source_url": "https://www.nature.com/articles/s41467-026-77651-5.pdf",
    "source_type": "paper",
    "publication_date": "2026-11-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-77651-5",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected gene signals and analytical branches were compared across independent cohorts.",
      "unknown": "This is a computational reanalysis of previously published, post-mortem cross-sectional datasets; it did not collect new tissue or perform perturbation experiments. The authors state the framework is for feature discovery rather than clinical prognosis, cannot establish temporal ordering, and requires functional perturbation studies to distinguish pathogenic drivers from protective or reactive signals. The reported cell-level AUCs are modest and should not be presented as patient-level diagnostic performance.",
      "ai_role": "The supervised machine-learning branch prioritized gene-expression features that predicted clinical Alzheimer’s diagnosis within cell types, conditional on demographics and APOE genotype. The study integrated those model-ranked features with donor-aware differential expression, pseudotime and GWAS–eQTL colocalization to nominate genetically supported cell-type programs.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-77651-5.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-spatial-glycan-imaging-and-machine-learning-identify-a-serum-n-glycan-panel-for-hcc-in-cirrhosis",
    "title": "Spatial glycan imaging and machine learning identify a serum N-glycan panel for HCC in cirrhosis",
    "summary": "Spatial N-glycan imaging of hepatocellular carcinoma (HCC) tissue and paired serum identified thirteen altered glycan structures; glycoproteomics linked them to more than 50 proteins, and sixteen glycoproteins were assessed in a separate patient set. Machine-learning algorithms combining glycan/glycoprotein features with clinical variables differentiated early-stage HCC from cirrhosis in a case-control cohort of 100 HCC and 101 cirrhosis-only patients, with reported AUROC values across the algorithms of 0.88–0.97.",
    "field": "medicine",
    "ai_technology": "Machine-learning algorithms using serum N-glycan and glycoprotein features, age, sex and AFP",
    "organization": "Medical University of South Carolina; GlycoPath, Inc.; HTX Technologies; University of California San Diego; University of Texas Southwestern Medical Center",
    "source_url": "https://www.nature.com/articles/s43856-024-00677-7.pdf",
    "source_type": "paper",
    "publication_date": "2024-12-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43856-024-00677-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Machine-learning algorithms combining glycan/glycoprotein features with clinical variables differentiated early-stage HCC from cirrhosis in a case-control cohort of 100 HCC and 101 cirrhosis-only patients, with reported AUROC values across the algorithms of 0.88–0.97.",
      "unknown": "The discovery samples included 53 HCC tissues and 23 matched sera; the 201-patient testing set was a case-control cohort, not prospective surveillance. The paper notes that sialic acid was removed before analysis, excluding an important class of glycan modification. It does not establish screening benefit or prospective clinical utility. Individual glycoforms were weak classifiers; performance came from combined feature algorithms.",
      "ai_role": "Researchers used spatial glycan imaging and glycoproteomics to select altered tissue and matched-serum features, then used machine-learning algorithms to combine selected glycan/glycoprotein measurements with clinical variables for HCC-versus-cirrhosis classification. The source reports model performance in an independent case-control patient set.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43856-024-00677-7.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-07-multimodal-ai-distinguishes-ten-dementia-etiologies-across-nine-clinical-datasets",
    "title": "Multimodal AI distinguishes ten dementia etiologies across nine clinical datasets",
    "summary": "A multimodal AI model analyzed demographics, personal and family history, medications, cognitive and functional testing, and neuroimaging across nine datasets totaling 51,269 participants. It assigned probabilities to ten primary or contributing dementia etiologies and addressed mixed cases. The paper reports microaveraged AUROC 0.96 for dementia-etiology differentiation; model probabilities aligned with etiology-specific biomarkers and postmortem pathology. In a review of 100 cases, AI-augmented neurologist assessment outperformed neurologist-only ratings.",
    "field": "medicine",
    "ai_technology": "Multimodal machine-learning model combining clinical, neuropsychological, functional and neuroimaging data",
    "organization": "Boston University; Stanford University; Peking Union Medical College Hospital; Columbia University Irving Medical Center; University of Nebraska Medical Center; additional clinical and research centers contributing cohorts",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38965435/",
    "source_type": "paper",
    "publication_date": "2024-07-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-024-03118-z",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A multimodal AI model analyzed demographics, personal and family history, medications, cognitive and functional testing, and neuroimaging across nine datasets totaling 51,269 participants.",
      "unknown": "The nine-dataset analysis is retrospective and includes in-house as well as external cohorts. The 100-case clinician comparison is small, and the study did not test prospective changes in care or patient outcomes. The authors call for prospective studies before claiming improved patient care. Model probabilities aligned with available biomarkers and postmortem evidence but are not histopathologic confirmation for every individual prediction.",
      "ai_role": "The model integrated routinely collected clinical variables and multimodal neuroimaging to estimate cognitive status and probabilities for ten dementia etiologies, including mixed etiologies and cases with incomplete features. It was evaluated across independent cohorts and compared with neurologist assessments; its output is the etiology-specific classification and probability result.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38965435/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-05-machine-learning-metabolomics-prioritizes-serum-metabolites-for-sle-diagnosis-and-activity-assessment",
    "title": "Machine-learning metabolomics prioritizes serum metabolites for SLE diagnosis and activity assessment",
    "summary": "Targeted serum metabolomics compared 89 patients with systemic lupus erythematosus (SLE) and 44 healthy participants, identifying 146 differential metabolites. The authors combined statistical screening with seven machine-learning models to prioritize eight diagnostic biomarker candidates (reported AUCs above 0.95). Two additional metabolites, N-methyl-L-glutamic acid and L-2-aminobutyric acid, were evaluated by ELISA in a 90-patient SLE validation cohort and associated with SLE Disease Activity Index scores.",
    "field": "medicine",
    "ai_technology": "Seven machine-learning classification models applied to targeted serum metabolomics",
    "organization": "The Affiliated Guangzhou Panyu Central Hospital of Guangzhou Medical University; Guangzhou University of Chinese Medicine; Sun Yat-sen Memorial Hospital, Sun Yat-sen University; Panyu District Central Hospital",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38710348/",
    "source_type": "paper",
    "publication_date": "2024-05-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.clim.2024.110235",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors combined statistical screening with seven machine-learning models to prioritize eight diagnostic biomarker candidates (reported AUCs above 0.95).",
      "unknown": "This retrospective case-control study has modest cohorts and does not establish a clinically deployed test, causal SLE mechanism, or patient benefit. Reported diagnostic AUCs and SLEDAI associations require independent prospective replication; the validation cohort described in the abstract consists of SLE patients and does not by itself establish performance against a new external healthy/control population. The authors’ pathway interpretation is hypothesis-generating.",
      "ai_role": "The study iteratively compared seven machine-learning models to select candidate metabolites for SLE classification and combined the selected features with statistical screening. N-methyl-L-glutamic acid and L-2-aminobutyric acid were subsequently assessed by ELISA in a patient validation cohort; the machine-learning analysis nominated candidates, while the assays supplied biological measurement follow-up.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38710348/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-02-ehr-machine-learning-and-genetic-colocalization-link-osteoporosis-to-female-alzheimers-risk-near-ms4a6a",
    "title": "EHR machine learning and genetic colocalization link osteoporosis to female Alzheimer’s risk near MS4A6A",
    "summary": "Random-forest models trained on UCSF records (749 expert-diagnosed Alzheimer’s disease cases and 250,545 controls) identified clinical features predictive up to seven years before the recorded onset; the seven-year model’s AUROC was 0.72 using clinical features alone. Matched-cohort follow-up in external University of California health-system records found faster Alzheimer’s diagnosis after osteoporosis in women. Genetic colocalization analyses supported a female-specific shared signal between Alzheimer’s disease, low heel bone mineral density and MS4A6A expression near the MS4A gene cluster. The paper also reports shared APOE-locus support for hyperlipidemia and Alzheimer’s disease.",
    "field": "medicine",
    "ai_technology": "Random-forest models on longitudinal electronic health records, interpreted with the SPOKE biomedical knowledge network",
    "organization": "University of California, San Francisco; Stanford University; University of California health-system collaborators; University of Oxford",
    "source_url": "https://www.nature.com/articles/s43587-024-00573-8.pdf",
    "source_type": "paper",
    "publication_date": "2024-02-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43587-024-00573-8",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Matched-cohort follow-up in external University of California health-system records found faster Alzheimer’s diagnosis after osteoporosis in women.",
      "unknown": "This is hypothesis-generating observational work. EHR features may reflect healthcare access, clinician behavior, coding and biological processes; the authors explicitly state that further validation across omics data types is needed. Genetic colocalization and matched cohort associations do not establish that osteoporosis causes Alzheimer’s disease or that treating osteoporosis prevents it. The prediction model is not a prospective clinical-impact evaluation, and early risk discrimination is modest with the low-prevalence outcome.",
      "ai_role": "Random-forest models ranked EHR-derived clinical features associated with future Alzheimer’s diagnosis. The team used matched models to prioritize hypotheses, then used SPOKE to connect clinical predictors to biological networks and independent health-system data plus GWAS/eQTL colocalization to examine the osteoporosis and hyperlipidemia associations. The model-generated prioritization led to follow-up of the female-specific osteoporosis signal near MS4A6A.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43587-024-00573-8.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-05-machine-learning-proteomics-identifies-a-replicated-plasma-signature-associated-with-alzheimers-disease",
    "title": "Machine-learning proteomics identifies a replicated plasma signature associated with Alzheimer’s disease",
    "summary": "A three-stage analysis of plasma proteomics from 1,270 people with clinical Alzheimer’s disease (AD) and 2,096 cognitively normal controls identified 416 proteins associated with clinical AD, including 294 that the paper classified as not previously associated with AD in large-scale plasma or CSF proteomic studies. A LASSO-selected seven-protein model predicted clinical AD and biomarker-defined AD status, with validation in external cohorts and orthogonal assays; the signature also stratified later progression to symptomatic AD among initially cognitively normal participants.",
    "field": "medicine",
    "ai_technology": "LASSO feature selection with logistic regression, using five-fold cross-validation and age/sex covariates",
    "organization": "Washington University School of Medicine; and Knight Alzheimer Disease Research Center; Stanford Alzheimer’s Disease Research Center; Rush Alzheimer’s Disease Center; Global Neurodegeneration Proteomics Consortium; collaborating Alzheimer research centers",
    "source_url": "https://pure.amsterdamumc.nl/ws/portalfiles/portal/140778915/Large-scale-plasma-proteomic-profiling-unveils-diagnostic-biomarkers-and-pathways-for-alzheimers-disease.pdf",
    "source_type": "paper",
    "publication_date": "2025-05-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43587-025-00872-8",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A LASSO-selected seven-protein model predicted clinical AD and biomarker-defined AD status, with validation in external cohorts and orthogonal assays; the signature also stratified later progression to symptomatic AD among initially cognitively normal participants.",
      "unknown": "The cohort analysis is observational and does not establish that the associated proteins cause AD. The model’s clinical-status AUC was 0.72 in replication; performance was comparable to p-tau217 in the study’s comparison, but the reported AUC for frontotemporal dementia was also 0.73, limiting disease specificity. Orthogonal platform validation covered only subsets of the seven proteins (four proteins on Olink and two on Alamar), and the authors state that further work is needed before clinical use. The reported “new” designation is relative to prior large-scale plasma/CSF proteomic studies, not proof that each protein is biologically novel.",
      "ai_role": "The team used LASSO to select a compact protein panel from the proteins associated with AD in discovery and replication analyses, then retrained logistic regression to predict clinical AD status. This model-driven panel was evaluated against amyloid and tau biomarkers, external cohorts and orthogonal protein platforms; the analysis also tested associations with progression to symptomatic AD and memory decline.",
      "evidence_urls": [
        "https://pure.amsterdamumc.nl/ws/portalfiles/portal/140778915/Large-scale-plasma-proteomic-profiling-unveils-diagnostic-biomarkers-and-pathways-for-alzheimers-disease.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-09-deep-learning-identifies-previously-unreported-protein-interactions-in-human-bacterial-pathogens",
    "title": "Deep learning identifies previously unreported protein interactions in human bacterial pathogens",
    "summary": "A RoseTTAFold2-Lite pipeline combined residue coevolution, deep-learning complex prediction and AlphaFold modelling to screen protein pairs across 19 human bacterial pathogens. In a validation set selected using the deep-learning pipeline, co-immunoprecipitation detected four of six predicted interactions. Three pairs lacked prior direct interaction evidence: E. coli UbiE–YcaR, P. aeruginosa PA4105–PA4106, and P. aeruginosa WbpG–HisF2. The PA4105–PA4106 structural similarity to proteins in 3-thiaglutamate biosynthesis and the WbpG–HisF2 result provide functional hypotheses, not established native-pathogen mechanisms.",
    "field": "biology",
    "ai_technology": "RoseTTAFold2-Lite integrated with direct coupling analysis and AlphaFold complex modelling",
    "organization": "University of Washington Institute for Protein Design; University of Texas Southwestern Medical Center; Broad Institute; Massachusetts General Hospital; University of Connecticut; Seoul National University; collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11445079/",
    "source_type": "paper",
    "publication_date": "2024-09-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41564-024-01791-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a validation set selected using the deep-learning pipeline, co-immunoprecipitation detected four of six predicted interactions.",
      "unknown": "The positive interaction evidence comes from heterologous co-immunoprecipitation/pull-down assays, with two independent experiments per positive pair, rather than native-pathogen systems. The validation subset was small (four positives among six AI-selected candidates); most of the thousands of predicted interactions remain untested, and false positives and false negatives are expected. Proposed roles for PA4105–PA4106 and UbiE–YcaR remain hypotheses; this paper does not establish their physiological function or a therapeutic target.",
      "ai_role": "RoseTTAFold2-Lite used residue–residue coevolution and predicted complex structures to rank candidate protein pairs in a proteome-wide screen. The authors selected six pairs from the final deep-learning pipeline for co-immunoprecipitation tests; four yielded positive interaction signals. Three were described as lacking previously reported direct interaction evidence, including UbiE–YcaR, PA4105–PA4106 and WbpG–HisF2. The models enabled the experimental target selection; they did not themselves establish physical interactions.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11445079/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-machine-learning-identifies-plasma-metabolite-panels-for-gastric-cancer-diagnosis-and-prognosis",
    "title": "Machine learning identifies plasma metabolite panels for gastric-cancer diagnosis and prognosis",
    "summary": "In a multicenter study of 702 participants, targeted plasma metabolomics and machine learning produced a 10-metabolite gastric-cancer diagnostic model and separate 28-metabolite survival-risk model. The diagnostic model was evaluated in an independent 95-person cohort (AUROC 0.920; sensitivity 0.905, specificity 0.75) and detected stage IA disease with AUROC 0.909. Among 181 patients followed for a median 40 months, the prognostic model achieved test-set concordance index 0.83. Succinate, uridine and lactate were among diagnostic features.",
    "field": "medicine",
    "ai_technology": "LASSO, random forest and random survival forest",
    "organization": "Tsinghua University; National Cancer Center/National Clinical Research Center for Cancer, China; Harbin Medical University Cancer Hospital; collaborating Chinese cancer centers",
    "source_url": "https://www.nature.com/articles/s41467-024-46043-y.pdf",
    "source_type": "paper",
    "publication_date": "2024-02-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-46043-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The diagnostic model was evaluated in an independent 95-person cohort (AUROC 0.920; sensitivity 0.905, specificity 0.75) and detected stage IA disease with AUROC 0.909.",
      "unknown": "Case-control cohorts do not establish population screening performance. External diagnostic test n=95; prognosis cohort median follow-up 40 months with right-censored outcomes. The paper says direct clinical use is premature and requires quality-control samples. Associations are not causal and no model-guided benefit was tested.",
      "ai_role": "LASSO selected ten plasma metabolites for a diagnostic model evaluated in internal and independent external cohorts. Random survival forest selected a 28-metabolite prognostic panel, evaluated in a held-out patient set.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-46043-y.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-07-machine-learning-selection-and-external-mass-spectrometry-testing-identify-an-ards-28-day-prognosis-protein-panel",
    "title": "Machine-learning selection and external mass-spectrometry testing identify an ARDS 28-day prognosis protein panel",
    "summary": "A multicenter observational serum multi-omics study used LASSO and Boruta to select candidate biomarkers for 28-day outcome in acute respiratory distress syndrome. Eight proteins (VCAM1, LDHB, MSN, FLG2, TAGLN2, LMNA, MBL2 and LBP) showed consistency in an independent cohort of 183 patients and were confirmed by targeted parallel-reaction-monitoring mass spectrometry. The protein model AUC was 0.802 in validation versus 0.738 for the clinical model. The discovery analysis also reported sphingosine-1-phosphate association with ARDS prognosis and an enriched sphingolipid pathway.",
    "field": "medicine",
    "ai_technology": "LASSO and Boruta feature selection; machine-learning prognostic model; targeted PRM validation",
    "organization": "Fudan University; Zhongshan Hospital, Fudan University; Shanghai Jiao Tong University School of Medicine; collaborating Shanghai hospitals",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38956604/",
    "source_type": "paper",
    "publication_date": "2024-07-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s13054-024-05000-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Eight proteins (VCAM1, LDHB, MSN, FLG2, TAGLN2, LMNA, MBL2 and LBP) showed consistency in an independent cohort of 183 patients and were confirmed by targeted parallel-reaction-monitoring mass spectrometry.",
      "unknown": "Observational prognostic associations do not show proteins cause outcomes or that biomarker-guided care improves survival. Discovery n=130 ARDS patients; validation n=183. PRM confirmation is not independent replication by another group; deployment and prospective benefit were not established.",
      "ai_role": "LASSO, fold-change screening and Boruta prioritized candidate 28-day ARDS prognostic proteins; candidates were tested by targeted PRM in an external cohort.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38956604/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-10-integrated-radiogenomic-machine-learning-predicts-neoadjuvant-chemotherapy-response-in-metastatic-ovarian-cancer",
    "title": "Integrated radiogenomic machine learning predicts neoadjuvant chemotherapy response in metastatic ovarian cancer",
    "summary": "The IRON ensemble combined baseline clinical data, CA-125, ctDNA and CT radiomics from primary and metastatic lesions to estimate tumor-volume change during neoadjuvant chemotherapy for high-grade serous ovarian carcinoma. In an independent 42-patient Barts cohort, the integrated model achieved AUC 0.78 for RECIST 1.1 response classification versus 0.47 for a clinical-only model; radiomics-containing scores significantly correlated with volume change in internal holdout. The study also measured spatial heterogeneity, with omental lesions showing better response than pelvic/ovarian disease.",
    "field": "medicine",
    "ai_technology": "IRON ensemble combining elastic net, support-vector regression and random forest",
    "organization": "University of Cambridge; and Cambridge University Hospitals; Barts Health NHS Trust; Medical University of Vienna; collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41467-023-41820-7.pdf",
    "source_type": "paper",
    "publication_date": "2023-10-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-41820-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In an independent 42-patient Barts cohort, the integrated model achieved AUC 0.78 for RECIST 1.1 response classification versus 0.47 for a clinical-only model; radiomics-containing scores significantly correlated with volume change in internal holdout. The model used historical Cambridge and Barts cohorts with baseline and post-treatment imaging, including an external retrospective test set.",
      "unknown": "Model analysis of two prospective observational cohorts (training n=72, internal holdout n=20, external n=42), not a decision trial. External ctDNA unavailable and imputed to training means. Sample size limits generalizability; treatment benefit from model-guided decisions was not tested.",
      "ai_role": "Pretreatment clinical, blood and multi-lesion radiomic features predicted volumetric response to neoadjuvant chemotherapy. Held-out and external evaluation showed radiomics added predictive information beyond clinical variables.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-023-41820-7.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-01-elastic-net-proteomic-risk-modeling-and-multi-cohort-validation-identify-proteins-associated-with-incident-diabetes",
    "title": "Elastic-net proteomic risk modeling and multi-cohort validation identify proteins associated with incident diabetes",
    "summary": "An untargeted analysis of 4,955 plasma proteins in the ARIC cohort identified 140 associations with incident diabetes; 64 were significant in an internal sample and 47 (including 22 reported as novel) were associated in the external Singapore Multi-Ethnic Cohort. Elastic-net Cox modeling selected a 27-protein panel for 20-year risk prediction. In internal validation, adding those proteins to conventional risk factors including fasting glucose and HbA1c increased the C statistic from 0.720 to 0.731. Genetic-instrument analyses supported possible causal roles for SHBG, ATP1B2 and GSTA1, although ATP1B2’s genetic and observational directions differed.",
    "field": "medicine",
    "ai_technology": "Elastic-net Cox regression for protein feature selection and incident-diabetes risk prediction",
    "organization": "Johns Hopkins Bloomberg School of Public Health; Atherosclerosis Risk in Communities study centers; National University of Singapore; New York University Grossman School of Medicine; collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10090896/",
    "source_type": "paper",
    "publication_date": "2023-01-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.2337/dc22-1830",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An untargeted analysis of 4,955 plasma proteins in the ARIC cohort identified 140 associations with incident diabetes; 64 were significant in an internal sample and 47 (including 22 reported as novel) were associated in the external Singapore Multi-Ethnic Cohort.",
      "unknown": "The 47-protein associations were externally tested in Singapore MEC; the elastic-net predictive model itself was internally validated in ARIC, not independently externally validated. Its incremental C statistic over covariates including fasting glucose and HbA1c was modest (0.011; P=0.04), and the protein model overestimated observed diabetes risk across internal-validation deciles. The causal claims are genetic-instrument inferences, not experimental proof; ATP1B2 had opposite directions in genetic-instrument versus observational analyses. Cohort associations do not establish clinical utility.",
      "ai_role": "Elastic net selected a 27-protein combination from candidate proteins to improve 20-year incident-diabetes risk prediction beyond clinical and cardiometabolic covariates. The model was trained in the ARIC discovery subset and internally evaluated in a held-out ARIC subset. Separate Cox association and genetic-instrument analyses reported the externally validated protein associations; the record does not attribute those analyses to machine learning.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10090896/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-10-machine-learning-models-identify-cerebrospinal-fluid-protein-signatures-for-multiple-sclerosis-activity-and-disability-progression",
    "title": "Machine-learning models identify cerebrospinal-fluid protein signatures for multiple-sclerosis activity and disability progression",
    "summary": "Using Olink proteomics of 1,463 proteins in CSF and plasma from two prospective Swedish cohorts, the study trained prediction models for multiple-sclerosis diagnosis, 2-year disease activity and long-term disability worsening. Lower CSF neurofilament light (NfL) predicted absence of disease activity at two years (replication AUC 0.77). A stepwise model using 11 CSF proteins (CXCL13, LTA, FCN2, ICAM3, LY9, SLAMF7, TYMP, CHI3L1, FYB1, TNFRSF1B and NfL) predicted normalized age-related MS severity; subgroup classification AUCs in the replication cohort ranged from 0.88 to 0.90.",
    "field": "medicine",
    "ai_technology": "Stepwise logistic regression for diagnosis and disease activity; stepwise linear regression for long-term disability severity",
    "organization": "Linköping University; University of Skövde; Karolinska Institutet and Karolinska University Hospital; collaborating Swedish institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10616092/",
    "source_type": "paper",
    "publication_date": "2023-10-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-42682-9",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using Olink proteomics of 1,463 proteins in CSF and plasma from two prospective Swedish cohorts, the study trained prediction models for multiple-sclerosis diagnosis, 2-year disease activity and long-term disability worsening.",
      "unknown": "The study included 143 people with early-stage MS and 43 healthy controls across discovery and replication cohorts; the replication MS cohort had 51 participants. The 2-year disease-activity comparison in replication was especially small (45 with activity versus 5 without). The 11-protein severity model predicted observational disability scores rather than treatment benefit. The plasma severity model had only 12 eligible replication samples and performed less convincingly. These are cohort-level results, not a clinically validated diagnostic assay or independent replication by another research group.",
      "ai_role": "The study used statistical differential-expression testing to select disease-associated candidate proteins, then fitted logistic-regression models to predict MS diagnosis and 2-year disease activity and a linear-regression model to predict long-term normalized age-related MS severity. Stepwise feature selection produced the protein combinations; the frozen models were evaluated in a separate prospective cohort.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10616092/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-03-unsupervised-clustering-identifies-three-blood-immune-endophenotypes-associated-with-distinct-multiple-sclerosis-trajectories",
    "title": "Unsupervised clustering identifies three blood immune endophenotypes associated with distinct multiple-sclerosis trajectories",
    "summary": "In two multicenter longitudinal cohorts of early multiple sclerosis (discovery n=309; validation n=232), unsupervised clustering of high-dimensional flow-cytometry and serum-proteomics profiles identified three peripheral-blood immune endophenotypes. Follow-up associated the groups with different trajectories of inflammation versus early structural damage. In the endophenotype-3 subgroup, patients subsequently treated with interferon beta had higher four-year progression and MRI activity than patients receiving other therapies.",
    "field": "medicine",
    "ai_technology": "Unsupervised clustering of high-dimensional flow-cytometry and serum-proteomics profiles (the reviewed abstract does not name the clustering algorithm)",
    "organization": "University of Münster; German Competence Network Multiple Sclerosis; Max Delbrück Center; Charité – Universitätsmedizin Berlin; collaborating German and international centers",
    "source_url": "https://edoc.mdc-berlin.de/id/eprint/24233/",
    "source_type": "paper",
    "publication_date": "2024-03-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/scitranslmed.ade8560",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In two multicenter longitudinal cohorts of early multiple sclerosis (discovery n=309; validation n=232), unsupervised clustering of high-dimensional flow-cytometry and serum-proteomics profiles identified three peripheral-blood immune endophenotypes.",
      "unknown": "The source reports cohort associations, not randomized treatment effects. The higher progression and MRI activity among endophenotype-3 patients treated with interferon beta may be affected by treatment selection and other confounding; it does not establish that interferon beta caused worse outcomes or that the signature can guide care. The validation cohort is an independent patient cohort within the study, not independent replication by another research group. The reviewed abstract does not provide the clustering algorithm name or establish prospective clinical utility.",
      "ai_role": "The researchers combined flow-cytometry immune measurements and serum-protein profiles, then used unsupervised clustering to define three immune endophenotypes in the discovery cohort and assess them in an independent multicenter validation cohort. The clustering generated the patient groups whose longitudinal clinical and paraclinical trajectories were compared; the repository abstract does not specify the algorithm name.",
      "evidence_urls": [
        "https://edoc.mdc-berlin.de/id/eprint/24233/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-lightgbm-feature-selection-identifies-a-four-protein-panel-associated-with-incident-chronic-kidney-disease",
    "title": "LightGBM feature selection identifies a four-protein panel associated with incident chronic kidney disease",
    "summary": "In a prospective UK Biobank cohort of 27,429 participants without baseline chronic kidney disease, the study evaluated 2,920 plasma proteins and found 205 associated with incident CKD after protein-wide Cox regression with Bonferroni correction. LASSO reduced these to 23 candidates and LightGBM feature selection identified EDA2R, GDF15, HAVCR1 and SCARB2 as a four-protein panel. The protein panel alone had an AUC of 0.796 versus 0.805 for demographic variables; adding the proteins to the demographic model raised AUC to 0.830 in the study data.",
    "field": "medicine",
    "ai_technology": "Least absolute shrinkage and selection operator (LASSO) and Light Gradient Boosting Machine (LightGBM) feature selection",
    "organization": "Nanjing Medical University, School of Public Health, Department of Epidemiology and Center for Global Health; UK Biobank",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42968231/",
    "source_type": "paper",
    "publication_date": "2026-11-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.23876/j.krcp.26.161",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a prospective UK Biobank cohort of 27,429 participants without baseline chronic kidney disease, the study evaluated 2,920 plasma proteins and found 205 associated with incident CKD after protein-wide Cox regression with Bonferroni correction. Protein selection and prediction were assessed internally within UK Biobank; prospective cohort collection does not establish prospective evaluation of a previously locked predictor.",
      "unknown": "The abstract reports a prospective UK Biobank cohort and internal analyses; the four-protein model did not outperform the demographic model alone (AUC 0.796 vs 0.805, p=0.14), though combining the proteins with demographics improved AUC to 0.830. The authors state external validation is needed before clinical application. Protein associations and model selection do not establish causal roles or clinical utility.",
      "ai_role": "After protein-wide Cox association screening, LASSO selected 23 candidate proteins and LightGBM-based feature selection identified four proteins for the incident-CKD prediction panel. The study assessed their predictive contribution with demographic covariates and sensitivity analyses; these algorithms selected risk markers rather than demonstrating causal protein effects.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42968231/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-external-validation-reveals-distinct-plasma-proteomic-signatures-across-biopsy-proven-primary-glomerular-diseases",
    "title": "External validation reveals distinct plasma proteomic signatures across biopsy-proven primary glomerular diseases",
    "summary": "Profiling 5,416 plasma proteins in Korean cohorts, the study found distinct signatures across four biopsy-proven primary glomerular diseases. An elastic-net model using 93 proteins differentiated disease subtypes and controls in discovery and external validation cohorts. It achieved AUCs above 0.8 for minimal change disease, membranous nephropathy and IgA nephropathy; the model correctly classified 14 of 15 minimal-change cases and 20 of 32 IgA-nephropathy cases, but only 3 of 14 focal-segmental glomerulosclerosis cases. Functional analysis highlighted hemostasis-related proteins in minimal change disease.",
    "field": "medicine",
    "ai_technology": "Elastic-net-regularized logistic regression using plasma proteomic features",
    "organization": "Seoul National University College of Medicine; and Seoul National University Hospital; KORNERSTONE investigators; collaborating Korean hospitals",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41779447/",
    "source_type": "paper",
    "publication_date": "2026-09-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1681/ASN.0000001054",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Profiling 5,416 plasma proteins in Korean cohorts, the study found distinct signatures across four biopsy-proven primary glomerular diseases.",
      "unknown": "The discovery cohort contained 147 participants and the external validation cohort 85; subtype-level validation samples were small. Classification was strong for minimal change disease, membranous nephropathy and IgA nephropathy but poor for focal segmental glomerulosclerosis (3/14 correctly classified). The findings support cohort-level proteomic differentiation and do not establish a clinically deployable noninvasive replacement for kidney biopsy or causal roles for the selected proteins.",
      "ai_role": "The model used elastic-net regularization to select and combine a 93-protein plasma signature for classifying biopsy-proven primary glomerular disease subtypes and controls. The model was evaluated in an independent external validation cohort. Functional analysis of key proteins was used to characterize subtype-associated pathways, including hemostasis in minimal change disease.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41779447/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-machine-learning-analysis-separates-genetic-from-environmental-drivers-of-stimulated-cytokine-responses-in-healthy-adults",
    "title": "Machine-learning analysis separates genetic from environmental drivers of stimulated cytokine responses in healthy adults",
    "summary": "In 406 healthy GEOCODE cohort participants, whole-blood cytokine responses to Toll-like-receptor (TLR) and T-cell-receptor (TCR) stimulation varied across people, including high- and low-IL-6-response profiles associated with monocyte counts, sex and sampling season. Feature-importance analysis of machine-learning models found TCR-induced cytokine predictions were mainly driven by genetic features, whereas TLR-induced predictions were driven more by environmental and biological features. Combining genetic with biological and environmental data improved prediction by about 0.2 on average, and the reported findings were evaluated in the independent Milieu Intérieur cohort.",
    "field": "biology",
    "ai_technology": "Machine-learning models including random forest, gradient-boosted trees, artificial neural networks and elastic net; random forest feature-importance analysis",
    "organization": "Université libre de Bruxelles; Hôpital Erasme; Milieu Intérieur cohort collaborators",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41735307/",
    "source_type": "paper",
    "publication_date": "2026-02-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41540-026-00671-w",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In 406 healthy GEOCODE cohort participants, whole-blood cytokine responses to Toll-like-receptor (TLR) and T-cell-receptor (TCR) stimulation varied across people, including high- and low-IL-6-response profiles associated with monocyte counts, sex and sampling season.",
      "unknown": "The discovery cohort was a selected healthy-adult sample (406 participants), and the correlations for TLR-induced response prediction were modest (average 0.27; TCR-induced 0.53). Genetic-only models may overestimate predictive accuracy, as the authors report. Feature importance identifies predictive contribution, not causal biological effects; results do not establish disease-specific immune mechanisms or treatment response.",
      "ai_role": "Models predicted measured cytokine levels following TLR or TCR stimulation from genetic, biological and environmental participant features. Random-forest permutation importance identified which input categories contributed to predictions; the study also compared predictions across cross-validation and a separate Milieu Intérieur cohort. Feature importance is associational and does not establish that a feature causally regulates cytokine production.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41735307/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-04-stability-based-machine-learning-selects-il-8-and-ccl3-as-a-two-protein-serum-signature-for-early-silicosis",
    "title": "Stability-based machine learning selects IL-8 and CCL3 as a two-protein serum signature for early silicosis",
    "summary": "A stability-based feature-selection workflow combining LASSO, random forest and SVM recursive feature elimination identified serum IL-8 and CCL3 as a two-protein signature distinguishing Stage I silicosis from controls. The study used a discovery cohort of 57 cases and 57 matched controls, then assessed the signature in an independent, unmatched cohort of 40 cases and 40 controls. The article reports validation AUC 0.973, specificity 95.0% and sensitivity 77.5%; both proteins were lower in silicosis cases.",
    "field": "medicine",
    "ai_technology": "Stability-based ensemble feature selection using LASSO, random forest and SVM recursive feature elimination, followed by logistic regression",
    "organization": "Nanjing Medical University; Jiangsu Provincial Center for Disease Control and Prevention; Southeast University; China University of Mining and Technology",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41937675/",
    "source_type": "paper",
    "publication_date": "2026-04-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jproteome.5c00987",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study used a discovery cohort of 57 cases and 57 matched controls, then assessed the signature in an independent, unmatched cohort of 40 cases and 40 controls.",
      "unknown": "The discovery cohort was small (57 cases/57 matched controls); the validation cohort was also small and unmatched (40/40), with 77.5% sensitivity despite high specificity. This is retrospective biomarker validation and does not establish population screening performance, clinical utility, or causal involvement of IL-8 or CCL3 in silicosis.",
      "ai_role": "Across 100 iterations, the machine-learning feature-selection framework selected a stable two-protein signature from targeted serum proteomics. A logistic-regression classifier used those selected features and was evaluated by cross-validation and an independent validation cohort. The study reports reduced IL-8 and CCL3 as the signature components.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41937675/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-04-computer-vision-and-metabolomics-fusion-links-paired-serumurine-flux-mismatch-to-early-diabetic-kidney-disease",
    "title": "Computer-vision and metabolomics fusion links paired serum–urine flux mismatch to early diabetic kidney disease",
    "summary": "In a two-center cohort of 364 participants with type 2 diabetes and controls, targeted serum/urine metabolite measurements found a high-serum/low-urine pattern for indoxyl sulfate and related metabolites in early diabetic kidney disease (DKD), consistent with reduced tubular secretion. Computer vision quantified urine foam stability and chromaticity; Random Forest recursive feature elimination selected these image features together with metabolite ratios and clinical variables for the FluxPro-DKD classifier. It distinguished DKD from diabetes without kidney disease with AUC 0.90 in discovery and 0.85 in a geographically distinct external cohort of 82.",
    "field": "medicine",
    "ai_technology": "Computer-vision urine phenotyping and Random Forest recursive feature elimination (RF-RFE) for multimodal feature selection",
    "organization": "Beijing University of Chinese Medicine; Dongfang Hospital; Third Affiliated Hospital of Beijing University of Chinese Medicine; Tianjin Institute of Nephropathy",
    "source_url": "https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2026.1802447/full",
    "source_type": "paper",
    "publication_date": "2026-04-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3389/fendo.2026.1802447",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a two-center cohort of 364 participants with type 2 diabetes and controls, targeted serum/urine metabolite measurements found a high-serum/low-urine pattern for indoxyl sulfate and related metabolites in early diabetic kidney disease (DKD), consistent with reduced tubular secretion.",
      "unknown": "The external cohort validation was cross-sectional; the paper says longitudinal follow-up for hard kidney outcomes is ongoing. Its projected three-year event rates are theoretical Monte Carlo simulations based on cross-sectional data and are not observed outcomes. The study did not measure proximal-tubule transporter expression or compare directly against KIM-1 or NGAL, and generalization beyond the two Chinese centers and population requires further validation. Metabolite flux supports, but does not prove, tubular transport dysfunction.",
      "ai_role": "Computer vision quantified urine foam half-life and chromaticity from standardized sample images. RF-RFE selected eight predictors, including digital urine features and serum-to-urine metabolite ratios, for the fusion classifier distinguishing early DKD from type 2 diabetes without DKD. The paper reports external cohort evaluation. The biochemical interpretation of flux mismatch as reduced tubular secretory capacity is based on measured metabolite ratios and literature-supported interpretation; transporter expression was not measured.",
      "evidence_urls": [
        "https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2026.1802447/full"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-sonar-identifies-3utr-sequence-features-that-raise-t-cell-receptor-protein-expression-and-cd8-cell-killing",
    "title": "SONAR identifies 3′UTR sequence features that raise T-cell receptor protein expression and CD8-cell killing",
    "summary": "The XGBoost-based SONAR pipeline modeled protein abundance from thousands of sequence features across human immune cells and cell lines, identifying cell-type- and activation-dependent regulatory patterns. Reporter assays found that CT-rich 3′UTR motifs increased protein output while AG-rich motifs were generally repressive. SONAR-informed synthetic 3′UTRs raised CDK4-specific T-cell-receptor protein without increasing its mRNA; engineered CD8+ T cells carrying the CT-rich constructs killed CDK4-expressing melanoma cells more effectively in a coculture assay.",
    "field": "biology",
    "ai_technology": "SONAR, an XGBoost machine-learning pipeline using thousands of sequence features to predict protein abundance",
    "organization": "Sanquin Blood Supply Foundation; University of Amsterdam; Amsterdam UMC; Oncode Institute",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12285716/",
    "source_type": "paper",
    "publication_date": "2025-07-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.ads0510",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Reporter assays found that CT-rich 3′UTR motifs increased protein output while AG-rich motifs were generally repressive.",
      "unknown": "The target-cell killing experiment used engineered CD8+ cells in vitro and three donors; no animal or patient treatment outcome was tested. Protein abundance prediction used limited cell and immune-subset datasets, and model feature importance is not itself causal evidence. The assay supports effects for the tested synthetic 3′UTRs and CDK4-specific TCR in the reported system, not all proteins or therapeutic settings.",
      "ai_role": "SONAR learned sequence-feature patterns associated with protein abundance in immune cells, including cell-type- and activation-dependent feature importance. Its outputs guided selection of 3′UTR motifs and synthetic sequence design; reporter assays and a CDK4-specific TCR construct then tested their effects on protein output and CD8+ T-cell cytotoxicity. The model generated design hypotheses; the downstream experiments established the measured motif and T-cell results.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12285716/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-few-shot-protein-language-model-design-yields-high-output-bicyclogermacrene-synthase-variants",
    "title": "Few-shot protein-language-model design yields high-output bicyclogermacrene synthase variants",
    "summary": "After experimentally screening activity-enhancing single mutants of LdTPS5, the researchers used PIFS-PLM, a few-shot model that selects features from protein-language-model embeddings using active-site and conserved-residue information, to propose mutation combinations. All 10 tested model-predicted combinations were active; nine produced more than 100 mg/L bicyclogermacrene in yeast, and the top variant produced 213.59 mg/L versus 2.93 mg/L for wild type (about 73-fold higher yield). A related selectivity-optimized combination increased bicyclogermacrene selectivity by 15%.",
    "field": "biology",
    "ai_technology": "PIFS-PLM (Physics-Inspired Feature Selection of Protein Language Models), using ESM-2 embeddings with a few-shot mutation-combination classifier",
    "organization": "Tsinghua University; Tianjin University; Inner Mongolia University; Zhejiang Institute of Tianjin University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13502649/",
    "source_type": "paper",
    "publication_date": "2026-06-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/anie.7768514",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "After experimentally screening activity-enhancing single mutants of LdTPS5, the researchers used PIFS-PLM, a few-shot model that selects features from protein-language-model embeddings using active-site and conserved-residue information, to propose mutation combinations.",
      "unknown": "The 73-fold result is product yield in the reported yeast system, reflecting effects of enzyme activity and expression; it is not a 73-fold increase in catalytic rate. The selectivity improvement was 15%. The model relies on a defined active site/conserved-site set and a small, enzyme-specific experimental dataset. Product profiles differed between cell-free extracts and yeast; these results do not establish universal performance across enzymes or production settings.",
      "ai_role": "After single substitutions were screened and activity-enhancing mutations experimentally validated, PIFS-PLM used 60–100 measured combinations to learn from selected ESM-2 features at active-site and conserved residues and predict beneficial combinations. Ten predicted LdTPS5 variants were tested; all were active, and the top variants substantially increased measured yeast product yield. The model also guided a selectivity optimization test. Initial single-mutation nomination used MM/GBSA or sequence conservation analysis, separate from PIFS-PLM combinatorial predictions.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13502649/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-tree-based-machine-learning-nominates-parabacteroides-merdae-and-n-acetyl-d-mannosamine-as-ulcerative-colitis-alleviators-in-a-mouse-model",
    "title": "Tree-based machine learning nominates Parabacteroides merdae and N-acetyl-D-mannosamine as ulcerative-colitis alleviators in a mouse model",
    "summary": "The researchers combined personalized gut-microbiome community metabolic models from human iHMP data with tree-based machine-learning feature selection to identify microbial strains and metabolites associated with ulcerative-colitis samples. Parabacteroides merdae ATCC 43,184 and N-acetyl-D-mannosamine were among the most predictive features; the strain was also the main modeled microbial contributor to the metabolite. In a DSS-induced mouse colitis experiment, oral administration of the bacterium or metabolite improved reported colitis measures.",
    "field": "biology",
    "ai_technology": "Tree-based supervised classifiers including random forest, gradient-boosted regression trees, XGBoost, XGBoost-random forest and LightGBM, combined with individualized genome-scale microbiome community models",
    "organization": "Jiangnan University; Wuxi Translational Medicine Research Center; Jiangsu Translational Medicine Research Institute Wuxi Branch; Wuxi People’s Hospital",
    "source_url": "https://www.tandfonline.com/doi/full/10.1080/19490976.2024.2336877",
    "source_type": "paper",
    "publication_date": "2024-04-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1080/19490976.2024.2336877",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In a DSS-induced mouse colitis experiment, oral administration of the bacterium or metabolite improved reported colitis measures.",
      "unknown": "The human data are observational microbiome samples, while intervention was tested in a DSS-induced mouse model, not in people. The animal groups contained eight mice each; findings do not establish treatment efficacy or safety in human ulcerative colitis. Metabolite-production associations came from community-model simulations and were not direct evidence of individual strain production in each person. The model provides candidate prioritization, not a causal human mechanism.",
      "ai_role": "Personalized community metabolic models estimated strain-level contributions to secreted metabolites in human iHMP samples. Tree-based classifiers selected discriminative strain and metabolite features; XGBRF feature attribution ranked P. merdae ATCC 43,184 and N-acetyl-D-mannosamine highly, motivating the mouse validation. The AI analysis nominated these targets; it did not establish human therapeutic efficacy.",
      "evidence_urls": [
        "https://www.tandfonline.com/doi/full/10.1080/19490976.2024.2336877"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-xgboost-guided-splice-switching-identifies-nedd4l-exon-13-as-a-tnbc-cell-vulnerability",
    "title": "XGBoost-guided splice switching identifies NEDD4L exon 13 as a TNBC cell vulnerability",
    "summary": "SpliceCore’s XGBoost models used splicing-factor RNA-binding and spliceosome information to nominate NEDD4L exon 13 (NEDD4Le13) and the SSO-0205 binding site. In triple-negative breast-cancer MDA-MB-231 cells, SSO-0205 increased exon inclusion (67% versus 15% control), reduced proliferation and decreased migration by 40–50%; the effects were not observed in the non-tumorigenic MCF10A comparator under the reported assays.",
    "field": "medicine",
    "ai_technology": "SpliceCore XGBoost models with SHAP and out-of-bag feature analysis",
    "organization": "Envisagenics, Inc.",
    "source_url": "https://link.springer.com/article/10.1038/s44320-024-00034-9",
    "source_type": "paper",
    "publication_date": "2024-04-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44320-024-00034-9",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In triple-negative breast-cancer MDA-MB-231 cells, SSO-0205 increased exon inclusion (67% versus 15% control), reduced proliferation and decreased migration by 40–50%; the effects were not observed in the non-tumorigenic MCF10A comparator under the reported assays.",
      "unknown": "Experiments were in vitro, including three breast-cell lines and limited biological replicates; the migration assay used two to three replicates. These results do not establish tumor control in animals, patient benefit, safety or clinical efficacy. The source reports the target as novel in the studied context; this is not evidence that the SSO is a treatment.",
      "ai_role": "The models prospectively nominated the NEDD4L exon-13 splice event, predicted functional SSO binding sites, and prioritized implicated splicing factors. Researchers synthesized selected oligonucleotides and tested splicing, proliferation and migration in cell lines.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1038/s44320-024-00034-9"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-machine-learning-multi-omics-links-gut-metabolites-to-gpcr-targets-and-tau-changes-in-alzheimers-patient-derived-neurons",
    "title": "Machine-learning multi-omics links gut metabolites to GPCR targets and tau changes in Alzheimer’s patient-derived neurons",
    "summary": "A machine-learning and multi-omics framework screened 1.09 million pairs across 408 human GPCRs and 335 gut microbial metabolites. It prioritized orphan receptor GPR84 and the metabolite triacanthine, which activated GPR84 in a cell assay. In Alzheimer’s patient-derived iPSC neurons, agmatine reduced C3AR and tau markers, while phenethylamine reduced phosphorylated and total tau in the reported experiments.",
    "field": "biology",
    "ai_technology": "Extra Trees machine-learning model integrated with molecular docking, multi-omics and Mendelian-randomization analyses",
    "organization": "Cleveland Clinic; Case Western Reserve University; Frederick National Laboratory for Cancer Research; Tel Aviv University and collaborating institutions",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S221112472400456X",
    "source_type": "paper",
    "publication_date": "2024-04-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.celrep.2024.114128",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "It prioritized orphan receptor GPR84 and the metabolite triacanthine, which activated GPR84 in a cell assay.",
      "unknown": "The metabolite–receptor screen is computational and many pairs remain predictions. Receptor activation was tested in a cell assay and tau-marker effects in patient-derived neurons; these experiments do not demonstrate clinical benefit, in-vivo efficacy or that altered tau markers translate to disease modification. Mendelian-randomization results and model scores are not proof of causality.",
      "ai_role": "The ML model ranked predicted metabolite–GPCR associations over a large search space; multi-omics and genetic analyses prioritized Alzheimer’s-related receptors and metabolites. These predictions guided follow-up assays, including triacanthine activation of GPR84 and testing agmatine and phenethylamine in patient-derived neurons.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S221112472400456X"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-07-machine-learning-selection-identifies-a-six-metabolite-signature-for-one-year-cardiovascular-events-after-pci-in-type-2-diabetes",
    "title": "Machine-learning selection identifies a six-metabolite signature for one-year cardiovascular events after PCI in type 2 diabetes",
    "summary": "VSURF feature selection prioritized six serum metabolites—ADPR, 1-methylnicotinamide, D-ribose 5-phosphate, nicotinamide, L-tryptophan and phosphatidylcholine PC(36:2)—for one-year major adverse cardiovascular events (MACEs) after PCI in patients with type 2 diabetes. Random-forest and XGBoost models were tested in a nested case-control study and an independent prospective cohort of 301 patients; in the external cohort, reported AUCs were 0.92 and 0.94 for MACEs, respectively. The study also linked NAD+ biosynthesis perturbation to altered bioenergetics, inflammation and proliferation in high-glucose-cultured human aortic smooth-muscle cells.",
    "field": "medicine",
    "ai_technology": "VSURF random-forest variable selection; random forest and XGBoost classifiers",
    "organization": "The First Affiliated Hospital of Zhengzhou University; Zhongnan Hospital of Wuhan University",
    "source_url": "https://link.springer.com/article/10.1186/s12933-022-01561-1",
    "source_type": "paper",
    "publication_date": "2022-07-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s12933-022-01561-1",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Random-forest and XGBoost models were tested in a nested case-control study and an independent prospective cohort of 301 patients; in the external cohort, reported AUCs were 0.92 and 0.94 for MACEs, respectively. For the prognostic panel, the external cohort was collected before the development cohort. Separate cell experiments support the reported pathway mechanism, not prospective clinical performance of the panel.",
      "unknown": "The discovery component was a matched nested case-control sample (108 cases and 108 controls) from a single center; external validation was in a separate Chinese cohort of 301 patients, so transportability to other populations is unknown. AUCs describe retrospective/prospective cohort validation, not demonstrated clinical utility or improved patient outcomes. HASMC experiments do not prove the metabolites cause clinical events, and the paper states further mechanistic work is needed.",
      "ai_role": "VSURF selected a six-metabolite feature set from serum metabolomic profiles. Random-forest and XGBoost models used that panel to predict one-year post-PCI outcomes; targeted metabolite testing evaluated the panel in a separate prospective cohort.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s12933-022-01561-1"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-07-machine-learning-prioritizes-a-protective-arabidopsis-leaf-fungus-that-suppresses-oomycete-infection",
    "title": "Machine learning prioritizes a protective Arabidopsis leaf fungus that suppresses oomycete infection",
    "summary": "Four classifiers (random forest, support-vector machine, logistic regression and multilayer perceptron) used six years of natural Arabidopsis leaf-microbiome data to prioritize taxa associated with Albugo infection status. Follow-up plant assays found that selected health-associated microbes reduced infection; Cystofilobasidium macerans OTU126 gave the strongest individual protection, reducing Albugo infection by an average of 73%. In a 15-member synthetic leaf community, adding this fungus increased the reported infection reduction to 40.25%, compared with 19.36% for the community alone.",
    "field": "biology",
    "ai_technology": "Random forest, support-vector machine, logistic regression and multilayer perceptron classifiers with cross-validated feature selection",
    "organization": "University of Tübingen; Université Claude Bernard Lyon 1 and collaborating institutions",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42521687/",
    "source_type": "paper",
    "publication_date": "2026-07-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-75789-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Follow-up plant assays found that selected health-associated microbes reduced infection; Cystofilobasidium macerans OTU126 gave the strongest individual protection, reducing Albugo infection by an average of 73%.",
      "unknown": "This is a controlled Arabidopsis–Albugo plant-pathogen model, not a field crop efficacy trial. ML features are associations with infection status; they do not establish the microbes caused naturally observed health status. Isolates were matched to predicted OTUs using marker sequences, which the authors note do not fully establish strain-level functional identity. Genomic terpene-cluster hypotheses were not directly tested. No commercial or field-ready biocontrol performance is established.",
      "ai_role": "Models classified infected versus uninfected plants from microbial community profiles and prioritized disease-associated and health-associated taxa for follow-up. Researchers then tested selected cultured isolates in planta and tested Cystofilobasidium in a synthetic community.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42521687/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-12-machine-learning-identifies-bacterial-strains-and-combinations-that-reduce-pathogen-colonization-in-arabidopsis",
    "title": "Machine learning identifies bacterial strains and combinations that reduce pathogen colonization in Arabidopsis",
    "summary": "Random forest and elastic-net models analyzed 136 synthetic communities, each composed of five bacterial strains, and identified strain identity as the strongest predictor of protection against Pseudomonas syringae DC3000 in Arabidopsis. Independent test communities confirmed predictive performance, and follow-up plant assays showed that Acidovorax Leaf76, Rhizobium Leaf68 and Pseudomonas Leaf15 each reduced pathogen colonization by about two orders of magnitude versus axenic controls; a Rhizobium Leaf371–Arthrobacter Leaf337 combination also reduced colonization.",
    "field": "biology",
    "ai_technology": "Random forest and elastic-net generalized linear models (GLMNet)",
    "organization": "ETH Zurich; Institute of Microbiology",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38042924/",
    "source_type": "paper",
    "publication_date": "2023-12-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-43793-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Independent test communities confirmed predictive performance, and follow-up plant assays showed that Acidovorax Leaf76, Rhizobium Leaf68 and Pseudomonas Leaf15 each reduced pathogen colonization by about two orders of magnitude versus axenic controls; a Rhizobium Leaf371–Arthrobacter Leaf337 combination also reduced colonization.",
      "unknown": "The experiments used gnotobiotic Arabidopsis thaliana and P. syringae DC3000, not field crops. Validation was in the same study, with 24 plants per treatment in the key individual-strain validation assay. The work establishes protection in this model system; it does not establish the underlying mechanisms, efficacy in other plants or field performance.",
      "ai_role": "Models predicted pathogen outcomes from strain presence or abundance across the 136 screened communities and ranked strains by feature importance. Researchers tested the leading strains and combinations in additional plant experiments, validating the model-prioritized protection results.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38042924/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-11-machine-learning-guided-pbmc-proteomics-identifies-sle-diagnosis-and-activity-biomarker-panels",
    "title": "Machine-learning-guided PBMC proteomics identifies SLE diagnosis and activity biomarker panels",
    "summary": "A random-forest and OPLS-DA prioritization pipeline applied to peripheral-blood mononuclear-cell proteomics identified candidate protein combinations for distinguishing systemic lupus erythematosus (SLE) from healthy and rheumatoid-arthritis controls and for separating active from inactive SLE. ELISA in a separate 80-person cohort supported a six-protein diagnosis panel (IFIT3, MX1, TOMM40, STAT1, STAT2 and OAS3; AUC 0.723 versus healthy controls and 0.815 versus rheumatoid arthritis) and a nine-protein activity panel (AUC 0.990). Single-cell RNA-sequencing analysis mapped several associated interferon-signature proteins to specific immune-cell populations.",
    "field": "medicine",
    "ai_technology": "POC-SLE: random-forest feature ranking with 1,000 bootstrap iterations, combined with OPLS-DA variable-importance selection",
    "organization": "Jinan University; Shenzhen People’s Hospital; collaborating institutions",
    "source_url": "https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2022.969509/full",
    "source_type": "paper",
    "publication_date": "2022-11-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3389/fimmu.2022.969509",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "ELISA in a separate 80-person cohort supported a six-protein diagnosis panel (IFIT3, MX1, TOMM40, STAT1, STAT2 and OAS3; AUC 0.723 versus healthy controls and 0.815 versus rheumatoid arthritis) and a nine-protein activity panel (AUC 0.990).",
      "unknown": "The study is cross-sectional and the ELISA validation cohort contained 80 people from the same clinical setting; it does not establish longitudinal flare prediction, treatment response, causal mechanisms or clinical utility. The six-protein panel had moderate discrimination versus healthy controls (AUC 0.723) and better but still cohort-specific discrimination versus rheumatoid arthritis (AUC 0.815). The activity-panel AUC of 0.990 was reported for active versus inactive groups in this validation cohort and needs broader independent validation. Authors note limited sample size/ethnic diversity and need for mechanistic study.",
      "ai_role": "The random forest ranked differentially expressed proteins from discovery-cohort PBMC proteomics; its selected proteins were intersected with OPLS-DA features to construct candidate diagnosis and activity panels. Researchers evaluated those panels with ELISA in a separate cohort and used single-cell transcriptomic data to examine immune-cell sources.",
      "evidence_urls": [
        "https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2022.969509/full"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-12-integrated-proteomics-and-metabolomics-with-lasso-nominate-a-four-analyte-early-iga-nephropathy-plasma-panel",
    "title": "Integrated proteomics and metabolomics with LASSO nominate a four-analyte early IgA-nephropathy plasma panel",
    "summary": "A study of biopsy-confirmed IgA nephropathy (IgAN) integrated plasma proteomics and untargeted metabolomics and used LASSO logistic regression to select PRKAR2A, IL6ST, SOS1 and palmitoleic acid as candidate discriminators from healthy controls. The panel had AUC 0.994 in the training set and 0.977 in the test split; ELISA assays in a separate 9-case/10-control sample set yielded an AUC of 0.91. Proteomic and metabolomic analyses also indicated complement/immune activation and altered energy and amino-acid metabolism in the studied cases.",
    "field": "medicine",
    "ai_technology": "LASSO logistic regression (glmnet), with tenfold cross-validation and additional random-forest/SVM model checks",
    "organization": "Zhejiang Provincial People’s Hospital; and collaborating institutions",
    "source_url": "https://link.springer.com/article/10.1186/s12014-022-09387-5",
    "source_type": "paper",
    "publication_date": "2022-12-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s12014-022-09387-5",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A study of biopsy-confirmed IgA nephropathy (IgAN) integrated plasma proteomics and untargeted metabolomics and used LASSO logistic regression to select PRKAR2A, IL6ST, SOS1 and palmitoleic acid as candidate discriminators from healthy controls.",
      "unknown": "This is an early candidate biomarker result, not a validated clinical test or biopsy replacement. The discovery analysis included 33 biopsy-confirmed IgAN cases and 20 healthy controls; the separate ELISA group contained only 9 IgAN cases and 10 controls from the same hospital. The test split is small, and the study did not test specificity against other kidney diseases or demonstrate prospective/clinical utility. The proposed biomarker biology remains speculative.",
      "ai_role": "LASSO feature selection on integrated proteomic and metabolomic measurements nominated the four-analyte panel; the study evaluated its classification in a held-out split and measured the selected analytes by ELISA in an independent sample batch.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s12014-022-09387-5"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-machine-learning-proteomics-prioritizes-four-plasma-markers-associated-with-ankylosing-spondylitis-activity",
    "title": "Machine-learning proteomics prioritizes four plasma markers associated with ankylosing-spondylitis activity",
    "summary": "A plasma-proteomics study of 48 active ankylosing-spondylitis (AS) patients, 21 stable-stage patients and 35 healthy controls combined random-forest and OPLS-DA scores to prioritize candidate markers. SAA1, FERMT3, ILK and TLN1 were elevated in active-stage AS and returned toward control levels in stable-stage patients; ELISA in a separate cohort (27 active, 28 stable, 24 controls) reproduced these patterns. A combined ELISA model distinguished active AS from healthy controls with AUC 0.932. FERMT3, ILK and TLN1 are integrin-pathway proteins.",
    "field": "medicine",
    "ai_technology": "Machine-learning biomarker scoring matrix combining random forest and OPLS-DA",
    "organization": "Guang’anmen Hospital, China Academy of Chinese Medical Sciences; Beijing Institute of Lifeomics; collaborating institutions",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40436140/",
    "source_type": "paper",
    "publication_date": "2025-05-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.jare.2025.05.052",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "SAA1, FERMT3, ILK and TLN1 were elevated in active-stage AS and returned toward control levels in stable-stage patients; ELISA in a separate cohort (27 active, 28 stable, 24 controls) reproduced these patterns.",
      "unknown": "This is an observational, cross-sectional plasma-biomarker study; it does not establish causal integrin-pathway activity, prognosis over time, treatment response, or clinical utility. The independent ELISA cohort had 79 participants and was drawn from the same clinical setting. The combined AUC 0.932 applies to active AS versus healthy controls; discrimination against other inflammatory diseases was not demonstrated. SAA1 has been previously reported as an AS-associated marker; the AI-supported novelty is the combined panel and the integrin-related proteins, not discovery of SAA1 itself.",
      "ai_role": "Random-forest classification accuracy/Gini importance and OPLS-DA variable-importance scores were combined with proteomic fold-change and significance terms to rank candidate proteins. The team then measured the four prioritized proteins by ELISA in an independent cohort.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40436140/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-12-machine-learning-guided-electrochemical-screening-isolates-five-new-reaction-products",
    "title": "Machine-learning-guided electrochemical screening isolates five new reaction products",
    "summary": "A machine-learning-guided, chemist-in-the-loop screen predicted electrochemical reactivity for 38,865 potential partners with 1,4-dicyanobenzene (DCB). Automated tests of 20 prioritized, previously untested partners yielded DCB conversion in 16; five produced an isolable major product judged of synthetic or mechanistic interest. The reported examples include a silyl-enol-ether coupling and electrochemical benzylic arylation of benzyltrimethylsilane with DCB, plus three mechanistically unexpected products.",
    "field": "chemistry",
    "ai_technology": "Machine-learning-guided reaction-competency classifier using a DFT-informed molecular representation",
    "organization": "Massachusetts Institute of Technology; collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9756344/",
    "source_type": "paper",
    "publication_date": "2022-12-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/jacs.2c08997",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Automated tests of 20 prioritized, previously untested partners yielded DCB conversion in 16; five produced an isolable major product judged of synthetic or mechanistic interest.",
      "unknown": "This is an early reaction-discovery demonstration under one fixed DCB electrolysis system, not a general reaction predictor or optimized synthetic platform. Seven of the 16 DCB-converting selected reactions gave complex mixtures without an appreciable single isolated product; five yielded isolable major products, reported at low yields and without method optimization. DCB conversion alone does not establish a product identity, and the three mechanistically ambiguous outcomes require further study.",
      "ai_role": "The model screened a 38,865-member virtual set for reaction competency and, with chemist selection, prioritized 20 new partners for automated electrolysis. The experiments measured DCB conversion, then product analysis isolated and characterized five major products from the selected set.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9756344/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-machine-learning-guided-photoredox-design-experimentally-enables-three-transformations",
    "title": "Machine-learning-guided photoredox design experimentally enables three transformations",
    "summary": "A workflow combining a variational-autoencoder latent space of 11,069 literature photoredox reactions with random-forest catalyst/reagent recommendations was used to propose three transformations. Experiments synthesized a previously inaccessible substituted dihydroquinolinone 43 using fac-Ir(ppy)3, a related oxidative cross-coupling product 46 in 64% yield, and heterocycle product 49 in 74% yield. The paper presents these as experimentally validated transformations identified through model-assisted precedent retrieval and condition selection.",
    "field": "chemistry",
    "ai_technology": "Variational autoencoder for reaction-space navigation and random-forest multilabel catalyst/reagent recommendation",
    "organization": "Wuhan University; HOURS Inc.; collaborating institutions",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42953598/",
    "source_type": "paper",
    "publication_date": "2026-10-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/jacsau.6c00471",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experiments synthesized a previously inaccessible substituted dihydroquinolinone 43 using fac-Ir(ppy)3, a related oxidative cross-coupling product 46 in 64% yield, and heterocycle product 49 in 74% yield.",
      "unknown": "The work is a human-guided photoredox design workflow, not autonomous synthesis. The three examples were selected by researchers and use a literature-trained model; the paper establishes feasibility for these tested transformations, not general transfer or broad yield improvement. Reported 64% and 74% yields apply only to the specific demonstrated products; further optimization and substrate-scope work are needed.",
      "ai_role": "The VAE organized 11,069 literature-derived reactions to retrieve traceable precedents; random-forest models recommended catalyst and reagent classes for proposed reactant-product pairs. Chemists used the model-suggested conditions to test three transformations, including two modifications beyond the reported precedent and one previously undocumented photoredox synthesis, then measured isolated product yields.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42953598/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-06-cross-tissue-machine-learning-prioritizes-a-13-gene-rheumatoid-arthritis-activity-score",
    "title": "Cross-tissue machine learning prioritizes a 13-gene rheumatoid-arthritis activity score",
    "summary": "A machine-learning feature-selection pipeline integrated public whole-blood (1,885 samples) and synovial-tissue (284 samples) transcriptomic datasets and selected a 13-gene rheumatoid-arthritis score. Across five independent transcriptomic datasets, the score tracked DAS28 disease activity, distinguished rheumatoid arthritis from osteoarthritis, separated polyarticular juvenile idiopathic arthritis from healthy controls, and was associated with treatment effects. In an independent immunoblot cohort, two of six tested proteins, TNFAIP6/TSG-6 and HSP90AB1/HSP90, were confirmed.",
    "field": "medicine",
    "ai_technology": "Cross-tissue machine-learning feature-selection pipeline for transcriptomic data",
    "organization": "University of California, San Francisco; Oregon Health & Science University",
    "source_url": "https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2021.638066/pdf",
    "source_type": "paper",
    "publication_date": "2021-06-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3389/fimmu.2021.638066",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In an independent immunoblot cohort, two of six tested proteins, TNFAIP6/TSG-6 and HSP90AB1/HSP90, were confirmed.",
      "unknown": "The transcriptomic inputs were previously published datasets and the reported clinical associations are observational; the study does not establish a deployable diagnostic test or causal disease mechanism. The independent protein follow-up confirmed two of six proteins tested, not the entire 13-gene panel, and validation in larger prospective cohorts is needed. Treatment-effect association should not be interpreted as prospective prediction of individual response.",
      "ai_role": "The pipeline selected a compact 13-gene signature from public blood and synovial expression datasets; the study evaluated its association with disease activity and diagnostic distinctions across five independent transcriptomic datasets, then tested six selected proteins by immunoblot in a separate patient cohort.",
      "evidence_urls": [
        "https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2021.638066/pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-01-random-forest-metabolite-profiling-reduces-false-positive-newborn-screens-for-four-inherited-disorders",
    "title": "Random-forest metabolite profiling reduces false-positive newborn screens for four inherited disorders",
    "summary": "A 121-analyte LC–MS/MS panel combined established markers, untargeted-metabolomics features and chromatographically resolved isobars with random-forest classification of California newborn-screen dried blood spots. In 883 archived infant samples, the study reports that the second-tier approach maintained 100% sensitivity while reducing false positives by 83% for glutaric acidemia type I, 84% for methylmalonic acidemia, 100% for ornithine transcarbamylase deficiency and 51% for very-long-chain acyl-CoA dehydrogenase deficiency. Several long-chain acylcarnitine and hydroxy-acylcarnitine isobar features contributed to classification, although their chemical identities were unresolved.",
    "field": "medicine",
    "ai_technology": "Random Forest classifier trained on targeted LC–MS/MS metabolite profiles",
    "organization": "Stanford Health Care; Yale University; California Department of Public Health",
    "source_url": "https://onlinelibrary.wiley.com/doi/full/10.1002/jimd.12591",
    "source_type": "paper",
    "publication_date": "2023-01-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/jimd.12591",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In 883 archived infant samples, the study reports that the second-tier approach maintained 100% sensitivity while reducing false positives by 83% for glutaric acidemia type I, 84% for methylmalonic acidemia, 100% for ornithine transcarbamylase deficiency and 51% for very-long-chain acyl-CoA dehydrogenase deficiency.",
      "unknown": "This was a proof-of-principle second-tier simulation on residual archived California newborn-screen specimens, not a prospective implementation study. The validation sample was not a fully independent external set, and archived-sample degradation and processing batch effects limited biomarker discovery; some discriminating isobar identities and biological significance remain unknown. The reported sensitivity and false-positive reductions are cohort-specific and do not establish improved health outcomes or routine clinical utility.",
      "ai_role": "The classifier modeled multianalyte profiles from true-positive and false-positive screen-positive samples to distinguish the groups for four inherited metabolic disorders. Feature-importance analysis showed that complex profiles, including unresolved long-chain acylcarnitine/hydroxy-acylcarnitine isobars, contributed to classification; the model's second-tier performance was assessed against known case status.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/full/10.1002/jimd.12591"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-12-3d-hichat-predicts-embryonic-enhancer-hubs-validated-by-crispr-interference",
    "title": "3D-HiChAT predicts embryonic enhancer hubs validated by CRISPR interference",
    "summary": "A random-forest model integrating ATAC-seq, H3K27ac ChIP-seq and HiChIP data predicted enhancer effects across mouse trophoblast stem cells, embryonic stem cells and extraembryonic endoderm cells. CRISPRi tests supported a shared Enh14 hub regulating both Tfcp2l1 and Gli2 in embryonic stem cells, and an Enh4 hub linked to seven genes in extraembryonic endoderm; four of those seven showed expression reductions as predicted. Across 40 tested enhancer-promoter pairs, experiments yielded 12 true positives and 13 true negatives, with overall prediction accuracy 0.71.",
    "field": "biology",
    "ai_technology": "3D-HiChAT, an optimized Random Forest model combining chromatin accessibility, H3K27ac and 3D HiChIP connectivity features",
    "organization": "Weill Cornell Medicine; New York Genome Center; collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10897904/",
    "source_type": "paper",
    "publication_date": "2023-12-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41594-023-01130-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "CRISPRi tests supported a shared Enh14 hub regulating both Tfcp2l1 and Gli2 in embryonic stem cells, and an Enh4 hub linked to seven genes in extraembryonic endoderm; four of those seven showed expression reductions as predicted.",
      "unknown": "The experiments were performed in mouse-derived cell lines, not intact embryos or human cells. Only selected predictions were tested; the 40-pair validation had 12 true positives, 13 true negatives and an overall reported accuracy of 0.71, leaving discordant/uncertain pairs. The Enh14 and Enh4 results support specific cell-type-dependent regulatory hubs but do not establish the full genome-wide predicted map as experimentally validated or prove developmental phenotypes in vivo.",
      "ai_role": "The model used 1D and 3D chromatin features to rank genome-wide enhancer-promoter pairs by predicted expression impact after in-silico enhancer perturbation. Researchers selected predictions for CRISPRi validation and measured target-gene expression by RT-qPCR; the paper reports specific validated multi-gene enhancer hubs and positive/negative prediction rates across 40 pairs.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10897904/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-02-celloracle-guided-zebrafish-perturbations-identify-axial-mesoderm-effects-of-noto-and-lhx1a-loss",
    "title": "CellOracle-guided zebrafish perturbations identify axial-mesoderm effects of noto and lhx1a loss",
    "summary": "The CellOracle machine-learning workflow used single-cell gene-regulatory networks to prioritize transcription factors for perturbation in zebrafish development. CRISPR/Cas9 tests confirmed a model-predicted, previously unreported increase in prechordal-plate differentiation after loss of noto, alongside the known loss of notochord. Model-prioritized lhx1a loss also inhibited axial-mesoderm differentiation: single-cell RNA-seq found expansion of early notochord cells and reductions in later notochord and prechordal-plate populations; orthogonal RNA assays supported reduced notochord/prechordal markers.",
    "field": "biology",
    "ai_technology": "CellOracle, a machine-learning gene-regulatory-network model with in-silico transcription-factor perturbation",
    "organization": "Washington University in St. Louis; collaborating institutions",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9946838/",
    "source_type": "paper",
    "publication_date": "2023-02-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-022-05688-9",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "CRISPR/Cas9 tests confirmed a model-predicted, previously unreported increase in prechordal-plate differentiation after loss of noto, alongside the known loss of notochord.",
      "unknown": "Evidence is from zebrafish crispants during embryonic development, not stable mutants or mammalian/human systems; the authors report no gross morphological change despite transcriptomic/cell-population effects. The CellOracle model depends on inferred networks and makes cell-state-shift predictions, not direct causal proof of all inferred edges. A later unreviewed 2024 preprint raises implementation and benchmarking concerns, particularly about its use of ATAC-seq for distal regulatory interactions and comparisons with other GRN tools; this record relies only on the specific perturbation predictions tested experimentally and does not adopt the paper's broader benchmarking claims.",
      "ai_role": "The model built context-specific networks from single-cell data, simulated transcription-factor loss, and prioritized noto and lhx1a for experimental follow-up. Researchers used CRISPR/Cas9 crispants, single-cell RNA-seq and marker assays to test the predicted zebrafish cell-fate effects; the prechordal-plate increase after noto loss was reported as previously unrecognized.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9946838/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-05-acranker-guided-discovery-of-acriia20-and-acriia21-anti-crispr-proteins",
    "title": "AcRanker-guided discovery of AcrIIA20 and AcrIIA21 anti-CRISPR proteins",
    "summary": "AcRanker, an XGBoost ranking model using amino-acid sequence features, ranked proteins from prophage regions in self-targeting Streptococcus genomes. Experimental testing identified two previously unknown type II-A anti-CRISPRs: AcrIIA20 (ML1), which strongly inhibited Streptococcus iniae Cas9 and weakly inhibited S. pyogenes Cas9, and AcrIIA21 (ML8), which inhibited S. pyogenes, S. aureus and S. iniae Cas9 with lower potency.",
    "field": "biology",
    "ai_technology": "AcRanker, an XGBoost pairwise ranking model trained on amino-acid composition and grouped di- and tri-mer features",
    "organization": "University of California, Berkeley; Monash University; University of Melbourne",
    "source_url": "https://academic.oup.com/nar/article/48/9/4698/5819938",
    "source_type": "paper",
    "publication_date": "2020-05-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/nar/gkaa219",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimental testing identified two previously unknown type II-A anti-CRISPRs: AcrIIA20 (ML1), which strongly inhibited Streptococcus iniae Cas9 and weakly inhibited S. pyogenes Cas9, and AcrIIA21 (ML8), which inhibited S. pyogenes, S. aureus and S. iniae Cas9 with lower potency.",
      "unknown": "The authors selected 10 candidates for follow-up, successfully purified nine and reported three validated candidates across their experiments, including ML3/AcrIIA12, which had also been identified separately as an LmoCas9 inhibitor during the study. AcrIIA20 and AcrIIA21 were roughly tenfold less potent against SpyCas9 than the established AcrIIA4 control; tested potency and host range are assay-specific. The 2025 Evo paper reports distinct generated proteins EvoAcr1–5 and separate assays; shared anti-CRISPR function does not make these identical discoveries.",
      "ai_role": "The model ranked candidate proteins from predicted prophage regions for anti-CRISPR likelihood; the researchers selected candidates from those rankings and biochemically tested them. It directly prioritized the proteins later named AcrIIA20 and AcrIIA21, whose Cas9 inhibition was measured in vitro.",
      "evidence_urls": [
        "https://academic.oup.com/nar/article/48/9/4698/5819938"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-ai-prioritized-crispr-screen-identifies-alox5-and-oxtr-as-regulators-of-keratinocyte-il17ra-in-psoriasis",
    "title": "AI-prioritized CRISPR screen identifies ALOX5 and OXTR as regulators of keratinocyte IL17RA in psoriasis",
    "summary": "A genome-wide CRISPR knockout screen in primary human epidermal keratinocytes, combined with VirtualCRISPR language-model prioritization, identified ALOX5 and OXTR as previously underappreciated cell-intrinsic regulators of surface IL17RA. Targeted knockout in keratinocytes from two adult donors reduced IL17RA and inflammatory cytokine secretion. Topical inhibition of ALOX5 or OXTR also reduced disease signs in an imiquimod mouse model of psoriasiform dermatitis.",
    "field": "biology",
    "ai_technology": "VirtualCRISPR, a functional-genomics language-model framework trained on BioGRID-ORCS screens",
    "organization": "Biohub; Northwestern University; University of Chicago",
    "source_url": "https://www.nature.com/articles/s41467-026-75249-5",
    "source_type": "paper",
    "publication_date": "2026-07-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-75249-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A genome-wide CRISPR knockout screen in primary human epidermal keratinocytes, combined with VirtualCRISPR language-model prioritization, identified ALOX5 and OXTR as previously underappreciated cell-intrinsic regulators of surface IL17RA.",
      "unknown": "The genetic validation used primary cells from two donors, while therapeutic effects were tested in an imiquimod-induced mouse model, not patients. The screen provided the initial unbiased hits; the AI supplied an orthogonal novelty-prioritization layer and did not itself measure biological effects. The paper reports a provisional patent application covering methods and findings. No clinical efficacy, safety or topical treatment benefit in humans is established.",
      "ai_role": "After the unbiased genome-wide CRISPR screen, VirtualCRISPR scored whether perturbing each gene in primary adult keratinocytes would alter IL17RA expression. The authors combined screen enrichment with the model’s lower prior probability for noncanonical hits to prioritize high-novelty regulators; targeted CRISPR knockout and pharmacologic mouse experiments tested the resulting biology.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-75249-5"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-celloracle-guided-pancreatic-islet-differentiation-experiments-reveal-stat1-and-zeb1-lineage-roles",
    "title": "CellOracle-guided pancreatic islet differentiation experiments reveal STAT1 and ZEB1 lineage roles",
    "summary": "A digital model built from 400,603 cells across 61 single-cell datasets used CellOracle gene-regulatory-network inference and in-silico transcription-factor perturbations to nominate regulators of stem-cell-derived human islet differentiation. CRISPRi experiments supported a previously unreported role for STAT1 in exocrine specification: transient STAT1 knockdown reduced exocrine markers, and JAK–STAT inhibition preserved beta-cell markers and improved glucose-stimulated insulin secretion. CRISPRi also supported a stage-dependent role for ZEB1, whose early knockdown impaired endocrine specification and later knockdown reduced off-target serotonergic cells while increasing beta-cell proportion. The date is the first preprint posting; this entry summarizes the later version reviewed in the research record, which reports the specific STAT1 and ZEB1 experiments.",
    "field": "biology",
    "ai_technology": "CellOracle v0.12.0 within a digital model integrating single-cell transcriptomic and chromatin-accessibility data",
    "organization": "Washington University School of Medicine; University of Miami",
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.04.27.721124v2.full",
    "source_type": "paper",
    "publication_date": "2026-04-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.04.27.721124",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "CRISPRi experiments supported a previously unreported role for STAT1 in exocrine specification: transient STAT1 knockdown reduced exocrine markers, and JAK–STAT inhibition preserved beta-cell markers and improved glucose-stimulated insulin secretion.",
      "unknown": "This is a bioRxiv preprint (first posted 30 April 2026; the source lists a later version on 17 September 2026), not peer-reviewed at this source date. It integrates existing datasets and reports laboratory validation in stem-cell-derived islet cultures; it does not establish human therapeutic benefit or mature clinical transplantation outcomes. Regulatory-model predictions remain context- and stage-dependent.",
      "ai_role": "The integrated cell-state model inferred regulatory networks and performed in-silico transcription-factor knockout simulations to identify candidate lineage regulators and intervention windows. These model predictions guided subsequent CRISPRi and pharmacologic perturbations in stem-cell-derived islet differentiation cultures.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.64898/2026.04.27.721124v2.full"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-05-bacterai-experimentally-maps-amino-acid-growth-requirements-of-two-oral-streptococcus-species",
    "title": "BacterAI experimentally maps amino-acid growth requirements of two oral Streptococcus species",
    "summary": "BacterAI autonomously designed and ran growth experiments with robotic liquid handling to learn amino-acid requirements for the oral commensals Streptococcus gordonii and S. sanguinis. It learned separate growth rules for each organism and showed that the closely related species have different amino-acid auxotrophies. For S. gordonii, its neural-network growth predictions exceeded 90% accuracy after nine days and 3,024 experiments; the article also reports transfer-learning tests in larger media.",
    "field": "biology",
    "ai_technology": "BacterAI, a reinforcement-learning agent coupled to iterative neural-network growth models and automated laboratory robots",
    "organization": "University of Illinois Urbana-Champaign; University of Michigan",
    "source_url": "https://www.nature.com/articles/s41564-023-01376-0",
    "source_type": "paper",
    "publication_date": "2023-05-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41564-023-01376-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "BacterAI autonomously designed and ran growth experiments with robotic liquid handling to learn amino-acid requirements for the oral commensals Streptococcus gordonii and S. sanguinis.",
      "unknown": "The measured outcomes are in defined growth media under the tested laboratory conditions, not direct evidence about nutrient exchange or health effects in a human oral microbiome. The article establishes that the two species have different amino-acid requirements; transfer learning and high growth-prediction accuracy describe the method’s performance, not separate biological discoveries. Replicate growth assays showed high agreement, but inferred rules may not cover every environmental condition.",
      "ai_role": "The agent selected successive combinations of amino acids to test based on prior growth results, executed those tests through an automated pipeline, and distilled results into interpretable growth rules. It enabled an unbiased experimental mapping of nutrient requirements from a blank slate for both species.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41564-023-01376-0"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-prospective-multicenter-validation-of-triverity-links-a-29-mrna-host-response-to-infection-type-and-near-term-severity",
    "title": "Prospective multicenter validation of TriVerity links a 29-mRNA host response to infection type and near-term severity",
    "summary": "In the prospective SEPSIS-SHIELD study, a machine-learning blood test using isothermal amplification of 29 host-response mRNAs was evaluated in adults presenting to 22 emergency departments with suspected infection or sepsis. Among 1,222 patients with valid results, its bacterial and viral scores distinguished clinically adjudicated infection status with AUROCs of 0.83 and 0.91, respectively, and the system also produced a separate score for risk of ICU-level care within seven days. The test’s high/very-high and low/very-low interpretation bands had 92–99% sensitivities and specificities for selected infection/severity endpoints.",
    "field": "medicine",
    "ai_technology": "TriVerity on the Myrna instrument, using logistic-regression classifiers on expression measurements for 29 host-response mRNAs",
    "organization": "Inflammatix; Stanford University; Washington University in St. Louis; 22 emergency-department study sites",
    "source_url": "https://www.nature.com/articles/s41591-025-03933-y",
    "source_type": "paper",
    "publication_date": "2025-09-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-025-03933-y",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In the prospective SEPSIS-SHIELD study, a machine-learning blood test using isothermal amplification of 29 host-response mRNAs was evaluated in adults presenting to 22 emergency departments with suspected infection or sepsis.",
      "unknown": "This is prospective diagnostic/prognostic validation of a specific host-response test, not discovery of a causal infection mechanism and not evidence that use of the score improves patient outcomes or antibiotic decisions. Of 1,441 enrolled patients, 1,222 had valid test results; the main diagnostic analysis used 729 consensus-adjudicated patients, and 1,120 were evaluable for prognosis. Clinical adjudication is not a perfect microbiological gold standard, and the study reports lower accuracy when uncertain adjudications are included. Performance and predictive values may vary with local infection prevalence and case mix. This is distinct from the earlier VB10 10-gene signature paper (DOI 10.1016/j.ebiom.2021.103352), which used different genes, cohorts and a different diagnostic score.",
      "ai_role": "The locked machine-learning classifiers convert the amplified host-response mRNA profile into separate likelihood scores for bacterial infection, viral infection and need for critical-care interventions within seven days. The multicenter clinical study measured how those AI-generated scores performed against blinded physician adjudication and near-term clinical endpoints.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41591-025-03933-y"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2023-08-deep-learning-knee-imaging-phenotypes-reveal-osteoarthritis-genetic-and-fracture-associations",
    "title": "Deep-learning knee imaging phenotypes reveal osteoarthritis genetic and fracture associations",
    "summary": "Using knee DXA scans from the UK Biobank, a ResNet-101 model identified 1,931 more probable knee-osteoarthritis cases than health-record coding and a U-Net segmentation model estimated minimum joint-space width (mJSW). In 29,257 participants, the quantitative mJSW phenotype had substantially higher SNP heritability than the binary phenotypes and yielded 18 genome-wide-significant loci, compared with one for the image-based binary classifier and six for ICD-10 coding; only one mJSW locus was already significant in a large prior knee-OA GWAS. The AI-derived mJSW phenotype was also associated with fracture history in an epidemiological analysis, a signal not seen with the binary case labels.",
    "field": "biology",
    "ai_technology": "ResNet-101 convolutional neural network for knee-OA classification and U-Net image segmentation for joint-space measurement",
    "organization": "University of Texas at Austin; Dell Medical School; Columbia University; New York Genome Center",
    "source_url": "https://www.nature.com/articles/s41746-023-00903-x",
    "source_type": "paper",
    "publication_date": "2023-08-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41746-023-00903-x",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using knee DXA scans from the UK Biobank, a ResNet-101 model identified 1,931 more probable knee-osteoarthritis cases than health-record coding and a U-Net segmentation model estimated minimum joint-space width (mJSW).",
      "unknown": "The GWAS used 29,257 quality-controlled UK Biobank participants with knee DXA images, all from the analyzed white-British ancestry subset; ancestry transferability is unestablished. DXA images were non-weight-bearing. The 18 loci are genome-wide associations for a model-derived quantitative endophenotype, not 18 proven causal OA genes; the paper notes only one also reached significance in a previous large knee-OA GWAS and many loci overlap broad anthropometric associations. The fracture result is observational and does not show that OA or joint-space narrowing causes fractures. The medRxiv preprint DOI 10.1101/2023.03.07.23286909 is an earlier version of this same study and is not counted separately.",
      "ai_role": "The neural networks converted routine knee DXA scans into two disease phenotypes: a case–control OA classifier and a quantitative minimum joint-space-width measurement. The researchers used these AI-derived phenotypes in GWAS and downstream epidemiological association analyses, increasing statistical power and revealing genetic and fracture associations that were not significant with ICD-10 case coding or the binary classifier.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41746-023-00903-x"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-07-machine-learning-selected-plasma-markers-distinguish-lewy-body-and-frontotemporal-dementias",
    "title": "Machine-learning-selected plasma markers distinguish Lewy body and frontotemporal dementias",
    "summary": "A six-cohort plasma proteomics study used cross-validated LASSO to select a 21-protein multiplex panel for dementia differential diagnosis. In the new DLB/FTD results, ITGAV and ITGAM were reduced in dementia with Lewy bodies and also tracked Lewy-body pathology in independent autopsy-confirmed cohorts; a panel including ITGAV, ITGAM and ERBB3 distinguished DLB from controls and Alzheimer dementia in an independent multicenter cohort. For frontotemporal dementia, NfL and B4GAT1 contributed to the panel distinguishing FTD from controls. Validation-set AUCs were 0.83 for DLB versus controls, 0.78 for DLB versus Alzheimer dementia, 0.85 for FTD versus controls and 0.77 for FTD versus Alzheimer dementia.",
    "field": "medicine",
    "ai_technology": "Cross-validated LASSO regression and multivariable logistic-regression models applied to Olink proximity-extension plasma proteomics",
    "organization": "University of Perugia; Amsterdam University Medical Center; CSIRO; Charles University; Sant Pau Research Institute; collaborating international dementia centers",
    "source_url": "https://www.nature.com/articles/s43587-026-01162-7",
    "source_type": "paper",
    "publication_date": "2026-07-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43587-026-01162-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A six-cohort plasma proteomics study used cross-validated LASSO to select a 21-protein multiplex panel for dementia differential diagnosis.",
      "unknown": "These are observational plasma biomarker associations and retrospective diagnostic-validation results, not evidence that the proteins cause disease or improve patient outcomes. The work reports moderate rather than definitive discrimination, especially DLB versus Alzheimer dementia (validation AUC 0.78) and FTD versus Alzheimer dementia (0.77); external prospective utility remains unestablished. The later study reuses the bPRIDE cohorts and includes Alzheimer-disease analyses overlapping the group’s earlier 2025 bPRIDE plasma-proteomics paper; this entry is limited to the distinct DLB/FTD differential-marker and independent 21-plex validation findings. No individual biomarker is presented as a stand-alone clinical test.",
      "ai_role": "The analysis used 100-fold cross-validated LASSO feature selection to prioritize a limited set of disease-associated proteins and choose 21 markers for a custom multiplex assay. Pairwise logistic-regression models used selected markers to distinguish DLB and FTD from controls and Alzheimer dementia; the targeted panel was then measured in an independent multicenter validation cohort.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43587-026-01162-7"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-deep-learning-designed-enhancers-drive-target-tissue-reporter-activity-in-mouse-embryos",
    "title": "Deep-learning-designed enhancers drive target-tissue reporter activity in mouse embryos",
    "summary": "Researchers trained convolutional neural networks on E11.5 mouse heart, limb and midbrain chromatin-accessibility data, then fine-tuned the models using validated human and mouse enhancers and used model-guided sequence optimization to design 15 synthetic enhancers. In vivo transgenic reporter assays found all 15 reproducibly active in their intended tissue. Four of five heart designs and four of five CNS designs were exclusive to the target tissue; all five limb designs were strongly active in limb and had weaker activity in other TWIST1-expressing mesenchymal tissues.",
    "field": "biology",
    "ai_technology": "Convolutional neural networks for sequence-to-accessibility and sequence-to-activity prediction, transfer learning, and Ledidi gradient-based sequence design",
    "organization": "Research Institute of Molecular Pathology (IMP), Vienna BioCenter; University of California, Irvine; Vienna BioCenter PhD Program, University of Vienna and Medical University of Vienna",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13553301/",
    "source_type": "paper",
    "publication_date": "2026-08-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41588-026-02729-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers trained convolutional neural networks on E11.5 mouse heart, limb and midbrain chromatin-accessibility data, then fine-tuned the models using validated human and mouse enhancers and used model-guided sequence optimization to design 15 synthetic enhancers.",
      "unknown": "The functional test was a transgenic reporter assay at embryonic day 11.5, not editing or activation of an endogenous native gene locus. Fifteen selected sequences were tested (five per target), so the reported all-positive result does not estimate success across every generated candidate. Some designs had weaker off-target activity, particularly in related mesenchymal tissues. These results do not establish therapeutic efficacy or safe gene therapy.",
      "ai_role": "Models learned tissue-specific chromatin-accessibility patterns from genome-wide ATAC-seq, then transfer learning adapted them to validated enhancer activity. The trained models guided de novo enhancer sequence edits; the authors selected 15 candidates for transgenic mouse-embryo reporter testing, where each showed reproducible activity in its intended target tissue.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13553301/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-enzgfm-agent-experimentally-prioritizes-activity-enhancing-secreted-alkaline-phosphatase-variants",
    "title": "EnzGFM-Agent experimentally prioritizes activity-enhancing secreted alkaline phosphatase variants",
    "summary": "EnzGFM-Agent scored all possible single amino-acid substitutions in secreted alkaline phosphatase (SEAP) and selected top candidates from EnzGFM-1.5B and ESM2-3B for testing in transfected Chinese hamster ovary cells. Among the five variants uniquely prioritized by EnzGFM, all five had significantly higher measured activity than wild type in the reported assay; three exceeded 1.5-fold, and A46I reached 6.52-fold. This is an experimentally measured enzyme-engineering result for the tested SEAP variants.",
    "field": "biology",
    "ai_technology": "EnzGFM-Agent with EnzGFM-1.5B and ESM2-3B",
    "organization": "Henan Medical University; Shanghai Jiao Tong University; Shanghai AI Laboratory",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13494029/",
    "source_type": "paper",
    "publication_date": "2026-07-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-75283-3",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "EnzGFM-Agent scored all possible single amino-acid substitutions in secreted alkaline phosphatase (SEAP) and selected top candidates from EnzGFM-1.5B and ESM2-3B for testing in transfected Chinese hamster ovary cells.",
      "unknown": "The evidence is limited to single substitutions in one SEAP reporter protein tested after transfection in CHO-S cells, with three independent biological replicates; it does not establish improved activity for other enzymes, industrial process performance, or a naturally occurring enzyme function. The five EnzGFM-specific variants all differed significantly from wild type, but the study also evaluated ESM2-ranked variants and five variants shared by both models. This is a paper-specific experimental validation, not independent replication.",
      "ai_role": "The workflow scored every possible SEAP single-point substitution, ranked candidates using EnzGFM-1.5B and ESM2-3B, and selected 15 unique variants for experimental validation. The distinct EnzGFM-prioritized group comprised five variants; the paper reports significantly increased SEAP activity for all five, with A46I at 6.52-fold wild type.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13494029/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-routine-laboratory-lightgbm-screening-identifies-monoclonal-immunoglobulin-positivity-across-independent-hospitals",
    "title": "Routine-laboratory LightGBM screening identifies monoclonal immunoglobulin positivity across independent hospitals",
    "summary": "A LightGBM classifier trained on routine clinical and laboratory data distinguished serum immunofixation-positive monoclonal immunoglobulin cases from negative cases. In an independent three-hospital cohort of 2,270 patients referred for immunofixation testing, the model achieved an AUC of 0.945, with 87.0% sensitivity and 89.6% specificity. In a subset with serum protein electrophoresis results, model sensitivity was 84.6% versus 75.3% for electrophoresis; in a separate 627-patient time-extended cohort, sensitivity was 88.2%, above each of three nonspecialist clinicians in that study.",
    "field": "medicine",
    "ai_technology": "LightGBM classifier selected from eight machine-learning models",
    "organization": "Xijing Hospital, Fourth Military Medical University; 3201 Hospital; Shaanxi Provincial People’s Hospital; Xi’an Medical University; Xianyang Normal University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13314525/",
    "source_type": "paper",
    "publication_date": "2026-06-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3389/fimmu.2026.1856505",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In an independent three-hospital cohort of 2,270 patients referred for immunofixation testing, the model achieved an AUC of 0.945, with 87.0% sensitivity and 89.6% specificity.",
      "unknown": "This is retrospective diagnostic validation, not evidence that screening with the model improves health outcomes or shortens diagnosis in practice. All participants had been referred for immunofixation testing at tertiary or regional hospitals; an unselected community or primary-care population was not studied. The 627-patient later cohort was time-extended but not strictly prospective. The model is an adjunct for flagging patients for confirmatory tests; immunofixation remained the reference standard, and the paper states broader population validation and clinical utility remain to be established.",
      "ai_role": "Researchers used five feature-selection algorithms to identify shared predictors and trained/evaluated eight classifiers on routine clinical and laboratory measurements. The selected LightGBM model produced patient-level MIg-positive probabilities; its diagnostic performance was measured against serum immunofixation results in independent multi-center and later-collected cohorts, including direct comparisons with serum protein electrophoresis and nonspecialist clinicians.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13314525/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-03-hmd-amp-mining-uncovers-evolutionarily-remote-antimicrobial-peptides-with-validated-antibacterial-activity",
    "title": "HMD-AMP mining uncovers evolutionarily remote antimicrobial peptides with validated antibacterial activity",
    "summary": "A protein-language-model pipeline, HMD-AMP, screened host and gut-microorganism genomes from nine mammalian species and nominated antimicrobial peptides, including candidates with little similarity to known AMPs. Of 91 high-confidence peptides selected for experiments, 74 showed strong antibacterial activity and 48 were evolutionarily remote from known AMPs. Four tested peptides had broad-spectrum activity at low effective concentrations and low toxicity; the most potent reported candidate showed efficacy in a mouse peritoneal E. coli infection model.",
    "field": "medicine",
    "ai_technology": "HMD-AMP protein-language-model pipeline",
    "organization": "Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences; Chinese University of Hong Kong; Hong Kong Polytechnic University; Broad Institute and collaborating institutions",
    "source_url": "https://www.nature.com/articles/s41551-026-01630-w",
    "source_type": "paper",
    "publication_date": "2026-03-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41551-026-01630-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Of 91 high-confidence peptides selected for experiments, 74 showed strong antibacterial activity and 48 were evolutionarily remote from known AMPs.",
      "unknown": "These are in-vitro peptide activity results plus a single reported mouse peritoneal E. coli infection experiment; they do not establish human safety or efficacy, clinical antibiotic utility, or activity against every pathogen. Only 91 of the many predicted sequences were tested, and 74 showed strong activity under the authors’ assays. The entry is one paper-level finding; other AMP discovery records in this ledger use different candidate sequences, sources or design workflows. The model acronym overlaps an earlier 2021 HMD-AMP classifier preprint, but this 2026 paper is a separate study with mammalian metagenomic mining and experimental outcomes; they are not conflated.",
      "ai_role": "HMD-AMP used protein language-model representations and a hierarchical classifier to identify and rank candidate antimicrobial peptides from host and gut-microorganism metagenomic/genomic sequences. The researchers experimentally tested 91 high-confidence candidates; 74 showed strong antibacterial activity, including 48 evolutionarily remote sequences, and four had the reported broad-spectrum/low-toxicity profile. The strongest candidate was then evaluated in a mouse infection model.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41551-026-01630-w"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-ai-designed-bacterial-anti-defense-proteins-enable-phage-infection-and-plasmid-retention",
    "title": "AI-designed bacterial anti-defense proteins enable phage infection and plasmid retention",
    "summary": "RFdiffusion-based de novo binder design and experimental screening produced synthetic proteins that inhibited three bacterial defense systems: Thoeris and Avs1 in Bacillus subtilis assays, and DdmDE in Vibrio cholerae. Five designed proteins inhibited Thoeris and nine inhibited Avs1 in the reported screens. An engineered SBSphiJ phage carrying both anti-Thoeris and anti-Avs1 proteins replicated in B. subtilis cells expressing both defenses. In native V. cholerae O1 El Tor, two of four tested DdmDE binders improved plasmid resistance to clearance.",
    "field": "biology",
    "ai_technology": "RFdiffusion de novo binder design, ProteinMPNN sequence design and AlphaFold2/3 structure-guided candidate scoring",
    "organization": "Weizmann Institute of Science; École Polytechnique Fédérale de Lausanne (EPFL)",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13402655/",
    "source_type": "paper",
    "publication_date": "2026-06-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-74301-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "RFdiffusion-based de novo binder design and experimental screening produced synthetic proteins that inhibited three bacterial defense systems: Thoeris and Avs1 in Bacillus subtilis assays, and DdmDE in Vibrio cholerae.",
      "unknown": "These are laboratory proof-of-concept results in three bacterial defense systems and specific strains; they do not demonstrate therapeutic phage efficacy in animals or humans. Binder success was low relative to the computationally designed candidate sets, and the authors note that a verified binder may not work against homologous defense proteins. The native V. cholerae result was tested for plasmid retention, not general transformation across clinical isolates. The earlier bioRxiv version is the same study and is not a separate record.",
      "ai_role": "RFdiffusion generated candidate binder backbones for selected target sites in the Thoeris, Avs1 and DdmDE defense proteins; ProteinMPNN designed sequences and AlphaFold-based co-folding/scoring prioritized candidates for experiments. The selected synthetic binders were tested for defense inhibition, direct interactions and ability to restore phage infection or plasmid retention. The measured anti-defense functions and engineered-phage outcomes are the paper-specific results enabled by AI-guided protein design.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13402655/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-03-ai-guided-promoter-editing-raises-maize-zmvte4-expression-and-kernel-alpha-tocopherol",
    "title": "AI-guided promoter editing raises maize ZmVTE4 expression and kernel alpha-tocopherol",
    "summary": "Basenji2-based deep-learning models and editing-plasticity analysis guided CRISPR editing of maize ZmVTE4, a gene involved in vitamin E biosynthesis. Researchers isolated a previously unobserved four-base deletion in the gene’s 5′ untranslated region; the edited allele significantly increased ZmVTE4 expression and α-tocopherol content in kernels. A separate promoter-variant assay supported the four-base deletion, rather than a co-occurring 82-base deletion, as the key change associated with increased reporter activity.",
    "field": "biology",
    "ai_technology": "Basenji2-long and Basenji2-3K-NAM sequence-to-expression deep-learning models with editing-plasticity analysis",
    "organization": "Huazhong Agricultural University; Hubei Hongshan Laboratory; Yazhouwan National Laboratory; Institute of Agricultural Sciences of Xishuangbanna Prefecture; Max Planck Institute of Molecular Plant Physiology",
    "source_url": "https://link.springer.com/article/10.1186/s13059-025-03516-7",
    "source_type": "paper",
    "publication_date": "2025-03-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s13059-025-03516-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A separate promoter-variant assay supported the four-base deletion, rather than a co-occurring 82-base deletion, as the key change associated with increased reporter activity.",
      "unknown": "The demonstrated editing result concerns one ZmVTE4 allele and maize material in the reported experiments; it does not establish effects on yield, nutritional outcomes in consumers, or performance across environments and genetic backgrounds. The four-base deletion is newly engineered and was not found in the 721 accessions examined, rather than a naturally observed allele. The paper reports limitations in prediction for rice and tomato and incomplete modeling of environmental influences; claims here are restricted to the maize ZmVTE4 test.",
      "ai_role": "Deep-learning models predicted plant gene expression from regulatory DNA and used interpretability/occlusion and in-silico editing to identify promoter/5′UTR sequences likely to alter expression. The editing-plasticity predictions guided CRISPR-Cas9 targeting of ZmVTE4; the resulting four-base 5′UTR deletion was experimentally linked to increased gene expression and maize-kernel α-tocopherol. This is a measured crop-specific molecular/metabolite outcome, not merely an in-silico editing proposal.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s13059-025-03516-7"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-machine-learning-derived-seven-protein-panel-distinguishes-remission-from-active-anca-associated-vasculitis",
    "title": "Machine-learning-derived seven-protein panel distinguishes remission from active ANCA-associated vasculitis",
    "summary": "Confounder-aware machine-learning feature selection on plasma proteomics identified a seven-protein signature—AHSG, CLEC3B, COMP, F9, LRG1, MCAM and MRC1—to classify remission versus active ANCA-associated vasculitis (AAV). After targeted PRM mass-spectrometry assay development, a generalized linear model was evaluated in an independent cohort of 108 AAV patients and achieved AUC 0.94 (95% CI 0.90–0.98); combined discovery and validation performance was AUC 0.97. In the study cohorts, the panel outperformed routine CRP and ANCA titer measurements.",
    "field": "medicine",
    "ai_technology": "MetaDeconfoundR confounder-aware feature analysis, LASSO and Boruta feature selection, and a generalized linear model applied to targeted PRM mass-spectrometry data",
    "organization": "Max Delbrück Center for Molecular Medicine; Charité – Universitätsmedizin Berlin; Berlin Institute of Health; Charles University and General University Hospital Prague; University of Oslo",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42481524/",
    "source_type": "paper",
    "publication_date": "2026-07-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-75755-6",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "After targeted PRM mass-spectrometry assay development, a generalized linear model was evaluated in an independent cohort of 108 AAV patients and achieved AUC 0.94 (95% CI 0.90–0.98); combined discovery and validation performance was AUC 0.97.",
      "unknown": "This is cross-sectional classification of disease state in patients already diagnosed with AAV, not a prospective prediction of relapse, a validated basis for stopping immunosuppression, or evidence of improved clinical outcomes. The two cohorts together included just over 200 patients with a rare disease; the authors note uncertainty from sample size and treatment heterogeneity, and the panel remains a starting point for clinical-tool development. It is an adjunct biomarker result, not proof that the seven proteins cause remission or vasculitis activity.",
      "ai_role": "Machine-learning methods accounted for clinical covariates including CRP, kidney function and ANCA-related measures, prioritized candidate plasma proteins, and reduced these to a seven-protein panel. Targeted PRM-MS quantified the selected proteins, and the GLM classified AAV remission versus active disease in an independent cohort. The measured validation performance and protein signature are the specific AI-enabled diagnostic finding.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42481524/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-10-crispr-perturbation-and-bayesian-network-inference-map-immune-disease-regulatory-cascades-in-primary-cd4-t-cells",
    "title": "CRISPR perturbation and Bayesian network inference map immune-disease regulatory cascades in primary CD4+ T cells",
    "summary": "The study used CRISPR knockouts of 84 genes in primary human CD4+ T cells and a new linear latent causal Bayes (LLCB) network-inference method to estimate 211 regulatory connections. Integrating the inferred network with immune GWAS analyses identified an interconnected module in which KMT2A was inferred upstream of JAK-STAT and IL-2 signaling regulators including IRF4, STAT5A/B and IL2RA, linking an epigenetic regulator to immune-disease-associated pathways.",
    "field": "biology",
    "ai_technology": "Linear latent causal Bayes (LLCB), a Bayesian extension of linear latent causal network inference, applied to CRISPR knockout and RNA-sequencing data",
    "organization": "Gladstone-UCSF Institute of Genomic Immunology; University of California, San Francisco; Johns Hopkins University; Stanford University; University of California, Berkeley",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11605694/",
    "source_type": "paper",
    "publication_date": "2024-10-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.xgen.2024.100671",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study used CRISPR knockouts of 84 genes in primary human CD4+ T cells and a new linear latent causal Bayes (LLCB) network-inference method to estimate 211 regulatory connections.",
      "unknown": "The study perturbed 84 selected genes in primary CD4+ T cells from three donors, so this is not a complete T-cell regulatory map. LLCB “direct effects” are statistical effects adjusted for other perturbed genes and are not evidence of physical binding; effects from unperturbed genes can be attributed to modeled edges. The GWAS integration identifies pathway links, not proof that the inferred cascade causes clinical disease. A prior bioRxiv version (10.1101/2023.09.17.557749) is the same study and is not counted separately.",
      "ai_role": "LLCB inferred direct and indirect regulatory relationships from perturbation-associated expression effects across 84 targeted genes. The model-derived network and gene-program analysis, combined with immune GWAS statistics, supported the specific KMT2A-to-JAK-STAT/IL-2 regulatory finding. CRISPR perturbation supplied the experimental perturbation data; the network algorithm converted those measurements into the reported regulatory map.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11605694/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-11-multicenter-targeted-metabolomics-and-machine-learning-identify-a-six-metabolite-rheumatoid-arthritis-diagnostic-panel",
    "title": "Multicenter targeted metabolomics and machine learning identify a six-metabolite rheumatoid-arthritis diagnostic panel",
    "summary": "Across seven cohorts and five medical centers, targeted metabolomics confirmed six metabolites—imidazoleacetic acid, ergothioneine, N-acetyl-L-methionine, 2-keto-3-deoxy-D-gluconic acid, 1-methylnicotinamide and dehydroepiandrosterone sulfate—as a panel for classifying rheumatoid arthritis against healthy controls and osteoarthritis. LASSO selected the six features, and logistic regression, random forest and XGBoost models were externally evaluated in geographically distinct cohorts; reported AUCs were 0.8375–0.9280 for RA versus healthy controls and 0.7340–0.8181 for RA versus osteoarthritis in three validation cohorts.",
    "field": "medicine",
    "ai_technology": "LASSO feature selection and logistic regression, random forest, support vector machine and XGBoost classifiers applied to untargeted and targeted metabolomics",
    "organization": "Tongren Hospital, Shanghai Jiao Tong University School of Medicine; The First Affiliated Hospital of Fujian Medical University; Guanghua Hospital Affiliated to Shanghai University of Traditional Chinese Medicine; The First Hospital of Lanzhou University; Xinhua Hospital, Shanghai Jiao Tong University School of Medicine",
    "source_url": "https://link.springer.com/article/10.1186/s12967-025-07265-w",
    "source_type": "paper",
    "publication_date": "2025-11-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s12967-025-07265-w",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Across seven cohorts and five medical centers, targeted metabolomics confirmed six metabolites—imidazoleacetic acid, ergothioneine, N-acetyl-L-methionine, 2-keto-3-deoxy-D-gluconic acid, 1-methylnicotinamide and dehydroepiandrosterone sulfate—as a panel for classifying rheumatoid arthritis against healthy controls and osteoarthritis.",
      "unknown": "This is a cross-sectional case–control diagnostic study, not prospective screening or evidence of improved patient outcomes. The authors note that disease activity and treatment effects were not assessed, so prognostic and treatment-monitoring uses are not established. Performance varied across cohorts and was lower for RA-versus-OA than RA-versus-healthy-control classification; stronger platform-specific than sample-type effects were observed. Some mechanistic relevance of the six metabolites remains uncertain.",
      "ai_role": "Machine-learning feature selection narrowed targeted-validated metabolites to a six-analyte panel, and supervised classifiers combined the measurements to distinguish RA from healthy controls and osteoarthritis in multiple independent cohorts. Targeted mass spectrometry provided absolute quantification; the models produced the reported cross-cohort diagnostic discrimination.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s12967-025-07265-w"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-02-prospective-swedish-cohort-validates-multimodal-ai-prostate-biopsy-biomarker-for-recurrence-and-adverse-pathology-after-prostatectomy",
    "title": "Prospective Swedish cohort validates multimodal AI prostate-biopsy biomarker for recurrence and adverse pathology after prostatectomy",
    "summary": "A previously developed multimodal AI score was retrospectively applied to archived diagnostic biopsies and follow-up from a prospectively collected Swedish prostatectomy cohort. The score, derived from digitized diagnostic prostate-biopsy images plus age, PSA and clinical T-stage, was associated with post-surgical biochemical recurrence and adverse pathology. Among 143 men with evaluable data and median 8.8-year follow-up, the continuous score was associated with recurrence (subdistribution HR 2.45, 95% CI 1.77–3.38); among 88 evaluable for adverse pathology, OR was 4.85 (95% CI 2.54–10.78). Estimated 5-year recurrence was 25% in intermediate/high score versus 4% in low score.",
    "field": "medicine",
    "ai_technology": "Artera multimodal artificial-intelligence (MMAI) prostate biomarker combining deep-learning features from digitized biopsy histopathology with age, PSA and clinical T-stage",
    "organization": "Skåne University Hospital; Lund University; Artera Inc.; University of California, San Francisco",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11995000/",
    "source_type": "paper",
    "publication_date": "2025-02-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1158/1078-0432.CCR-24-3656",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A previously developed multimodal AI score was retrospectively applied to archived diagnostic biopsies and follow-up from a prospectively collected Swedish prostatectomy cohort.",
      "unknown": "This is retrospective external prognostic validation using specimens and follow-up from a prospectively collected cohort; it is not prospective AI-guided care, a randomized test of biomarker-guided treatment, or evidence that using the score improves outcomes. Only 143 patients had evaluable images and complete covariates, and 88 were evaluable for adverse pathology; it is a single tertiary Swedish center. The model’s earlier development setting involved men receiving radiation in North America, so this tests transport to a different population/treatment context. A 2025 conference abstract describing this cohort is an earlier presentation of the same study, not another record.",
      "ai_role": "The locked MMAI score converted diagnostic biopsy images and clinical variables into patient-level prognostic scores. The prospective Swedish cohort tested whether that model-derived score associated with the distinct post-prostatectomy outcomes biochemical recurrence and adverse pathology; associations persisted after adjustment for standard NCCN/CAPRA clinical risk measures.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11995000/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-08-machine-learning-guided-proteomics-identifies-apoc3-rab39a-and-kng1-for-plaque-rupture-classification-in-acute-coronary-syndrome",
    "title": "Machine-learning-guided proteomics identifies APOC3, RAB39A and KNG1 for plaque-rupture classification in acute coronary syndrome",
    "summary": "In an acute-coronary-syndrome derivation cohort matched for plaque rupture (PR) versus intact fibrous cap (IFC), quantitative mass spectrometry measured 1,121 plasma proteins. Five machine-learning algorithms screened candidate proteins; APOC3, RAB39A and KNG1 were then significantly different between PR and IFC in an independent 85-patient cohort using ELISA. The three proteins discriminated plaque rupture better than the study’s conventional biomarkers and risk factors, suggesting a candidate biomarker panel for identifying PR among ACS patients.",
    "field": "medicine",
    "ai_technology": "Unbiased machine-learning feature screening of quantitative mass-spectrometry plasma proteomics using five algorithms",
    "organization": "Beijing Anzhen Hospital, Capital Medical University; Beijing Institute of Heart Lung and Blood Vessel Diseases; First Affiliated Hospital of Dalian Medical University",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S0009898124021570",
    "source_type": "paper",
    "publication_date": "2024-08-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cca.2024.119904",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In an acute-coronary-syndrome derivation cohort matched for plaque rupture (PR) versus intact fibrous cap (IFC), quantitative mass spectrometry measured 1,121 plasma proteins.",
      "unknown": "This was a small matched derivation cohort (110 ACS patients) and an 85-patient validation cohort, not prospective outcome prediction or a clinical-impact trial. The proteins are candidate diagnostic markers; the study does not show that using them improves management or outcomes, nor establish causal roles in plaque rupture. The publisher reports electronic publication 2024-08-06 and issue date 2024-09-15; the record uses online-first date.",
      "ai_role": "The ML algorithms selected nine candidate proteins from the proteomic measurements; the study subsequently tested candidates in an independent ACS cohort and confirmed three analytes—APOC3, RAB39A and KNG1—as significantly different in plaque-rupture versus intact-cap groups. ELISA provided the independent protein validation. The AI-enabled result is this prioritized, externally tested three-protein panel for a specific plaque phenotype.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S0009898124021570"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-03-machine-learning-links-circulating-tyrosine-and-distinct-lipid-metabolite-features-to-diabetic-kidney-disease-and-retinopathy-detection",
    "title": "Machine learning links circulating tyrosine and distinct lipid-metabolite features to diabetic kidney disease and retinopathy detection",
    "summary": "In 2,772 adults with diabetes from Singapore’s population-based SEED study, LASSO and gradient-boosting models screened 220 circulating metabolites and 19 clinical factors for prevalent diabetic kidney disease (DKD) and retinopathy (DR), then were externally assessed in 5,843 UK Biobank participants. Both outcomes shared diabetes duration, insulin use, age and tyrosine as important model features. DKD additionally selected cardiovascular disease history, antihypertensive use, lactate, citrate and the intermediate-density-lipoprotein cholesterol-ester/total-lipid ratio; DR additionally selected HbA1c, glucose, pulse pressure and alanine. External AUCs were 0.791 for DKD and 0.778 for DR, versus 0.691 and 0.760 for conventional adjusted logistic-regression comparators.",
    "field": "medicine",
    "ai_technology": "Least absolute shrinkage and selection operator logistic regression and gradient-boosting decision trees applied to metabolomics and clinical risk factors",
    "organization": "Singapore Eye Research Institute; Singapore National Eye Centre; National University of Singapore; Duke-NUS Medical School",
    "source_url": "https://www.jmir.org/2024/1/e41065/",
    "source_type": "paper",
    "publication_date": "2024-03-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.2196/41065",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In 2,772 adults with diabetes from Singapore’s population-based SEED study, LASSO and gradient-boosting models screened 220 circulating metabolites and 19 clinical factors for prevalent diabetic kidney disease (DKD) and retinopathy (DR), then were externally assessed in 5,843 UK Biobank participants.",
      "unknown": "The analysis is retrospective and cross-sectional; it classifies prevalent complications and cannot establish temporal prediction or causal effects of tyrosine, lactate, citrate, alanine or lipid features. The Singapore cohort was recruited in 2004–2011 and was multiethnic Asian; about 95% had type 2 diabetes, but the study did not separate diabetes types. UK Biobank definitions and available eye-severity measures differed, and validation of moderate-or-worse DR was not feasible. Performance is modest, particularly external AUC 0.791 DKD/0.778 DR, and does not establish clinical utility.",
      "ai_role": "ML feature selection prioritized a shared set of clinical/metabolic factors and distinct DKD- and DR-associated metabolites, then built detection models externally evaluated in UK Biobank. These models outperformed traditional adjusted logistic-regression comparators in the reported internal and external validation. The scientific result is the disease-specific feature pattern and its validated detection performance, not merely an algorithm benchmark.",
      "evidence_urls": [
        "https://www.jmir.org/2024/1/e41065/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-prospective-silent-trial-validation-of-a-fine-tuned-pathology-foundation-model-for-egfr-mutation-screening-in-lung-adenocarcinoma",
    "title": "Prospective silent-trial validation of a fine-tuned pathology foundation model for EGFR mutation screening in lung adenocarcinoma",
    "summary": "A fine-tuned Prov-GigaPath ViT-g model, named EAGLE, predicted EGFR mutation status from routine H&E slides plus clinical features in lung adenocarcinoma. Across an international dataset of 8,461 slides it achieved mean AUC 0.847 internally and 0.870 in external cohorts. In a four-month prospective silent trial on 197 primary lung-adenocarcinoma samples, EAGLE achieved AUC 0.890. Simulated AI-assisted triage thresholds reduced rapid molecular EGFR tests by up to 43% while maintaining the study’s performance criteria.",
    "field": "medicine",
    "ai_technology": "EAGLE, a task-specific model fine-tuned from the Prov-GigaPath ViT-g pathology foundation model, combining H&E whole-slide image features with clinical variables for EGFR mutation prediction",
    "organization": "Icahn School of Medicine at Mount Sinai; Memorial Sloan Kettering Cancer Center; Technical University of Munich; Sahlgrenska University Hospital and University of Gothenburg; SUNY Upstate Medical University; University of Montreal Hospital Centre",
    "source_url": "https://www.nature.com/articles/s41591-025-03780-x",
    "source_type": "paper",
    "publication_date": "2025-07-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-025-03780-x",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Across an international dataset of 8,461 slides it achieved mean AUC 0.847 internally and 0.870 in external cohorts.",
      "unknown": "The four-month prospective trial was silent: AI outputs did not guide care or replace standard testing. The 43% reduction is a simulated triage workflow at a selected score threshold, not an observed reduction in tests or evidence of improved outcomes. The cohort was 197 primary cases at one real-time deployment site, while broader retrospective/external cohorts supported generalization. EAGLE predicts presence of EGFR mutation but does not identify the mutation subtype; sequencing remains necessary before treatment. Performance was lower on metastatic samples. The paper is a distinct 2025 model iteration and validation study in the lineage of the group’s 2022 arXiv proof of concept (10.48550/arXiv.2206.10573); the prior work is cited as antecedent, not treated as a second alias or hidden as the first-ever pathology-image EGFR result.",
      "ai_role": "The AI model predicted tumor EGFR mutation status directly from routine H&E digital slides, using sequencing results as ground truth. Fine-tuning the pathology foundation model produced cross-institution validation, and real-time inference in a prospective silent trial tested generalization on newly processed clinical slides. The model score was evaluated as a screening/triage layer alongside rapid molecular testing; it did not replace genomic sequencing.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41591-025-03780-x"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2025-10-a-six-metabolite-machine-learning-panel-and-isovalerylcarnitine-findings-for-gastric-cancer",
    "title": "A six-metabolite machine-learning panel and isovalerylcarnitine findings for gastric cancer",
    "summary": "In two multicenter cohorts totaling 597 people, LASSO feature selection and five machine-learning classifiers produced a six-metabolite plasma panel for gastric-cancer diagnosis, with AUCs of 0.947–0.982 in discovery and 0.920–0.951 in validation. The panel included isovalerylcarnitine (C5), which was lower in cancer plasma and paired tumor tissue; the paper further reported a Mendelian-randomization association and cell-assay evidence linking C5 treatment to reduced gastric-cancer cell migration and invasion through calpain-associated changes in VE-cadherin and MMP2.",
    "field": "medicine",
    "ai_technology": "LASSO regression score feature selection with neural-network, support-vector-machine, ridge-regression, logistic-regression and naive-Bayes classifiers",
    "organization": "Zhejiang Cancer Hospital; and six collaborating clinical centers; Shanghai Jiao Tong University; Zhejiang Chinese Medical University; Wenzhou Medical University; National Cancer Center/National Clinical Research Center for Cancer, China",
    "source_url": "https://link.springer.com/article/10.1038/s44321-025-00325-0",
    "source_type": "paper",
    "publication_date": "2025-10-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44321-025-00325-0",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In two multicenter cohorts totaling 597 people, LASSO feature selection and five machine-learning classifiers produced a six-metabolite plasma panel for gastric-cancer diagnosis, with AUCs of 0.947–0.982 in discovery and 0.920–0.951 in validation.",
      "unknown": "The cohorts were case-control clinical samples and do not establish population-screening performance or improved patient outcomes. The validation cohort contained 453 participants from seven centers; no prospective care-impact trial was reported. Mendelian-randomization evidence is observational genetic-instrument analysis, and cell migration/invasion experiments do not establish a treatment effect in animals or people. The 2024 IDDF abstract (DOI 10.1136/gutjnl-2024-IDDF.267) is an earlier disclosure of this same project, not a separate discovery. Source checked; independent replication not assessed.",
      "ai_role": "Machine-learning feature selection narrowed replicated plasma metabolite signals to a six-feature diagnostic panel, and the specified classifiers constructed/evaluated diagnostic models. The C5 mechanistic work was follow-up metabolomics, Mendelian-randomization, proteomic-correlation and in-vitro experimentation; the record does not imply that machine learning alone established the mechanism.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1038/s44321-025-00325-0"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-10-a-seven-metabolite-ai-model-detected-renal-cell-carcinoma-across-six-validation-cohorts",
    "title": "A seven-metabolite AI model detected renal-cell carcinoma across six validation cohorts",
    "summary": "An SVM-based screen of untargeted plasma metabolomics identified candidate renal-cell-carcinoma metabolites, which were confirmed by targeted analysis and used in the RCAID AI diagnostic model. In 1,680 participants across five hospitals, seven plasma metabolites supported AUROCs of 0.977, 0.911, 0.945 and 0.972 in internal, external, multicenter and temporal validation cohorts, respectively; validation also included late-stage and non-clear-cell RCC. The authors report six associated dysregulated pathways from multi-omic analyses.",
    "field": "medicine",
    "ai_technology": "Support-vector-machine metabolite selection and the Renal Cell Carcinoma Artificial Intelligence Detector (RCAID)",
    "organization": "Peking University First Hospital; and Peking University Health Science Center; Beijing Miyun Hospital; Beijing Friendship Hospital; collaborating Chinese hospitals",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41047317/",
    "source_type": "paper",
    "publication_date": "2025-10-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.eururo.2025.09.4148",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An SVM-based screen of untargeted plasma metabolomics identified candidate renal-cell-carcinoma metabolites, which were confirmed by targeted analysis and used in the RCAID AI diagnostic model.",
      "unknown": "The study used diagnostic case-control cohorts with healthy controls, so reported discrimination does not establish population-screening performance, clinical utility, or improved outcomes. Most cases were clear-cell RCC (81%) and stage I (69%); subtype and cohort composition may affect generalizability. The abstract reports multi-omic pathway associations but does not establish causal mechanisms. Samples were collected 2019–2023; source publication date is the 2025 online-first date. Source checked; independent replication not assessed.",
      "ai_role": "The source states that an SVM was used to identify candidate metabolites from untargeted plasma profiles; targeted metabolomics confirmed selected metabolites. The resulting seven-feature AI-aided RCAID model was evaluated in six independent internal, external, multicenter, temporal, late-stage and non-clear-cell validation cohorts.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41047317/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-10-a-gradient-boosted-model-detected-untreated-infection-before-antimicrobial-administration",
    "title": "A gradient-boosted model detected untreated infection before antimicrobial administration",
    "summary": "The eDENTIFI gradient-boosted model estimated the probability of untreated infection during the next six hours from demographics, vital signs and laboratory results. Trained on 3,357 adult hospitalizations at two Illinois health systems and validated in 1,632 patients at a third, it achieved validation AUC 0.80 versus 0.64 for SIRS criteria. At the reported threshold it detected infection a median two hours before antimicrobial administration (47.6% sensitivity; 93.6% negative predictive value).",
    "field": "medicine",
    "ai_technology": "eDENTIFI, a gradient-boosted machine model using longitudinal discrete-time predictions",
    "organization": "University of Chicago Medical Center; Duke University; Emory University; Loyola University; Endeavor Health; University of Wisconsin-Madison",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39392375/",
    "source_type": "paper",
    "publication_date": "2024-10-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1097/CCE.0000000000001165",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Trained on 3,357 adult hospitalizations at two Illinois health systems and validated in 1,632 patients at a third, it achieved validation AUC 0.80 versus 0.64 for SIRS criteria.",
      "unknown": "Retrospective derivation and validation used historical hospital records from 2006–2018, and the infection reference standard was clinician chart review. The reported 10.8% absolute/46.4% relative reduction in unnecessary IV antibiotics is a counterfactual model comparison, not observed intervention evidence. The authors state prospective validation is needed; no patient benefit or implementation outcome was established.",
      "ai_role": "The model used routinely available demographics, vital signs and laboratory results to predict clinician-chart-reviewed untreated infection over the next six hours. The study tested temporal detection performance in a separate third health system; model estimates were not prospectively used to direct care.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39392375/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-10-explainable-neural-network-analysis-nominated-gpc6-as-a-parkinson-disease-dopaminergic-neuron-marker",
    "title": "Explainable neural-network analysis nominated GPC6 as a Parkinson disease dopaminergic-neuron marker",
    "summary": "An explainable machine-learning workflow applied to four post-mortem midbrain single-nucleus RNA-sequencing datasets identified cell-type-specific gene signatures that distinguished Parkinson disease from control cells across datasets (mean balanced accuracy 0.92) and nominated ten candidate genes. GPC6 emerged as a marker in dopaminergic neurons; rare GPC6 variants were enriched among Parkinson disease cases in analyses spanning three case-control cohorts.",
    "field": "biology",
    "ai_technology": "Explainable neural-network classifiers with LIME feature attribution applied to single-nucleus RNA-sequencing data",
    "organization": "McGill University; Montreal Neurological Institute-Hospital",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/41120302/",
    "source_type": "paper",
    "publication_date": "2025-10-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41531-025-01147-0",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "GPC6 emerged as a marker in dopaminergic neurons; rare GPC6 variants were enriched among Parkinson disease cases in analyses spanning three case-control cohorts.",
      "unknown": "The expression analysis used post-mortem datasets and the rare-variant association is not a functional perturbation or proof of causality. GPC6 remains a candidate marker/association requiring further validation. The reported balanced accuracy is cell-level classifier performance, not a clinical diagnostic test. Source checked; independent replication beyond the reported cohorts was not assessed.",
      "ai_role": "The neural-network classifiers identified cell-type-specific genes important for distinguishing Parkinson disease from control nuclei across four datasets; LIME feature attribution prioritized candidate genes. The researchers then used external rare-variant case-control analyses to validate GPC6 enrichment. The model nominates markers and candidates; it does not establish that GPC6 causes Parkinson disease.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/41120302/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-machine-learning-protein-signatures-predicted-crohn-disease-years-before-diagnosis",
    "title": "Machine-learning protein signatures predicted Crohn disease years before diagnosis",
    "summary": "Regularized logistic-regression and other machine-learning analyses of 178 Olink proteins identified a 29-protein preclinical Crohn disease signature. It distinguished future cases from controls with AUC 0.85 in a Swedish discovery cohort and 0.87 in an independent preclinical cohort; a separate inception cohort produced AUC 1.0. The study also reported a less predictive ulcerative-colitis signature and found that downregulated proteins related to gut-barrier integrity and macrophage function correlated with time to Crohn diagnosis.",
    "field": "medicine",
    "ai_technology": "Regularized logistic regression and additional machine-learning methods (L1-penalized regression, elastic net, support-vector machines and random forests) applied to Olink protein measurements",
    "organization": "Örebro University; Karolinska Institutet; Uppsala University; Linköping University; University of Gothenburg; Umeå University; Swedish population-based cohorts and BIOIBD consortium collaborators",
    "source_url": "https://www.gastrojournal.org/article/S0016-5085%2824%2905741-X/pdf",
    "source_type": "paper",
    "publication_date": "2024-11-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1053/j.gastro.2024.11.006",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "It distinguished future cases from controls with AUC 0.85 in a Swedish discovery cohort and 0.87 in an independent preclinical cohort; a separate inception cohort produced AUC 1.0.",
      "unknown": "This is a nested case-control study using biobanked samples, not prospective screening or intervention evidence. The paper reports that the Crohn signature’s AUC fell to 0.58 when preclinical twins were compared with their healthy twin siblings, suggesting inherited/shared-environment factors affect discrimination. The ulcerative-colitis validation AUC was lower (0.67). Protein associations do not establish causal mechanisms; authors call for mechanistic studies. Source checked; independent replication beyond reported cohorts was not assessed.",
      "ai_role": "Machine-learning feature selection/modeling produced preclinical Crohn and ulcerative-colitis protein signatures in the discovery cohort. The fitted signature was evaluated in an external population-based preclinical cohort, an inception cohort at diagnosis, and twin comparisons. The result is predictive association from cohort data; the paper says mechanistic studies are needed.",
      "evidence_urls": [
        "https://www.gastrojournal.org/article/S0016-5085%2824%2905741-X/pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-11-a-16-feature-ai-model-distinguished-sjogren-disease-from-clinically-similar-conditions-using-routine-tests",
    "title": "A 16-feature AI model distinguished Sjögren disease from clinically similar conditions using routine tests",
    "summary": "The retrospective multicenter SMFIF study screened routine laboratory data and used multi-algorithm feature selection, SHAP ranking and a stacked ensemble to construct a 16-feature Sjögren disease model. Across a 34,958-patient source population from three Chinese hospitals, reported AUCs were 0.923 in testing, 0.934 in internal validation and 0.964 in a 545-person external cohort. The study also compared the model with conventional autoantibody and laboratory markers and evaluated discrimination from rheumatoid arthritis, lupus, systemic sclerosis and osteoarthritis.",
    "field": "medicine",
    "ai_technology": "SMFIF, a stacked ensemble of five selected classifiers built from multi-algorithm feature selection and SHAP-ranked routine laboratory features",
    "organization": "Nanjing Drum Tower Hospital/Nanjing University Medical School; Suzhou Municipal Hospital/Nanjing Medical University; Huai’an First People’s Hospital/Nanjing Medical University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12660861/",
    "source_type": "paper",
    "publication_date": "2025-11-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41746-025-02110-2",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Across a 34,958-patient source population from three Chinese hospitals, reported AUCs were 0.923 in testing, 0.934 in internal validation and 0.964 in a 545-person external cohort.",
      "unknown": "All cohorts were retrospective and drawn from three hospitals in China; the external cohort had 545 people, including 139 Sjögren cases. External calibration showed greater variability, and the authors call for prospective and multinational validation. Case groups had longer recorded symptom duration than controls and higher organ involvement, which may contribute to discrimination. The study does not show clinical impact, improved outcomes, or replacement of diagnostic criteria/clinical assessment.",
      "ai_role": "The AI pipeline screened 70 candidate features from routine laboratory tests, selected a 16-feature set using multiple feature-selection methods and SHAP ranking, and combined five base classifiers in a stacking model. The resulting score estimated Sjögren disease probability and was evaluated on a held-out test set, internal validation set and external-hospital cohort.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12660861/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-05-proteomic-profiling-and-machine-learning-identified-an-externally-validated-diagnostic-panel-and-severity-markers-for-infective-endocarditis",
    "title": "Proteomic profiling and machine learning identified an externally validated diagnostic panel and severity markers for infective endocarditis",
    "summary": "The study profiled plasma and excised vegetation samples from 238 infective-endocarditis patients and 100 controls, then used an ensemble machine-learning pipeline to select a 10-protein plasma diagnostic panel. The paper reports AUCs of 0.97 in the held-out discovery test set and 0.98 and 0.94 in two external plasma cohorts. Integrating proteomic and clinical features also produced a prognostic model for severe heart-failure status (abstract AUC 0.87). LRG1 and NDUFB4 were associated with infection severity in the cohort analyses.",
    "field": "medicine",
    "ai_technology": "Ensemble machine-learning diagnostic and prognostic models; the diagnostic workflow compared six algorithms and used recursive feature elimination with cross-validation to select a 10-protein panel",
    "organization": "Fudan University; Guangdong Provincial People's Hospital/Southern Medical University; Fuwai Hospital/Chinese Academy of Medical Sciences; collaborating hospitals in China",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40447567/",
    "source_type": "paper",
    "publication_date": "2025-05-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-60184-8",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports AUCs of 0.97 in the held-out discovery test set and 0.98 and 0.94 in two external plasma cohorts.",
      "unknown": "The study is an observational proteomic cohort analysis; it does not demonstrate that use of the panel improves patient outcomes or can replace current diagnostic criteria. The development comparison included 238 IE patients and 100 controls, so case-control separation may not reflect performance in an unselected real-world diagnostic population. External evaluation used two additional plasma cohorts (combined n=328); the prognostic endpoint and biomarkers are associations, not causal mechanisms. The paper's proposed NET-related therapeutic implications are hypothesis-generating and are not counted as demonstrated treatment benefit.",
      "ai_role": "Machine-learning feature-importance analysis and recursive feature elimination selected a 10-protein plasma combination for distinguishing infective endocarditis from non-endocarditis samples. The models were evaluated in a held-out discovery cohort and two external plasma cohorts. A separate model combined proteomic and clinical features to classify severe heart-failure status.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40447567/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-08-blood-dna-methylation-machine-learning-signatures-classified-brain-insulin-resistance-across-three-cohorts",
    "title": "Blood DNA methylation machine learning signatures classified brain insulin resistance across three cohorts",
    "summary": "A machine-learning framework applied to blood DNA-methylation profiles identified 540 CpG sites that classified brain insulin resistance in adults without type 2 diabetes. In the 167-person discovery cohort, the selected methylation signature was tested in two separate replication cohorts of 33 and 24 participants, with reported classification accuracy of 83–94%. The CpGs mapped to 445 genes enriched for neuronal-development and axonogenesis annotations; 98 sites showed significant blood–brain methylation correlation.",
    "field": "medicine",
    "ai_technology": "Supervised machine-learning framework using blood DNA-methylation profiles; a multi-fold elastic-net approach was described in the associated institutional and conference materials",
    "organization": "Institute for Diabetes Research and Metabolic Diseases, Helmholtz Munich/University of Tübingen; German Center for Diabetes Research; German Institute of Human Nutrition Potsdam-Rehbruecke; University Hospital Ulm; University of Lübeck; University of Potsdam",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40768600/",
    "source_type": "paper",
    "publication_date": "2025-08-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/scitranslmed.adv7834",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In the 167-person discovery cohort, the selected methylation signature was tested in two separate replication cohorts of 33 and 24 participants, with reported classification accuracy of 83–94%.",
      "unknown": "This is a biomarker-classification study, not a prospective clinical screening or intervention study. The discovery cohort had 167 participants and the replication cohorts were small (n=33 and n=24); replication accuracy is reported as 83–94%, and larger independent cohorts are needed to establish transportability. Blood–brain methylation correlations at 98 sites support a proxy relationship but do not show that peripheral methylation causes brain insulin resistance. The authors describe potential future diagnostic use rather than demonstrated clinical utility.",
      "ai_role": "The supervised model selected a 540-CpG methylation signature from peripheral blood measurements to classify participants with and without brain insulin resistance, and the signature was assessed in two replication cohorts. The model-derived CpG set was then used for differential-methylation, gene-annotation and blood–brain correlation analyses.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40768600/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-06-automated-methylome-analysis-identified-candidate-blood-epigenetic-markers-for-first-episode-schizophrenia",
    "title": "Automated methylome analysis identified candidate blood epigenetic markers for first-episode schizophrenia",
    "summary": "An automated machine-learning workflow screened public blood-methylation datasets and selected IGF2BP1, CENPI and PSME4 as candidate schizophrenia-associated methylation markers. The researchers then measured these genes by targeted qMSP in 30 first-episode, drug-naive schizophrenia patients and 30 controls: IGF2BP1 methylation was higher and PSME4 lower in cases, while CENPI did not differ. A five-feature model combining the three genes with age and sex classified cases versus controls with AUC 0.755 (95% CI 0.636–0.862).",
    "field": "medicine",
    "ai_technology": "JADBio AutoML platform, comparing classifiers including support-vector machines, random forests, ridge logistic regression and decision trees with feature selection and cross-validation",
    "organization": "Democritus University of Thrace; Aristotle University of Thessaloniki; University of Crete; Hellenic Mediterranean University; Foundation for Research and Technology – Hellas",
    "source_url": "https://www.nature.com/articles/s41398-024-02946-4.pdf",
    "source_type": "paper",
    "publication_date": "2024-06-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41398-024-02946-4",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The researchers then measured these genes by targeted qMSP in 30 first-episode, drug-naive schizophrenia patients and 30 controls: IGF2BP1 methylation was higher and PSME4 lower in cases, while CENPI did not differ.",
      "unknown": "The patient validation was a small case-control pilot (30 first-episode cases and 30 healthy controls); its AUC of 0.755 is modest and has a wide confidence interval. CENPI methylation did not differ in the clinical samples, so it should not be presented as independently validated. The study does not establish diagnostic performance in clinical differential-diagnosis populations, prospective utility or causality, and the authors describe the signature as potentially useful for future diagnostics rather than clinically established.",
      "ai_role": "AutoML selected gene-methylation features from genome-wide public schizophrenia methylome datasets and built predictive signatures. The authors then measured the selected markers using qMSP in a separate 30-case/30-control first-episode drug-naive cohort and trained a second model on methylation plus demographic variables.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41398-024-02946-4.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-12-picnic-machine-learning-predictions-revealed-previously-poorly-characterized-human-proteins-in-cellular-condensates",
    "title": "PICNIC machine learning predictions revealed previously poorly characterized human proteins in cellular condensates",
    "summary": "The PICNIC gradient-boosting model ranked proteins for membership in biomolecular condensates using sequence-pattern and AlphaFold2-derived structural features. To test its predictions, researchers expressed 24 poorly characterized human proteins in U2OS cells: 21 formed mesoscale fluorescent foci, and 18 met the paper’s high-confidence condensate threshold. The experiments mapped examples to cytoplasmic bodies, nuclear bodies, nucleoli or P-bodies; for example, TYW5 and RAD51AP1 showed distinct localization patterns. Across the tested set, the authors found condensate localization was not limited to highly disordered proteins.",
    "field": "biology",
    "ai_technology": "PICNIC, a gradient-boosting machine-learning classifier using protein sequence co-occurrence patterns, sequence complexity/disorder features and AlphaFold2-derived structural features",
    "organization": "Max Planck Institute of Molecular Cell Biology and Genetics; Center for Systems Biology Dresden; TU Dresden Cluster of Excellence Physics of Life",
    "source_url": "https://www.nature.com/articles/s41467-024-55089-x.pdf",
    "source_type": "paper",
    "publication_date": "2024-12-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-55089-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "To test its predictions, researchers expressed 24 poorly characterized human proteins in U2OS cells: 21 formed mesoscale fluorescent foci, and 18 met the paper’s high-confidence condensate threshold.",
      "unknown": "The experimental test used fluorescently tagged proteins expressed in U2OS cells, so observed localization may depend on expression level, tag, cell type or condition. The study did not establish each protein’s endogenous localization or biological function; the authors explicitly state that condensate localization alone makes no functional claim. Three of 24 proteins did not form detectable mesoscale foci, and three additional results were low-confidence under the paper’s size threshold. This is targeted experimental validation of model-selected candidates, not an independently replicated atlas.",
      "ai_role": "PICNIC scored human proteins for likely cellular-condensate membership and selected poorly characterized positive predictions for experimental testing. Its rankings directly guided the 24-protein U2OS localization validation; the model was built to predict condensate members (including clients), not only proteins that self-drive phase separation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-55089-x.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-ai-guided-subtilisin-variants-improve-aqueous-amide-formation",
    "title": "AI-guided subtilisin variants improve aqueous amide formation",
    "summary": "The study reports that wild-type subtilisin BPN′ directly catalyzed condensation of free carboxylic acids and amines in aqueous PBS (pH 8, 40 °C), forming a peptide-bond amide without stoichiometric coupling reagents. The authors combined AI-guided mutation ranking with a FRET assay, reducing 7,999 in-silico variants at three active-site positions to five experimental candidates. The top triple mutant N155G/N218D/M222Q had reported relative activity of 753 ± 14% of wild type and also coupled biotin to a peptide fragment in water, extending beyond peptide–peptide ligation.",
    "field": "biology",
    "ai_technology": "AI-guided enzyme mutation prediction combining binding-free-energy analysis of active-site substitutions with an experimental FRET-based rapid product-detection platform",
    "organization": "Shanghai Institute of Organic Chemistry, Chinese Academy of Sciences; Shanghai Artificial Intelligence Laboratory; collaborating Chinese research institutions",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S1385894725106992",
    "source_type": "paper",
    "publication_date": "2025-12-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cej.2025.169856",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reports that wild-type subtilisin BPN′ directly catalyzed condensation of free carboxylic acids and amines in aqueous PBS (pH 8, 40 °C), forming a peptide-bond amide without stoichiometric coupling reagents.",
      "unknown": "The reported chemistry is demonstrated for specific peptide fragments and a biotin–peptide coupling example under the stated aqueous conditions; the paper does not establish broad substrate generality or production-scale utility. The 753% figure is reported relative catalytic activity with wild type as the reference, not a yield increase or 753-fold improvement. The reaction requires follow-up for broader substrate scope and process optimization.",
      "ai_role": "The AI-guided pipeline evaluated combinations at residues N155, N218 and M222, narrowing 7,999 in-silico variants to five prioritized mutants for FRET-based experimental validation. This reduced the candidate set and enabled selection of the N155G/N218D/M222Q variant; the AI component was applied after the paper established the wild-type aqueous amidation reaction.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S1385894725106992"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-machine-learning-analysis-linked-two-sepsis-associated-neutrophil-clusters-to-five-mouse-validated-marker-genes",
    "title": "Machine-learning analysis linked two sepsis-associated neutrophil clusters to five mouse-validated marker genes",
    "summary": "Reanalysis of single-cell profiles from four sepsis patients and two healthy controls identified two neutrophil clusters enriched in the sepsis samples and a set of cluster-associated gene modules. LASSO and seven additional machine-learning classifiers prioritized eight hub genes, including ALPL, CD177, S100A8, S100A9 and STXBP2. Expression analyses in separate public sepsis transcriptome cohorts supported several markers, and qRT-PCR in cecal-ligation/puncture and LPS mouse models found increased neutrophil expression of those five genes. The result links sepsis-associated neutrophil states to a small marker panel, with experimental expression validation in mice.",
    "field": "biology",
    "ai_technology": "LASSO feature selection followed by seven machine-learning classifiers (k-nearest neighbors, linear discriminant analysis, logistic regression, naive Bayes, random forest, recursive partitioning and support-vector machine), combined with hdWGCNA and single-cell transcriptomic analysis",
    "organization": "Department of General Surgery, General Hospital of Northern Theater Command, Shenyang, China",
    "source_url": "https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2024.1377817/full",
    "source_type": "paper",
    "publication_date": "2024-05-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3389/fimmu.2024.1377817",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Expression analyses in separate public sepsis transcriptome cohorts supported several markers, and qRT-PCR in cecal-ligation/puncture and LPS mouse models found increased neutrophil expression of those five genes.",
      "unknown": "The single-cell discovery comparison was very small (four sepsis cases and two controls) and the two clusters are analysis-derived states in this dataset; the study does not establish that they are distinct universal neutrophil lineages. The public bulk cohorts validate expression/classification associations, not an independent single-cell subtype. qRT-PCR confirmed expression changes for five genes in mouse models; it did not test whether the genes cause sepsis phenotypes or are therapeutic targets. The article itself notes earlier reports of other sepsis-related neutrophil subtypes, so this record is bounded to its specific clusters/markers and assays.",
      "ai_role": "Machine-learning feature selection prioritized eight hub genes from expression modules associated with the two sepsis-enriched neutrophil clusters. The models were assessed on separate public training and validation transcriptome cohorts; the resulting marker set guided qRT-PCR validation of five genes in two sepsis mouse models.",
      "evidence_urls": [
        "https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2024.1377817/full"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-machine-learning-uses-historical-warming-patterns-to-constrain-future-climate-projections",
    "title": "Machine learning uses historical warming patterns to constrain future climate projections",
    "summary": "A CatBoost model trained on CMIP6 simulations used the observed 1971–2020 grid-cell warming pattern to identify tropical and polar regions that best constrain later global warming. Applying these spatial constraints reduced projection error variance by about 70%, compared with about 48% using global-mean warming alone; the results depend on model-based emergent relationships and are projections rather than direct observations of future climate.",
    "field": "climate",
    "ai_technology": "CatBoost machine-learning emergent-constraint model",
    "organization": "East China Normal University; Nanjing University of Information Science and Technology; Pacific Climate Impacts Consortium; University of Victoria",
    "source_url": "https://www.nature.com/articles/s41467-026-70205-9",
    "source_type": "paper",
    "publication_date": "2026-03-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-70205-9",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A CatBoost model trained on CMIP6 simulations used the observed 1971–2020 grid-cell warming pattern to identify tropical and polar regions that best constrain later global warming.",
      "unknown": "The study is a climate-model emergent-constraint analysis; uncertainty reduction was assessed with leave-one-model-out cross-validation, not by validating future outcomes. The reported probabilities are scenario-conditional projections.",
      "ai_role": "Learned grid-cell-scale relationships between historical warming fields and future global mean warming across CMIP6 simulations, then ranked grid cells by their influence on constrained projections.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-70205-9"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-deep-learning-maps-six-high-potential-deep-earth-scattering-zones",
    "title": "Deep learning maps six high-potential deep-Earth scattering zones",
    "summary": "Researchers used deep learning with iterative manual optimization to identify 174,929 PKP precursors in more than two million earthquake waveforms recorded from 1990 to 2024. The expanded global map connected previously isolated sampling regions and highlighted six high-potential scattering zones near the core–mantle boundary that the authors report had not previously been documented. These are model-inferred targets for later seismic imaging, not fully characterized structures.",
    "field": "physics",
    "ai_technology": "Deep-learning PKP precursor detector with human iterative optimization",
    "organization": "Chinese Academy of Sciences; University of Chinese Academy of Sciences",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JB033195",
    "source_type": "paper",
    "publication_date": "2026-08-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025JB033195",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers used deep learning with iterative manual optimization to identify 174,929 PKP precursors in more than two million earthquake waveforms recorded from 1990 to 2024.",
      "unknown": "The six features are high-potential scattering zones inferred from probability maps and proposed as targets for future multi-phase inversions; they are not precise outlines or independently characterized structures. The proposed thermochemical explanations are interpretations. Cross-validation with independent seismic phases supports multiple mapped ULVZs, but does not individually validate all six newly highlighted zones.",
      "ai_role": "The deep-learning detector screened over two million earthquake waveforms to identify 174,929 weak PKP precursor signals; researchers iteratively optimized detections manually. This AI-enabled catalog increased global sampling density by an order of magnitude and supported the dual-probability map that identified six previously undocumented high-potential scattering zones.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JB033195"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-ai-guided-experiments-establish-feasible-500950-w-printing-configurations-for-grcop-42",
    "title": "AI-guided experiments establish feasible 500–950 W printing configurations for GRCop-42",
    "summary": "A closed-loop Bayesian experimental-design method, BEAM, found multiple defect-free directed-energy-deposition configurations for GRCop-42 copper alloy at infrared laser powers from 950 down to 500 W. Researchers report that these experimentally successful settings solved a process-development problem that had resisted several months of manual trials, enabling GRCop-42 fabrication on lower-power platforms. This is a manufacturing-process result for an existing alloy, not a new alloy composition.",
    "field": "materials",
    "ai_technology": "BEAM (Bayesian Experimental design for Additive Manufacturing)",
    "organization": "Washington State University",
    "source_url": "https://onlinelibrary.wiley.com/doi/10.1002/aaai.70069",
    "source_type": "paper",
    "publication_date": "2026-08-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/aaai.70069",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A closed-loop Bayesian experimental-design method, BEAM, found multiple defect-free directed-energy-deposition configurations for GRCop-42 copper alloy at infrared laser powers from 950 down to 500 W.",
      "unknown": "The paper establishes feasible configurations in one directed-energy-deposition system for one alloy. It does not demonstrate aerospace-qualified components, service life or broader transfer to other machines and alloys. The reported defect-free characterization is bounded to the study's print assessment.",
      "ai_role": "BEAM uses a probabilistic surrogate and adaptive experimental-design strategy to select small batches of process configurations for laboratory testing; print outcomes update its feasibility model and guide subsequent tests. The loop identified successful configurations at 950, 700, 600 and 500 W, where prior manual trials had failed for months.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/10.1002/aaai.70069"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-01-machine-learning-screening-and-experiments-identify-two-new-superconductors-be2hfnb2-and-be2hf2nb",
    "title": "Machine-learning screening and experiments identify two new superconductors, Be₂HfNb₂ and Be₂Hf₂Nb",
    "summary": "An AI-accelerated workflow screened more than 1.3 million material structures, combining a graph-neural-network model of superconducting properties with density-functional calculations and stability filters. The researchers experimentally synthesized and confirmed superconductivity in Be₂HfNb₂ and Be₂Hf₂Nb, reporting two previously unreported compounds among 741 computationally screened stable candidates. The two compounds were confirmed in the Be₂Nb₃ substitution series; other listed candidates remain computational predictions.",
    "field": "materials",
    "ai_technology": "BEE-NET equivariant graph neural networks in an AI-accelerated high-throughput screening workflow",
    "organization": "University of Florida; Oak Ridge National Laboratory; Quantum Formatics",
    "source_url": "https://www.nature.com/articles/s41524-026-01964-8",
    "source_type": "paper",
    "publication_date": "2026-01-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-026-01964-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An AI-accelerated workflow screened more than 1.3 million material structures, combining a graph-neural-network model of superconducting properties with density-functional calculations and stability filters.",
      "unknown": "The AI model accelerated candidate selection but did not itself perform the synthesis; DFT and laboratory confirmation were essential. The paper reports an overall screening precision of 86%, so most candidate predictions were not experimentally verified in this study.",
      "ai_role": "Predicted Eliashberg spectral functions and superconducting transition temperatures to prioritize candidate structures, followed by DFT stability calculations and experimental synthesis/characterization by the research team.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-026-01964-8"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-machine-learning-triage-identifies-a-dormant-wandering-black-hole-in-a-galaxys-outskirts",
    "title": "Machine-learning triage identifies a dormant wandering black hole in a galaxy’s outskirts",
    "summary": "The off-nuclear transient TDE 2025abcr revealed a dormant supermassive black hole about 9.3 kiloparsecs from its host galaxy’s center, with a mass described as similar to the Milky Way’s central black hole. A machine-learning program searching for tidal-disruption flares anywhere in the sky selected the event for follow-up three months after deployment; astronomers then collected multi-instrument evidence and interpreted it as the strongest known wandering-black-hole case. Its precise origin remains uncertain, with stripped remnant and black-hole-ejection scenarios both discussed.",
    "field": "physics",
    "ai_technology": "Off-nuclear tidal-disruption-event classifier (tdescore)",
    "organization": "University of Maryland; NASA Goddard Space Flight Center; University of California Observatories and collaborators",
    "source_url": "https://doi.org/10.3847/2041-8213/ae77f3",
    "source_type": "paper",
    "publication_date": "2026-07-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The off-nuclear transient TDE 2025abcr revealed a dormant supermassive black hole about 9.3 kiloparsecs from its host galaxy’s center, with a mass described as similar to the Milky Way’s central black hole.",
      "unknown": "The specific off-nuclear transient was AI-selected, but the paper’s evidence and interpretation are observational follow-up, not a model-generated discovery. The black hole’s formation history is unresolved; the host has no obvious surrounding galaxy at the black-hole position.",
      "ai_role": "Ranked the transient as a candidate tidal-disruption event away from a galaxy nucleus, prompting spectroscopic and multi-wavelength follow-up; the astrophysical interpretation relied on subsequent human analysis and observations.",
      "evidence_urls": [
        "https://doi.org/10.3847/2041-8213/ae77f3"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-ai-guided-autonomous-experiments-map-phase-domains-in-bismuthtitanium-oxide-films",
    "title": "AI-guided autonomous experiments map phase domains in bismuth–titanium oxide films",
    "summary": "A robotic active-learning campaign using SARA-H mapped processing regions that stabilize δ-Bi₂O₃ and Bi₂Ti₂O₇ in thin films and examined dwell-dependent phase behavior in the Bi–Ti–O system. The study also experimentally supported the prediction that bismuth substitution in TiO₂ inhibits the anatase-to-rutile transformation. This is a human-in-the-loop experimental materials result; the compounds/phases were not all newly discovered, while the mapped processing domains and comparative evidence are the study-specific contribution.",
    "field": "materials",
    "ai_technology": "SARA-H (Scientific Autonomous Reasoning Agent with human input) and automated probabilistic phase labeling",
    "organization": "Cornell University; California Institute of Technology; Liquid Sunlight Alliance",
    "source_url": "https://journals.aps.org/prxintelligence/abstract/10.1103/c362-349b",
    "source_type": "paper",
    "publication_date": "2026-09-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/c362-349b",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A robotic active-learning campaign using SARA-H mapped processing regions that stabilize δ-Bi₂O₃ and Bi₂Ti₂O₇ in thin films and examined dwell-dependent phase behavior in the Bi–Ti–O system.",
      "unknown": "This is a published, human-guided experimental study, but the phases themselves are not claimed as newly discovered compounds. The study-specific finding is experimentally mapped processing/dwell behavior in the Bi–Ti–O campaign and evidence that Bi substitution inhibits anatase-to-rutile transformation. Scope is thin films and tested processing conditions, not bulk phase stability or universal behavior.",
      "ai_role": "Used phase-identification outputs to select and interpret robotic laser-annealing experiments, with human domain guidance improving search efficiency.",
      "evidence_urls": [
        "https://journals.aps.org/prxintelligence/abstract/10.1103/c362-349b"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-multiscale-machine-learning-microscopy-identifies-robust-paleomagnetic-carriers-in-meteorite-wis-91600",
    "title": "Multiscale machine-learning microscopy identifies robust paleomagnetic carriers in meteorite WIS 91600",
    "summary": "A multiscale microscopy study of carbonaceous chondrite WIS 91600 used a deep-learning classifier to segment magnetic remanence-carrying ensembles and quantify their volume and texture. Combined chemical mapping, 3D nanotomography, and diffraction showed fine matrix-hosted magnetite has paleomagnetic recording potential comparable to magnetite framboids but is less susceptible to strong interactions; transient heating also reduced the contribution of matrix pyrrhotite through its vacancy-ordering state. This is one meteorite case study, and the mineralogical conclusions draw on the full microscopy and diffraction workflow, not AI alone.",
    "field": "physics",
    "ai_technology": "UNET segmentation, SIGMA interactive machine-learning mineral analysis, and few-shot image segmentation",
    "organization": "University of Cambridge; and collaborators",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GC012882",
    "source_type": "paper",
    "publication_date": "2026-07-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025GC012882",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A multiscale microscopy study of carbonaceous chondrite WIS 91600 used a deep-learning classifier to segment magnetic remanence-carrying ensembles and quantify their volume and texture.",
      "unknown": "The source is a single meteorite case study. The AI classified/quantified image ensembles, while mineral identities and remanence interpretation use additional microscopy, chemical mapping, nanotomography, and diffraction; extrapolation to other carbonaceous chondrites requires more samples.",
      "ai_role": "Classified remanence-carrying ensembles in microscopy images and quantified their volumes and morphologies; complementary imaging and diffraction supported mineral identification and interpretation.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GC012882"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-a-machine-learning-assisted-tess-search-confirms-a-previously-unknown-hot-jupiter",
    "title": "A machine-learning-assisted TESS search confirms a previously unknown hot Jupiter",
    "summary": "The T16 pipeline processed 83.7 million TESS Cycle 1 full-frame-image light curves and reported 10,091 new transiting-planet candidates, alongside previously known candidates. Radial-velocity follow-up confirmed TIC 183374187 b as a newly identified hot Jupiter, demonstrating that the semi-automated search yielded at least one bona fide new planet. Most entries in the catalog remain candidates rather than confirmed planets.",
    "field": "physics",
    "ai_technology": "T16 semi-automated transit search with random-forest machine-learning vetting",
    "organization": "Princeton University; and collaborating institutions",
    "source_url": "https://arxiv.org/abs/2604.18579",
    "source_type": "paper",
    "publication_date": "2026-05-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-4365/ae5b6c",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The T16 pipeline processed 83.7 million TESS Cycle 1 full-frame-image light curves and reported 10,091 new transiting-planet candidates, alongside previously known candidates.",
      "unknown": "The 10,091 entries are planet candidates, most unconfirmed. The cited discovery claim here is limited to TIC 183374187 b, which the authors say was confirmed with radial-velocity follow-up; the pipeline is described as machine-learning-assisted, not fully autonomous.",
      "ai_role": "Machine-learning-assisted pipeline classified light curves and prioritized possible transit signals; human vetting and Magellan/PFS radial-velocity observations confirmed the highlighted planet.",
      "evidence_urls": [
        "https://arxiv.org/abs/2604.18579"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-machine-learning-x-ray-analysis-provides-bulk-evidence-for-a-bragg-glass-phase-in-pdxerte3",
    "title": "Machine-learning X-ray analysis provides bulk evidence for a Bragg-glass phase in PdₓErTe₃",
    "summary": "Researchers combined extensive X-ray diffraction with X-ray Temperature Clustering (X-TEC) to identify Bragg-glass signatures in the charge-density-wave material PdₓErTe₃. Their analysis extracted diffraction features across about 20,000 Brillouin zones and found a diverging correlation length and a phase diagram consistent with a Bragg-glass transition. The result provides bulk diffraction evidence for a phase that had been difficult to establish experimentally.",
    "field": "physics",
    "ai_technology": "X-ray Temperature Clustering (X-TEC), a machine-learning analysis of X-ray diffraction peaks",
    "organization": "Cornell University; Stanford University; Argonne National Laboratory; University of Illinois Urbana-Champaign",
    "source_url": "https://www.nature.com/articles/s41567-023-02380-1",
    "source_type": "paper",
    "publication_date": "2024-02-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41567-023-02380-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers combined extensive X-ray diffraction with X-ray Temperature Clustering (X-TEC) to identify Bragg-glass signatures in the charge-density-wave material PdₓErTe₃.",
      "unknown": "The authors frame the result as evidence for the Bragg-glass phase based on characteristic diffraction behavior, rather than direct imaging of all underlying defects. Machine learning enabled the scale of peak extraction; physical interpretation and measurements were done by the team.",
      "ai_role": "Automatically extracted and clustered charge-density-wave diffraction peaks across temperature and momentum, enabling high-throughput estimation of peak spread/correlation length from a large set of Brillouin zones; researchers interpreted these features as evidence for Bragg-glass order.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41567-023-02380-1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-10-machine-learning-image-search-helps-identify-pavo-an-isolated-star-forming-dwarf-galaxy",
    "title": "Machine-learning image search helps identify Pavo, an isolated star-forming dwarf galaxy",
    "summary": "A search combining low-surface-brightness galaxy detection and machine-learning candidate classification identified Pavo in Dark Energy Camera Legacy Survey imaging. Follow-up Magellan imaging resolved young and old stellar populations and measured a tip-of-the-red-giant-branch distance of about 1.99 Mpc, establishing Pavo as an extremely isolated, low-mass star-forming dwarf galaxy just beyond the Local Group.",
    "field": "physics",
    "ai_technology": "Machine-learning image candidate classifier used with low-surface-brightness galaxy search algorithms",
    "organization": "University of Arizona; University of Chicago; University of Minnesota Duluth; and collaborators",
    "source_url": "https://arxiv.org/abs/2310.01478",
    "source_type": "paper",
    "publication_date": "2023-10-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/2041-8213/ad0130",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A search combining low-surface-brightness galaxy detection and machine-learning candidate classification identified Pavo in Dark Energy Camera Legacy Survey imaging.",
      "unknown": "The AI contributed candidate classification within a broader image-search workflow. The discovery was verified by follow-up observations and astrophysical analysis. AAS conference abstract from 2024 also describes this same Pavo discovery and is not a separate result.",
      "ai_role": "Classified image-search candidates in DECaLS imaging; the search team selected Pavo for follow-up, and telescope imaging and color-magnitude analysis established its distance and galaxy properties.",
      "evidence_urls": [
        "https://arxiv.org/abs/2310.01478"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-deep-learning-reveals-more-than-500-previously-unrecognized-earthquakes-beneath-david-glacier",
    "title": "Deep learning reveals more than 500 previously unrecognized earthquakes beneath David Glacier",
    "summary": "A deep-learning analysis of regional Antarctic seismic records identified more than 500 previously unrecognized intermediate-depth earthquakes roughly 100–150 km beneath David Glacier in East Antarctica. The earthquakes occur within a tectonic plate rather than at a plate boundary. The authors explain the concentration through bending stresses associated with variable thermally driven uplift and an abrupt change in lithospheric strength; the ice sheet may also affect regional stress.",
    "field": "physics",
    "ai_technology": "Deep-learning seismic-event detector applied to regional seismic records",
    "organization": "University of Alabama; Complutense University of Madrid; University of Oklahoma",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42207895/",
    "source_type": "paper",
    "publication_date": "2026-05-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.aea9895",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A deep-learning analysis of regional Antarctic seismic records identified more than 500 previously unrecognized intermediate-depth earthquakes roughly 100–150 km beneath David Glacier in East Antarctica.",
      "unknown": "The discovery concerns a newly recognized concentration of intraplate events, not a new earthquake mechanism conclusively proven. Proposed stress sources are interpretations; the paper notes similar events elsewhere may remain undetected.",
      "ai_role": "Detected intermediate-depth earthquake signals in regional seismic data that were not recognized in prior catalogs; the authors then located and interpreted the events using seismological and geodynamic analysis.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42207895/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-swin-transformer-discovery-of-planetary-nebulae-with-spectroscopic-confirmation-in-vphas",
    "title": "Swin-Transformer discovery of planetary nebulae with spectroscopic confirmation in VPHAS+",
    "summary": "A Swin-Transformer search of VPHAS+ Hα images identified faint, resolved planetary-nebula candidates in crowded Galactic fields that prior point-source searches missed. Follow-up spectroscopy and HASH classification of 31 targets found 16 true, two likely and four possible planetary nebulae, alongside contaminants and a supernova-remnant fragment. This is one validated search-sample result, not 22 equally secure confirmations.",
    "field": "physics",
    "ai_technology": "Swin-Transformer deep-learning classifier trained on known planetary nebulae",
    "organization": "The University of Hong Kong; Taiyuan University of Technology; Peng Cheng Laboratory",
    "source_url": "https://www.aanda.org/articles/aa/full_html/2024/12/aa50998-24/aa50998-24.html",
    "source_type": "paper",
    "publication_date": "2024-12-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202450998",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Swin-Transformer search of VPHAS+ Hα images identified faint, resolved planetary-nebula candidates in crowded Galactic fields that prior point-source searches missed.",
      "unknown": "The 31 follow-up targets were selected from more than 800 high-quality candidates, so the sample is not a completeness or population estimate. Spectroscopic classifications comprise 16 true, two likely and four possible PNe; these confidence levels are not merged into a claim that all 22 are confirmed. Two non-emitter late-type contaminants and other emission-line contaminants were found. The result counts the paper-level AI-assisted search and validation campaign once.",
      "ai_role": "Scanned survey images and ranked candidate planetary nebulae; follow-up spectroscopy and standard HASH classification established which selected targets were nebulae and which were contaminants.",
      "evidence_urls": [
        "https://www.aanda.org/articles/aa/full_html/2024/12/aa50998-24/aa50998-24.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-06-active-learning-identifies-28-new-dft-validated-pressure-induced-phase-transitions",
    "title": "Active learning identifies 28 new DFT-validated pressure-induced phase transitions",
    "summary": "Across 13 active-learning cycles, a graph neural network screened phase pairs for pressure-induced transitions and selected cases for density-functional-theory (DFT) calculations. DFT confirmed 28 previously unreported transformations; 14 high-pressure phases were dynamically stable at transition pressure, including 10 also predicted dynamically stable at zero pressure. The study also identifies a previously unrecorded dynamically stable monoclinic P2₁/c phase of Cr₂O₃.",
    "field": "materials",
    "ai_technology": "MEGNet graph neural network with five-model ensemble and iterative active learning",
    "organization": "Purdue University",
    "source_url": "https://www.nature.com/articles/s41524-025-01682-7",
    "source_type": "paper",
    "publication_date": "2025-06-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-025-01682-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Across 13 active-learning cycles, a graph neural network screened phase pairs for pressure-induced transitions and selected cases for density-functional-theory (DFT) calculations.",
      "unknown": "These are computational predictions corroborated by DFT, not experimentally synthesized or observed phases. Of 28 transitions, 14 high-pressure structures were dynamically stable at transition pressure; only 10 were also predicted stable at zero pressure. Two selected phase-pair calculations remained difficult to converge. The paper's discovery count excludes 18 rediscoveries, and individual stability remains subject to DFT approximations.",
      "ai_role": "The GNN screened 7,677 phase pairs for consensus transition-pressure predictions. Active learning selected high-confidence pairs for DFT equation-of-state calculations, and the resulting DFT outcomes were added to training data for subsequent cycles. The AI search was the discovery and prioritization mechanism; DFT calculations supplied the physical verification of the transitions.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01682-7"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-representation-learning-uncovers-the-extragalactic-x-ray-transient-xrt-200515-in-chandra-archives",
    "title": "Representation learning uncovers the extragalactic X-ray transient XRT 200515 in Chandra archives",
    "summary": "A representation-learning pipeline searched archived Chandra event data and identified XRT 200515, a previously unnoticed extragalactic fast X-ray transient in the Large Magellanic Cloud. Its light curve has a sharp sub-10-second hard burst followed by an approximately 800-second oscillating tail with spectral softening. The paper discusses possible origins including a low-energy giant magnetar flare or a Type I burst from a faint low-mass X-ray binary; the source's physical nature is unresolved.",
    "field": "physics",
    "ai_technology": "Unsupervised representation-learning pipeline using sparse autoencoders, dimensionality reduction and density-based clustering",
    "organization": "Stanford University; Harvard-Smithsonian Center for Astrophysics; Smithsonian Astrophysical Observatory; collaborators",
    "source_url": "https://arxiv.org/abs/2412.01150",
    "source_type": "paper",
    "publication_date": "2024-12-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stae2808",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A representation-learning pipeline searched archived Chandra event data and identified XRT 200515, a previously unnoticed extragalactic fast X-ray transient in the Large Magellanic Cloud.",
      "unknown": "The transient is a real source-reported observational event, but its astrophysical progenitor remains unresolved. The study also catalogs candidate flares and dips; those catalog candidates are not counted as separately confirmed discoveries.",
      "ai_role": "Converted Chandra photon event files into time-energy representations, learned low-dimensional embeddings and clustered similar events; anomalous transient-dominated clusters surfaced XRT 200515 for subsequent human inspection and astrophysical analysis.",
      "evidence_urls": [
        "https://arxiv.org/abs/2412.01150"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-data-driven-equations-reveal-rainfall-linked-north-atlantic-decadal-variability",
    "title": "Data-driven equations reveal rainfall-linked North Atlantic decadal variability",
    "summary": "Using SINDy equation discovery on monthly observational indices from 1950–2022, researchers derived three linked low-order equations for Atlantic Multidecadal Variability, the North Atlantic Oscillation and North Atlantic precipitation. The model recovered a distinct roughly 20-year component and highlighted precipitation-linked terms as robust components of the learned dynamics; tests against the independent 1900–1939 record found skill for statistics of the variability indices and rainfall. The equations generate hypotheses about ocean-atmosphere feedbacks; they do not establish a single causal mechanism or guarantee forecasts.",
    "field": "climate",
    "ai_technology": "Sparse Identification of Nonlinear Dynamics (SINDy) equation-discovery algorithm",
    "organization": "University of Oxford; UK Centre for Ecology & Hydrology; UK Met Office",
    "source_url": "https://esd.copernicus.org/articles/17/1061/2026/",
    "source_type": "paper",
    "publication_date": "2026-08-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.5194/esd-17-1061-2026",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using SINDy equation discovery on monthly observational indices from 1950–2022, researchers derived three linked low-order equations for Atlantic Multidecadal Variability, the North Atlantic Oscillation and North Atlantic precipitation.",
      "unknown": "The learned equations are compact models/hypotheses from observational associations. The paper and Oxford source note that proposed mechanisms need further evaluation; reported hindcast skill is not prospective climate prediction validation.",
      "ai_role": "Searched candidate dynamical models against observational time series and selected sparse, interpretable coupled equations; the researchers analyzed robust equation terms and compared the resulting model against a historical holdout period.",
      "evidence_urls": [
        "https://esd.copernicus.org/articles/17/1061/2026/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-lhcb-reports-first-single-experiment-evidence-for-b0-decay",
    "title": "LHCb reports first single-experiment evidence for B⁰ → μ⁺μ⁻ decay",
    "summary": "Using 18 fb⁻¹ of Run 3 data and earlier LHCb data, the collaboration reports first single-experiment evidence for the rare B⁰→μ⁺μ⁻ decay, with a combined significance of 3.2σ and measured branching fraction (1.14 +0.42/−0.39 stat. +0.06/−0.05 syst.)×10⁻¹⁰. A boosted decision tree classified candidates and suppressed combinatorial background in the fit. This is evidence below the 5σ discovery threshold; results agree with Standard Model expectations. The date is the original preprint submission; CERN announced the result on 3 November 2026.",
    "field": "physics",
    "ai_technology": "LHCb boosted decision tree (BDT) candidate classifier and multivariate trigger classifiers",
    "organization": "LHCb Collaboration; CERN",
    "source_url": "https://arxiv.org/abs/2611.00230",
    "source_type": "paper",
    "publication_date": "2026-10-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Using 18 fb⁻¹ of Run 3 data and earlier LHCb data, the collaboration reports first single-experiment evidence for the rare B⁰→μ⁺μ⁻ decay, with a combined significance of 3.2σ and measured branching fraction (1.14 +0.42/−0.39 stat.",
      "unknown": "This is a preprint submitted to Physical Review X and reports 3.2σ evidence, not a 5σ observation. Earlier joint LHCb–CMS evidence was not confirmed by later Run 1–2 analyses; this result is the first single-experiment evidence, not the first ever claim. The paper reports all measurements as consistent with Standard Model expectations. Machine learning contributes to candidate selection/background control along with substantial detector upgrades.",
      "ai_role": "The primary analysis uses a BDT to suppress combinatorial background, scores candidates and divides the dimuon mass fit into five response bins to improve signal sensitivity; it also uses multivariate classifiers in the software trigger. CERN states that machine-learning background suppression, together with detector upgrades, enabled the analysis to isolate these rare signals. The measured evidence comes from the collaboration's collision data and statistical fit, not an AI prediction.",
      "evidence_urls": [
        "https://home.cern/lhcb-finds-evidence-of-one-in-ten-billion-particle-decay/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-12-machine-learning-reveals-two-previously-undetected-phases-in-rydberg-atom-quantum-simulator-data",
    "title": "Machine learning reveals two previously undetected phases in Rydberg-atom quantum-simulator data",
    "summary": "An interpretable hybrid unsupervised/supervised learning analysis of experimental snapshots from a programmable Rydberg-atom array mapped phase regions on a square lattice and identified signatures of two previously undetected phases: boundary-ordered and rhombic. The rhombic pattern had been predicted theoretically, so the result is its experimental-data signature rather than a wholly unanticipated theoretical phase; the boundary-ordered phase was revealed from edge correlations.",
    "field": "physics",
    "ai_technology": "Hybrid-correlation convolutional neural network (Hybrid-CCNN) with PCA and clustering",
    "organization": "Harvard University; Massachusetts Institute of Technology; Cornell University; collaborators",
    "source_url": "https://arxiv.org/abs/2112.10789",
    "source_type": "paper",
    "publication_date": "2021-12-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevResearch.5.023005",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An interpretable hybrid unsupervised/supervised learning analysis of experimental snapshots from a programmable Rydberg-atom array mapped phase regions on a square lattice and identified signatures of two previously undetected phases: boundary-ordered and rhombic.",
      "unknown": "The study reanalyzes experimental quantum-simulator snapshots; the rhombic phase was previously predicted by theory, while the paper reports its finite-size experimental manifestation. The paper characterizes phases using data and numerical interpretation, not bulk material synthesis.",
      "ai_role": "Unsupervised clustering exposed phase regions in experimental snapshots; interpretable supervised correlator networks refined boundaries and identified edge and multi-point correlation motifs characteristic of the previously undetected boundary-ordered and rhombic regions.",
      "evidence_urls": [
        "https://arxiv.org/abs/2112.10789"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-deep-learning-resolves-subtle-seismic-wave-changes-across-the-2016-norcia-earthquake-cycle",
    "title": "Deep learning resolves subtle seismic-wave changes across the 2016 Norcia earthquake cycle",
    "summary": "A seven-layer convolutional neural network classified raw seismic waveforms recorded near Italy’s 2016 Mw 6.5 Norcia earthquake as pre- or post-mainshock and separated time-to-failure intervals with over 90% test accuracy. The authors interpret the model’s performance, frequency dependence and null tests as evidence that local waveforms encode evolving attenuation and elastic properties consistent with progressive fault damage before rupture and postseismic recovery.",
    "field": "physics",
    "ai_technology": "Seven-layer convolutional neural network trained on local seismic waveforms",
    "organization": "Sapienza University of Rome; Pennsylvania State University",
    "source_url": "https://www.nature.com/articles/s41467-024-54153-w",
    "source_type": "paper",
    "publication_date": "2024-11-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-54153-w",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A seven-layer convolutional neural network classified raw seismic waveforms recorded near Italy’s 2016 Mw 6.5 Norcia earthquake as pre- or post-mainshock and separated time-to-failure intervals with over 90% test accuracy.",
      "unknown": "Authors studied one earthquake sequence and state that generalization to other tectonic settings and prospective real-time precursor detection remain untested. Related changes in elastic properties were already reported by field studies; the paper's new contribution is applying deep learning to infer them from this local waveform sequence.",
      "ai_role": "Classified the waveform signals by their temporal relation to the mainshock; controlled tests using incorrect mainshock times, distant stations and filtered frequencies helped expose subtle local seismic-wave property changes associated with the fault cycle.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-54153-w"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-03-three-spectroscopically-confirmed-z6-quasars-found-through-contrastive-learning",
    "title": "Three spectroscopically confirmed z≈6 quasars found through contrastive learning",
    "summary": "A self-supervised contrastive-learning search of Dark Energy Survey imaging isolated a group of 11 sources that included seven known z≈6 quasars. Follow-up Gemini-South and archival NTT spectroscopy confirmed three additional quasars at redshifts 5.94–6.07. One source, J0109−5424, shows evidence of a foreground object and is a lensed-quasar candidate, but its multiple-image status requires high-resolution imaging.",
    "field": "physics",
    "ai_technology": "Self-supervised contrastive-learning neural network trained on Dark Energy Survey imaging data",
    "organization": "University of Cambridge; Max Planck Institute for Astronomy; Gemini Observatory/NSF NOIRLab; INAF; collaborators",
    "source_url": "https://arxiv.org/abs/2403.17903",
    "source_type": "paper",
    "publication_date": "2024-03-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stae902",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A self-supervised contrastive-learning search of Dark Energy Survey imaging isolated a group of 11 sources that included seven known z≈6 quasars.",
      "unknown": "The three quasar identifications are spectroscopically confirmed, but the possible gravitational lens interpretation of J0109−5424 remains unconfirmed pending high-resolution imaging. The reported 91% efficiency refers to ten quasars among the eleven-source island, including seven previously known objects; three of the four previously unclassified targets were confirmed quasars. It is not a universal completeness estimate.",
      "ai_role": "Grouped sources into a learned embedding-space island that prioritized four unclassified candidates; spectroscopy confirmed three as quasars, including one source missed by conventional color selection because foreground light contaminated the photometry.",
      "evidence_urls": [
        "https://arxiv.org/abs/2403.17903"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-ai-enhanced-seismic-catalog-exposes-a-ring-fault-system-at-campi-flegrei",
    "title": "AI-enhanced seismic catalog exposes a ring-fault system at Campi Flegrei",
    "summary": "A retrained machine-learning workflow applied to continuous seismic records from 2022–2025 expanded the Campi Flegrei earthquake catalog from about 12,000 to more than 54,000 events. The more complete locations delineated a sharply defined, shallow caldera ring-fault system, illuminated previously unclear northern faults, and identified very shallow hybrid events likely associated with the hydrothermal system. The authors report no direct signature of upward magma migration in the analyzed data.",
    "field": "physics",
    "ai_technology": "Retrained machine-learning seismic detection and phase-picking workflow developed at Stanford",
    "organization": "Stanford University; Istituto Nazionale di Geofisica e Vulcanologia – Osservatorio Vesuviano; University of Naples Federico II",
    "source_url": "https://doi.org/10.1126/science.adw9038",
    "source_type": "paper",
    "publication_date": "2025-09-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A retrained machine-learning workflow applied to continuous seismic records from 2022–2025 expanded the Campi Flegrei earthquake catalog from about 12,000 to more than 54,000 events.",
      "unknown": "The observations cover 21 January 2022–20 March 2025. The inferred hydrothermal source is described as likely, and the mapped faults do not establish imminent eruption or earthquake timing. Full Science article text was access-restricted in this review; the original abstract and reporting institution's article were read.",
      "ai_role": "Detected and located small earthquakes missed by the prior monitoring catalog, providing denser observations that made the ring-fault geometry and shallow seismic structures visible.",
      "evidence_urls": [
        "https://doi.org/10.1126/science.adw9038"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-machine-learning-guided-synthesis-confirms-superconductivity-in-mo20re6si4",
    "title": "Machine-learning-guided synthesis confirms superconductivity in Mo20Re6Si4",
    "summary": "Researchers used a deep-learning composition model to prioritize a sigma-phase ternary compound, synthesized Mo20Re6Si4, and measured bulk superconductivity with a critical temperature of 5.4 K. Magnetic susceptibility and a specific-heat anomaly support the bulk transition. The paper reports mixed model performance on its small test set and describes the model as guidance alongside chemical intuition.",
    "field": "materials",
    "ai_technology": "Deep-learning neural network that classifies compositions as superconducting and predicts critical temperature",
    "organization": "Rutgers University; Flatiron Institute; University of South Carolina; Princeton University; Michigan State University",
    "source_url": "https://arxiv.org/abs/2412.13012",
    "source_type": "paper",
    "publication_date": "2024-12-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Researchers used a deep-learning composition model to prioritize a sigma-phase ternary compound, synthesized Mo20Re6Si4, and measured bulk superconductivity with a critical temperature of 5.4 K.",
      "unknown": "The result is experimental and supported by bulk measurements, but the preprint reports a small, specialized evaluation set and mixed predictor performance. Chemical intuition also guided candidate selection; AI's contribution was a search hint, not independent autonomous discovery. The compound is a materials-physics finding rather than a new general superconducting mechanism.",
      "ai_role": "Provided composition-based hints that helped researchers select Mo20Re6Si4 for synthesis; laboratory resistivity, magnetic-susceptibility and specific-heat measurements established the superconducting result.",
      "evidence_urls": [
        "https://arxiv.org/abs/2412.13012"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-semi-supervised-image-classifiers-lead-to-confirmation-of-a-new-lensed-quasar",
    "title": "Semi-supervised image classifiers lead to confirmation of a new lensed quasar",
    "summary": "Semi-supervised image classifiers trained on Pan-STARRS and DESI imagery surfaced five quasar-lens candidates for Keck follow-up. Spectroscopy confirmed GRALJ140833.73+042229.98 as a previously undiscovered gravitationally lensed quasar: a z=2.998 quasar lensed by a z=0.542 early-type galaxy. Three other targets were quasar-plus-star or quasar-plus-galaxy interlopers, and one remained unresolved.",
    "field": "physics",
    "ai_technology": "Semi-supervised image-classifier pipeline using a convolutional autoencoder with a classifier, plus a convolutional network trained with virtual adversarial training",
    "organization": "University of Sydney; University of California, Irvine; California Institute of Technology; collaborators",
    "source_url": "https://arxiv.org/abs/2504.00054",
    "source_type": "paper",
    "publication_date": "2026-03-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stag445",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Semi-supervised image classifiers trained on Pan-STARRS and DESI imagery surfaced five quasar-lens candidates for Keck follow-up.",
      "unknown": "Only one of five observed candidates was confirmed; three were interlopers and one could not be resolved. The successful identification came from a previous model iteration during model development, so this is an AI-assisted discovery with limited follow-up yield, not a prospective performance demonstration.",
      "ai_role": "Image-only classifiers prioritized candidate lenses; a prior iteration of the autoencoder-classifier surfaced the object, which Keck spectroscopy subsequently confirmed. The paper notes the observed candidates were selected during model development, not from a strictly prospective frozen-model test.",
      "evidence_urls": [
        "https://arxiv.org/abs/2504.00054"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-self-supervised-imaging-search-identifies-16-spectroscopically-confirmed-quasars-at-z5-946-45",
    "title": "Self-supervised imaging search identifies 16 spectroscopically confirmed quasars at z≈5.94–6.45",
    "summary": "Contrastive learning on DESI Legacy Survey DR10 multiband images, followed by SED-fit prioritization, selected sources for spectroscopy; follow-up confirmed 16 quasars at z=5.94–6.45. Several show uncommon narrow Lyα lines, strong Lyα+N V equivalent widths or mildly red observed-frame NIR slopes, and three lie in regions traditional color cuts would miss. The 16-object reported sample includes 15 new literature discoveries and one independently confirmed DESI DR1 source; the 45% follow-up yield includes previously known quasars among 40 observed targets.",
    "field": "physics",
    "ai_technology": "Self-supervised contrastive-learning neural network applied to DESI Legacy Survey DR10 multiband images, with SED-fit prioritization",
    "organization": "Max Planck Institute for Astronomy; Pontificia Universidad Católica de Chile; Jet Propulsion Laboratory; Universidad Diego Portales; University of Geneva; collaborators",
    "source_url": "https://www.aanda.org/articles/aa/full_html/2026/04/aa57039-25/aa57039-25.html",
    "source_type": "paper",
    "publication_date": "2026-04-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202557039",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Contrastive learning on DESI Legacy Survey DR10 multiband images, followed by SED-fit prioritization, selected sources for spectroscopy; follow-up confirmed 16 quasars at z=5.94–6.45.",
      "unknown": "Only 40 candidates were followed up from 1,139 high-priority candidates, so the reported 45% follow-up yield is not completeness or precision over the full sample. That yield counts 18 quasars among 40 targets, including previously known recoveries. The publisher abstract calls all 16 new; the paper body distinguishes 15 new literature discoveries and one additional candidate independently confirmed in DESI DR1 but not reported in literature at the time. Do not overstate population completeness.",
      "ai_role": "Learned an image embedding where candidate quasars formed overdensities, allowing atypical sources beyond standard color selection to be prioritized for spectroscopic validation. Spectroscopy confirmed the quasars; AI did not determine their redshifts alone.",
      "evidence_urls": [
        "https://www.aanda.org/articles/aa/full_html/2026/04/aa57039-25/aa57039-25.html",
        "https://arxiv.org/html/2603.08830"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-08-deep-learning-and-spectroscopy-confirm-four-faint-quasars-at-redshift-about-5",
    "title": "Deep learning and spectroscopy confirm four faint quasars at redshift about 5",
    "summary": "A deep-learning and Bayesian-information-criterion search of Subaru Hyper Suprime-Cam photometry selected four faint high-redshift quasar candidates. Palomar spectroscopy found broad Lyman-alpha emission and breaks in all four, confirming redshifts between 4.7 and 5.6; one object also received a black-hole mass estimate.",
    "field": "physics",
    "ai_technology": "Deep-learning photometric classifier combined with Bayesian information criterion",
    "organization": "Seoul National University; Kyungpook National University; Kavli Institute for Astronomy and Astrophysics, Peking University",
    "source_url": "https://doi.org/10.5303/JKAS.2022.55.4.131",
    "source_type": "paper",
    "publication_date": "2022-08-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2208.00612",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A deep-learning and Bayesian-information-criterion search of Subaru Hyper Suprime-Cam photometry selected four faint high-redshift quasar candidates.",
      "unknown": "The paper spectroscopically confirms four targeted candidates, not the full 35-object catalog. The selection method combines deep learning with a Bayesian information criterion and conventional candidate filtering; AI was one component rather than the sole discovery mechanism.",
      "ai_role": "The classifier shortlisted high-redshift quasar candidates from survey photometry; human researchers selected four for observation and used telescope spectra to confirm all four as quasars.",
      "evidence_urls": [
        "https://doi.org/10.5303/JKAS.2022.55.4.131"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-10-the-qubrics-machine-learning-survey-spectroscopically-identifies-224-new-bright-high-redshift-quasars",
    "title": "The QUBRICS machine-learning survey spectroscopically identifies 224 new bright high-redshift quasars",
    "summary": "The QUBRICS survey applied a machine-learning selection pipeline to southern optical and infrared catalogs and followed candidates spectroscopically. The paper reports 224 newly identified bright quasars at z≥2.5, including 15 at z>4, and provides a large southern sample for cosmological observations.",
    "field": "physics",
    "ai_technology": "QUBRICS machine-learning photometric selection with Canonical Correlation Analysis redshift regression",
    "organization": "INAF Astronomical Observatory of Trieste; INAF Astronomical Observatory of Padova; Las Campanas Observatory; Sapienza University of Rome; other QUBRICS collaboration institutions",
    "source_url": "https://doi.org/10.3847/1538-4365/abafc1",
    "source_type": "paper",
    "publication_date": "2020-10-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2008.03865",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The QUBRICS survey applied a machine-learning selection pipeline to southern optical and infrared catalogs and followed candidates spectroscopically.",
      "unknown": "The 224 are cumulative QUBRICS discoveries across its pilot and follow-up publications. The paper estimates a 68% success rate for the current selection; contaminants include lower-redshift quasars/AGN. Five objects in the sample had already appeared in literature and were independently selected, so the total is a survey result, not 224 exclusive first publications in this paper.",
      "ai_role": "Machine-learning selection ranked photometric sources for follow-up; the team obtained and analyzed optical/infrared spectra to identify and measure redshifts. The paper reports the cumulative survey sample, not 224 objects newly found by a single autonomous run.",
      "evidence_urls": [
        "https://doi.org/10.3847/1538-4365/abafc1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-09-ai-assisted-seismic-mapping-reveals-segmented-fault-activity-at-the-discovery-transform-fault",
    "title": "AI-assisted seismic mapping reveals segmented fault activity at the Discovery transform fault",
    "summary": "A machine-learning earthquake-detection and relocation workflow applied to 2008 ocean-bottom seismometer records found that the western Discovery transform-fault strand may comprise seven patches with distinct seismicity. The authors also identify a seismicity-gap zone that they interpret as possibly linked to an active small off-ridge rift.",
    "field": "physics",
    "ai_technology": "PhaseNet phase picker, GaMMA phase association, and machine-learning-enhanced seismic event workflow",
    "organization": "University of California, Santa Cruz; Woods Hole Oceanographic Institution",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023GC011043",
    "source_type": "paper",
    "publication_date": "2023-09-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2023GC011043",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning earthquake-detection and relocation workflow applied to 2008 ocean-bottom seismometer records found that the western Discovery transform-fault strand may comprise seven patches with distinct seismicity.",
      "unknown": "The seven-patch segmentation and possible off-fault rift relationship are interpretations of one year of local 2008 seismic data, not a causal or universal transform-fault model. The location and tectonic interpretations are supported by the new catalog and follow-up cross-correlation relocation; the source describes the rift explanation as possible.",
      "ai_role": "The AI workflow detected P/S arrivals from noisy ocean-bottom recordings and associated event phases; cross-correlation relocation and geological interpretation by the research team produced the fault-pattern interpretation.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023GC011043"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-11-eqgpt-infers-a-previously-unreported-equation-for-steepening-surface-gravity-waves",
    "title": "EqGPT infers a previously unreported equation for steepening surface gravity waves",
    "summary": "A knowledge-guided generative model trained on mathematical handbook equations generated candidate PDEs, then evaluated and optimized them against data. In a real-data case study of strongly nonlinear surface gravity waves approaching breaking, the authors report an equation not previously reported for that process.",
    "field": "physics",
    "ai_technology": "EqGPT, a generative transformer using equation tokenization and a generation–evaluation–optimization loop",
    "organization": "China University of Petroleum (East China); Imperial College London; Southern University of Science and Technology; Hong Kong University of Science and Technology; University of Oxford",
    "source_url": "https://www.nature.com/articles/s41467-025-65114-2",
    "source_type": "paper",
    "publication_date": "2025-11-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-65114-2",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A knowledge-guided generative model trained on mathematical handbook equations generated candidate PDEs, then evaluated and optimized them against data.",
      "unknown": "The claimed new governing equation is an inference from a particular real-world laboratory dataset and remains a model-derived description, not a universally established law. Authors used human fluid-mechanics analysis; this was one new equation result within a methods paper that also rediscovers known PDEs.",
      "ai_role": "The model proposed free-form PDE candidates using structures learned from 221 handbook equations; data-driven fit evaluation selected a candidate for the breaking-wave dataset, after which researchers analyzed its physical interpretation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-65114-2"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-01-autonomous-x-ray-mapping-reveals-previously-unseen-block-copolymer-morphologies-including-a-nanoscale-ladder",
    "title": "Autonomous X-ray mapping reveals previously unseen block-copolymer morphologies, including a nanoscale ladder",
    "summary": "An autonomous experiment planner selected synchrotron X-ray measurements across a combinatorial block-copolymer blend sample. The resulting map and targeted electron microscopy revealed previously unreported alternating, skew/bilayer, and ladder-like self-assembled morphologies, and the authors infer a general mechanism of local subdomain partitioning.",
    "field": "materials",
    "ai_technology": "gpCAM autonomous experimentation framework for selecting high-value measurements",
    "organization": "Brookhaven National Laboratory Center for Functional Nanomaterials and National Synchrotron Light Source II; Lawrence Berkeley National Laboratory CAMERA",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9839324/",
    "source_type": "paper",
    "publication_date": "2023-01-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.add3687",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An autonomous experiment planner selected synchrotron X-ray measurements across a combinatorial block-copolymer blend sample.",
      "unknown": "The newly discovered results are nanoscale morphologies in one specific polymer-blend/template system; the paper does not show that each morphology is already useful in a device. AI selected measurements, while domain researchers performed electron-microscopy confirmation and interpretation.",
      "ai_role": "The algorithm autonomously chose where to collect synchrotron X-ray scattering data and updated its model online; researchers then used electron microscopy to validate and interpret the novel morphologies.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9839324/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-10-smdet-selected-ultracool-dwarf-candidates-include-two-objects-with-spectroscopically-confirmed-t-dwarf-types",
    "title": "SMDET-selected ultracool-dwarf candidates include two objects with spectroscopically confirmed T-dwarf types",
    "summary": "The SMDET image-level neural network identified 118 previously uncatalogued high-proper-motion ultracool-dwarf candidates in WISE time-series images. Follow-up near-infrared spectra for two selected objects yielded T5 and T3 classifications, confirming their estimated spectral types within one subtype.",
    "field": "physics",
    "ai_technology": "SMDET, a convolutional neural network for detecting faint moving sources in WISE time-series coadds",
    "organization": "Backyard Worlds: Planet 9 collaboration; NASA Goddard Space Flight Center; California Institute of Technology/IPAC; Northern Arizona University",
    "source_url": "https://iopscience.iop.org/article/10.3847/1538-3881/ad77d2",
    "source_type": "paper",
    "publication_date": "2024-10-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-3881/ad77d2",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The SMDET image-level neural network identified 118 previously uncatalogued high-proper-motion ultracool-dwarf candidates in WISE time-series images.",
      "unknown": "Most of the 118 objects remain candidates, and the study spectroscopically checks only two. Some candidates were also identified by citizen scientists, so the AI was part of a human/algorithm discovery process. The counted finding is limited to the two spectra and the candidate search result as one paper-level report.",
      "ai_role": "Scanned WISE image sequences and prioritized faint, fast-moving source candidates; researchers cross-checked catalogs, estimated photometric types and distances, and acquired spectra for two objects to verify their T-dwarf classifications.",
      "evidence_urls": [
        "https://iopscience.iop.org/article/10.3847/1538-3881/ad77d2"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-machine-learning-tremor-catalog-reveals-japan-trench-spatial-and-slow-slip-relationships",
    "title": "Machine-learning tremor catalog reveals Japan Trench spatial and slow-slip relationships",
    "summary": "A machine-learning monitoring system applied to eight years of S-net seafloor records detected seven times more tectonic tremors than the previous envelope-correlation catalog. The denser catalog resolved tremor activity extending along strike and dip, tremor patterns synchronized with slow-slip events, higher tremor energy rates near large-earthquake asperities, and a positive relationship between energy rate and recurrence interval; authors interpret the latter as evidence consistent with plate-boundary frictional heterogeneity.",
    "field": "physics",
    "ai_technology": "DiET convolutional neural network and GrASP graph-based station association, with hybrid envelope-correlation/energy-ratio hypocenter location",
    "organization": "Geological Survey of Japan, National Institute of Advanced Industrial Science and Technology; Tohoku University",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JB031348",
    "source_type": "paper",
    "publication_date": "2025-05-31",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025JB031348",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning monitoring system applied to eight years of S-net seafloor records detected seven times more tectonic tremors than the previous envelope-correlation catalog.",
      "unknown": "The paper reports a substantial catalog expansion and interprets energy-rate patterns as suggestive of frictional heterogeneity; this is not a prospective earthquake forecast or proof of a causal friction mechanism. The abstract states prior work had already observed complementary tremor/earthquake distributions in the region, so the new contribution is the expanded catalog and improved resolution of the relationships, not the first observation of any complementary distribution.",
      "ai_role": "The CNN classified station spectrograms as earthquake, tremor or noise; graph clustering associated station detections, enabling a higher-resolution catalog from continuous S-net data. Human researchers interpreted the spatial and temporal relationships.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JB031348"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-01-deep-learning-reveals-a-new-seismicity-distribution-along-the-kabaw-fault-in-central-myanmar",
    "title": "Deep learning reveals a new seismicity distribution along the Kabaw Fault in central Myanmar",
    "summary": "A deep-learning-based catalog pipeline processed 18 months of seismic-array data and located 1,891 shallow earthquakes, more than twice the 849 events in the prior manual catalog. The denser catalog shows that shallow Kabaw Fault seismicity disappears south of about 22.8°N while deeper events occur west of the fault, and it fills part of a previously under-sampled seismicity gap; the authors interpret this contrast as potentially reflecting along-strike variation in deformation associated with oblique Indian-plate subduction.",
    "field": "physics",
    "ai_technology": "DEPALORE pipeline using Earthquake Transformer, transfer-learned PickNet, REAL association, NonLinLoc location and HypoDD relocation",
    "organization": "Chinese Academy of Sciences; Zhejiang Ocean University; collaborators",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023GL105159",
    "source_type": "paper",
    "publication_date": "2024-01-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2023GL105159",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A deep-learning-based catalog pipeline processed 18 months of seismic-array data and located 1,891 shallow earthquakes, more than twice the 849 events in the prior manual catalog.",
      "unknown": "The source reports catalog-derived patterns and offers a tentative tectonic explanation; the catalog contains 1,891 localized events and location uncertainties of several kilometers. The AI enabled detections and resolution, while geological interpretation remains the authors’ analysis.",
      "ai_role": "Deep-learning detectors and phase-picking models extracted seismic arrivals from continuous array waveforms; association and location stages produced the higher-completeness catalog from which researchers interpreted fault patterns.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023GL105159"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-02-unsupervised-machine-learning-reveals-previously-missed-hydraulic-fracturing-seismicity-in-live-oak-county",
    "title": "Unsupervised machine learning reveals previously missed hydraulic-fracturing seismicity in Live Oak County",
    "summary": "An unsupervised repeating-signal detector, combined with template matching, exposed small earthquake bursts in a region of the Eagle Ford shale where conventional catalogs had detected little seismicity. The authors report that the new events temporally coincided with hydraulic-fracturing activity, and their regional analysis found detectable seismicity began after injection flux exceeded about two million barrels per month despite similar cumulative injection volume; this supports, but does not prove, a causal role for injection rate and is limited by sparse station coverage.",
    "field": "physics",
    "ai_technology": "Repeating Signal Detector (RSD), an unsupervised machine-learning waveform-family detector",
    "organization": "Miami University; University of Kansas",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022JB025436",
    "source_type": "paper",
    "publication_date": "2023-02-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2022JB025436",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An unsupervised repeating-signal detector, combined with template matching, exposed small earthquake bursts in a region of the Eagle Ford shale where conventional catalogs had detected little seismicity.",
      "unknown": "The RSD detections were manually inspected and some event locations were poorly constrained; the authors note station density limits detection and do not establish causation from temporal correlation alone. This paper extends an earlier Eagle Ford study but reports a distinct Live Oak County seismicity region and subsequent time-period catalog.",
      "ai_role": "RSD searched continuous seismograms without relying on preselected earthquake templates, clustered similar signals, and revealed small-magnitude event bursts missed by the previous catalog. Researchers manually inspected signals and analyzed their timing relative to hydraulic-fracturing records.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022JB025436"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-12-a-self-driving-laboratory-discovers-lens-and-grating-patterns-that-steer-spontaneous-emission",
    "title": "A self-driving laboratory discovers lens-and-grating patterns that steer spontaneous emission",
    "summary": "A self-driving nanophotonics experiment discovered that combining lens-like refractive-index curvature with grating-like spatial gradient can steer incoherent light emission from a semiconductor metasurface beyond conventional momentum-matching designs. Across a 74° field of view, optimized patterns produced average peak directivity 2.2 times that of sawtooth-grating controls, reaching 67% peak directivity; a fitted equation learner extracted a compact relationship from the experiments.",
    "field": "physics",
    "ai_technology": "Variational autoencoder, active-learning experimental planner and neural-network equation learner",
    "organization": "Sandia National Laboratories",
    "source_url": "https://www.nature.com/articles/s41467-025-66916-0",
    "source_type": "paper",
    "publication_date": "2025-12-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-66916-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A self-driving nanophotonics experiment discovered that combining lens-like refractive-index curvature with grating-like spatial gradient can steer incoherent light emission from a semiconductor metasurface beyond conventional momentum-matching designs.",
      "unknown": "The article reports local optima under a finite experiment budget, not a global optimum. Measurements were made on a specific GaAs/InAs quantum-dot metasurface; extensions to other emitters or devices remain untested in this study.",
      "ai_role": "The VAE generated pump-pattern configurations, an active-learning agent selected closed-loop physical experiments using measured photoluminescence, and an equation-learning network distilled observed structure–property relationships. The experimental setup measured and human researchers interpreted the results.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-66916-0"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-a-self-driving-lab-finds-a-polymer-shell-with-record-energy-absorbing-efficiency",
    "title": "A self-driving lab finds a polymer shell with record energy-absorbing efficiency",
    "summary": "A human-supervised self-driving laboratory selected and physically tested more than 25,000 additively manufactured polymer shell designs across an 11-dimensional parameter space. The campaign found a structure with 75.2% energy-absorbing efficiency, exceeding the paper’s previously reported maximum of 71.8% for balsa wood, and the authors derived transferable design heuristics from the resulting dataset. The result is a structural design for specified compression thresholds, not a universal material efficiency ranking.",
    "field": "materials",
    "ai_technology": "MAMA BEAR self-driving laboratory with Bayesian optimization and automated additive manufacturing/testing",
    "organization": "Boston University; U.S. Army Combat Capabilities Development Command Soldier Center",
    "source_url": "https://www.nature.com/articles/s41467-024-48534-4",
    "source_type": "paper",
    "publication_date": "2024-05-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-48534-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A human-supervised self-driving laboratory selected and physically tested more than 25,000 additively manufactured polymer shell designs across an 11-dimensional parameter space.",
      "unknown": "The result concerns generalized cylindrical shells made from a limited set of polymers and a defined nondimensional metric, whose value depends on threshold stress. The optimization was human-in-the-loop, and transfer to other geometries, materials or operational settings is not established.",
      "ai_role": "Bayesian optimization selected designs from over a trillion possible parameter combinations; the laboratory manufactured and mechanically tested candidates, while human researchers monitored and periodically modified the campaign.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-48534-4"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-machine-learning-helps-identify-61-new-cataclysmic-variables-in-sdss-v-spectroscopy",
    "title": "Machine learning helps identify 61 new cataclysmic variables in SDSS-V spectroscopy",
    "summary": "A convolutional neural network shortlisted cataclysmic-variable candidates from more than two million SDSS-V spectra, reducing the spectra requiring visual inspection to about 2% of the total. Human review and follow-up analysis of the resulting 504-object sample reported 61 previously unreported cataclysmic variables, spectroscopically confirmed 248 candidates and yielded 82 new or improved orbital periods; the article discusses selection biases and the CNN missed 27 spectra later found by human review.",
    "field": "physics",
    "ai_technology": "Convolutional neural network classifier for SDSS-V/BOSS spectra",
    "organization": "SDSS-V collaboration; University of Warwick and collaborating institutions",
    "source_url": "https://arxiv.org/abs/2406.19459",
    "source_type": "paper",
    "publication_date": "2024-06-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stae2524",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A convolutional neural network shortlisted cataclysmic-variable candidates from more than two million SDSS-V spectra, reducing the spectra requiring visual inspection to about 2% of the total.",
      "unknown": "This is one grouped paper-level survey result, not 61 separate records. The CNN was a triage aid rather than an autonomous confirmation system; authors report completeness/selection limitations and 27 misclassified spectra discovered through human inspection.",
      "ai_role": "Ranked over two million spectra for likely cataclysmic variables, reducing the manual-inspection workload; human investigators visually inspected candidate spectra and established source classifications and periods.",
      "evidence_urls": [
        "https://arxiv.org/abs/2406.19459"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-05-spectroscopic-follow-up-revises-gaia-machine-learning-white-dwarf-classifications-and-reveals-a-large-magnetic-and-carbon-rich-population",
    "title": "Spectroscopic follow-up revises Gaia machine-learning white-dwarf classifications and reveals a large magnetic and carbon-rich population",
    "summary": "Follow-up spectra for 255 Gaia white dwarfs were compared with machine-learning classifications from low-resolution Gaia spectra. The study reports that only 5 of 112 objects classified as massive DBs were genuine DB white dwarfs; most were magnetic white dwarfs or warm DQ stars. Its spectroscopy also catalogued 63 magnetic white dwarfs and 29 warm DQs in the selected sample, while confirming high classification accuracy for major spectral types within the classifier’s training set.",
    "field": "physics",
    "ai_technology": "Random-forest spectral classifier applied to Gaia low-resolution spectra",
    "organization": "Universitat Politècnica de Catalunya; Universitat de Barcelona and collaborators",
    "source_url": "https://arxiv.org/abs/2605.16493",
    "source_type": "paper",
    "publication_date": "2026-05-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202659763",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Follow-up spectra for 255 Gaia white dwarfs were compared with machine-learning classifications from low-resolution Gaia spectra.",
      "unknown": "The result is a follow-up characterization of a selected 500-pc sample, not a complete census of white dwarfs. It corrects a specific machine-learning-labelled ‘massive DB’ subpopulation; reported counts should not be extrapolated to Gaia’s full white-dwarf population.",
      "ai_role": "The classifier assigned initial white-dwarf spectral types and selected objects of interest; astronomers obtained Gran Telescopio Canarias medium-resolution spectra and visually inspected them to determine the objects’ types and magnetic features.",
      "evidence_urls": [
        "https://arxiv.org/abs/2605.16493"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-unsupervised-learning-reveals-mixed-frequency-seismic-precursors-before-axial-seamounts-2015-eruption",
    "title": "Unsupervised learning reveals mixed-frequency seismic precursors before Axial Seamount’s 2015 eruption",
    "summary": "An analysis of four months of Axial Seamount seismicity combined supervised event discrimination with unsupervised spectral clustering and found a mixed-frequency earthquake class whose counts rose rapidly about 15 hours before the 2015 eruption. The events migrated along mapped eruptive fissures; the authors interpret them as brittle crack opening associated with magma or volatile influx. This is a retrospective single-eruption finding, not an operationally validated forecast.",
    "field": "physics",
    "ai_technology": "Unsupervised machine-learning spectral clustering, with supervised event discrimination",
    "organization": "Lamont-Doherty Earth Observatory, Columbia University; University of Washington; Chinese University of Hong Kong",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024GL108631",
    "source_type": "paper",
    "publication_date": "2024-09-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2024GL108631",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An analysis of four months of Axial Seamount seismicity combined supervised event discrimination with unsupervised spectral clustering and found a mixed-frequency earthquake class whose counts rose rapidly about 15 hours before the 2015 eruption.",
      "unknown": "The result is retrospective and concerns a single 2015 eruption at one well-instrumented submarine volcano. The proposed crack-opening/magma-or-volatile explanation is the authors’ interpretation; applicability to future eruptions or other volcanoes remains untested.",
      "ai_role": "Discriminated volcanic earthquakes from other signal sources and clustered waveform spectral features, revealing a previously unrecognized mixed-frequency event pattern and its timing and migration before eruption.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024GL108631"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-08-machine-learning-identifies-regional-low-magnitude-seismic-unrest-before-the-ridgecrest-and-anchorage-earthquakes",
    "title": "Machine learning identifies regional low-magnitude seismic unrest before the Ridgecrest and Anchorage earthquakes",
    "summary": "A random-forest method using statistical features from earthquake catalogs identified abnormal low-magnitude seismicity spreading across roughly 15–25% of Southern California and Southcentral Alaska up to about three months before the 2019 Ridgecrest sequence and 2018 Anchorage M7.1 earthquake. Finite-element mechanics models were used to propose a pore-fluid-pressure explanation. This is a retrospective analysis of two target events, not a demonstrated reliable warning system.",
    "field": "physics",
    "ai_technology": "Random-forest machine-learning algorithm with earthquake-catalog statistical features",
    "organization": "Institute of Geophysics, Spanish National Research Council (CSIC); University of Edinburgh",
    "source_url": "https://www.nature.com/articles/s41467-024-51596-z",
    "source_type": "paper",
    "publication_date": "2024-08-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-51596-z",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A random-forest method using statistical features from earthquake catalogs identified abnormal low-magnitude seismicity spreading across roughly 15–25% of Southern California and Southcentral Alaska up to about three months before the 2019 Ridgecrest sequence and 2018 Anchorage M7.1 earthquake.",
      "unknown": "The study analyzes two retrospective target events and explicitly notes that earthquake precursors remain ambiguous and reliable alert systems are not established. The pore-pressure mechanism is proposed from modeling rather than directly measured, and the finding is not prospective validation.",
      "ai_role": "Detected regional nonlinear patterns in low-magnitude seismic catalogs that the authors report as precursory unrest before two large earthquake sequences; the AI analysis identified the statistical pattern, while finite-element models supported a proposed mechanism.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-51596-z"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-04-weakly-supervised-deep-learning-maps-seamount-morphology-patterns-associated-with-tectonic-setting",
    "title": "Weakly supervised deep learning maps seamount morphology patterns associated with tectonic setting",
    "summary": "A weakly supervised deep-learning analysis combining satellite-derived gravity and bathymetry produced a global seamount inventory and reported systematic morphology differences: broad edifices along fast-spreading ridges and steeper forms near intraplate hotspots. The authors interpret these differences as reflecting geodynamic controls on volcanic construction. Many detections rely on indirect gravity/bathymetry data rather than shipborne mapping, so individual feature identifications and causal interpretations require further validation.",
    "field": "physics",
    "ai_technology": "Weakly supervised deep-learning detector with vision foundation model, iterative pseudo-label refinement, and elliptical Gaussian fitting",
    "organization": "University of Science and Technology of China; Lawrence Berkeley National Laboratory",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JH000848",
    "source_type": "paper",
    "publication_date": "2026-04-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025JH000848",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A weakly supervised deep-learning analysis combining satellite-derived gravity and bathymetry produced a global seamount inventory and reported systematic morphology differences: broad edifices along fast-spreading ridges and steeper forms near intraplate hotspots.",
      "unknown": "The publisher abstract describes numerous previously undocumented features but does not establish shipborne confirmation for every mapped detection. The reported morphology-setting relationship is an observational pattern and should not be read as proof of a unique causal mechanism.",
      "ai_role": "Detected and morphologically characterized seamount features in global gravity and bathymetry data despite sparse labels, enabling the cross-setting comparison that revealed tectonically organized morphology patterns.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JH000848"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-08-machine-learning-identifies-predicted-zero-and-extreme-thermal-expansion-monolayers",
    "title": "Machine learning identifies predicted zero- and extreme-thermal-expansion monolayers",
    "summary": "High-throughput calculations and symbolic regression identified mechanical descriptors that classify positive- versus negative-thermal-expansion 2D crystals, then used those descriptors to predict monolayers with near-zero thermal expansion (2D Invar) and unusually large positive or negative expansion (anti-Invar). The paper reports multiple specific predicted crystals and explains the phonon/deformation-mode basis. The materials are computational predictions rather than experimentally synthesized monolayers.",
    "field": "materials",
    "ai_technology": "High-throughput calculations combined with symbolic regression machine learning",
    "organization": "Soochow University; Xi’an Jiaotong University",
    "source_url": "https://www.nature.com/articles/s41467-024-51379-6",
    "source_type": "paper",
    "publication_date": "2024-08-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-51379-6",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "High-throughput calculations and symbolic regression identified mechanical descriptors that classify positive- versus negative-thermal-expansion 2D crystals, then used those descriptors to predict monolayers with near-zero thermal expansion (2D Invar) and unusually large positive or negative expansion (anti-Invar).",
      "unknown": "The identified materials and thermal-expansion properties are from computational/high-throughput predictions and were not experimentally validated in this paper. This is a new 2D monolayer property result, distinct from previously cataloged bulk high-entropy Invar alloys.",
      "ai_role": "Learned interpretable relationships between in-plane tensile stiffness, out-of-plane bending stiffness and thermal-expansion class, then used them to screen 2D-crystal databases for predicted Invar and anti-Invar materials.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-51379-6"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-05-machine-learning-assisted-design-predicts-netsene-a-3d-carbon-nodal-surface-semimetal",
    "title": "Machine-learning-assisted design predicts Netsene, a 3D carbon nodal-surface semimetal",
    "summary": "A theory-guided inverse-design study combined a crystal diffusion variational autoencoder with a Crystal Transformer to propose a body-centered tetragonal C24 allotrope, Netsene. First-principles calculations report dynamical and mechanical stability and a nested nodal-surface semimetal band structure with drumhead surface states. This is a computationally predicted allotrope and electronic phase, not an experimentally synthesized material.",
    "field": "materials",
    "ai_technology": "Crystal diffusion variational autoencoder (CDVAE) and Crystal Transformer, guided by symmetry-engineering theory",
    "organization": "Authors’ institutions include Nanjing University and associated materials-physics groups (see arXiv record)",
    "source_url": "https://arxiv.org/abs/2605.22141",
    "source_type": "paper",
    "publication_date": "2026-05-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2605.22141",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A theory-guided inverse-design study combined a crystal diffusion variational autoencoder with a Crystal Transformer to propose a body-centered tetragonal C24 allotrope, Netsene.",
      "unknown": "The allotrope and its properties are computational predictions validated by first-principles calculations; the paper does not report synthesis or experimental observation. The finding is one proposed material/phase result from one paper, not a claim that AI alone established the physical existence of Netsene.",
      "ai_role": "Generated and ranked candidate crystal structures within a theory-guided inverse-design workflow; the resulting Netsene structure was then assessed with first-principles calculations for stability and topological band properties.",
      "evidence_urls": [
        "https://arxiv.org/abs/2605.22141"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-02-a-radio-survey-reveals-megaparsec-scale-diffuse-emission-around-a-cosmic-web-node",
    "title": "A radio survey reveals megaparsec-scale diffuse emission around a cosmic-web node",
    "summary": "Radio U-Net prompted reanalysis of LOFAR, XMM-Newton, DESI optical/near-infrared and weak-lensing data around cluster PSZ2 G083.29−31.03. The team reports previously unnoticed diffuse radio emission extending about 5 Mpc and aligned with X-ray emission and galaxies; lensing supports a complex mass distribution with two low-mass systems on either side of the cluster. The authors propose that these systems are merging substructures in a massive cosmic-web node, while noting alternative interpretations remain possible.",
    "field": "physics",
    "ai_technology": "Radio U-Net, a machine-learning detector for diffuse radio emission",
    "organization": "University of Bologna; INAF; University of Milano-Bicocca; National Central University, Taiwan; University of Tokyo; INAF Istituto di Radioastronomia; CINECA",
    "source_url": "https://arxiv.org/abs/2502.04823",
    "source_type": "paper",
    "publication_date": "2025-02-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202452581",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Radio U-Net prompted reanalysis of LOFAR, XMM-Newton, DESI optical/near-infrared and weak-lensing data around cluster PSZ2 G083.29−31.03.",
      "unknown": "The radio morphology and multiwavelength associations are reported observational results. The interpretation that the two faint components are subclusters merging with the central cluster is proposed rather than uniquely established; redshift/mass uncertainties and low radio resolution leave other emission classifications possible.",
      "ai_role": "Flagged diffuse low-surface-brightness radio emission in LoTSS data that had escaped prior human analysis, motivating the multiwavelength and weak-lensing study that characterized this new extended structure.",
      "evidence_urls": [
        "https://arxiv.org/abs/2502.04823"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-02-deep-learning-tracking-of-mantle-simulations-finds-continental-margin-subduction-zones-persist-longer",
    "title": "Deep-learning tracking of mantle simulations finds continental-margin subduction zones persist longer",
    "summary": "A fully convolutional network segmented subduction zones in two-dimensional mantle-convection simulations using temperature, vertical velocity and fineness fields. Across the modeled runs, zones forming near continental margins were generally longer-lived and less sensitive to internal heating than oceanic zones, supporting the hypothesis that continental lithosphere can stabilize subduction during early-Earth-like conditions. This is a computational-model result rather than direct observation of ancient Earth.",
    "field": "physics",
    "ai_technology": "Attention-based fully convolutional network with custom segmentation loss",
    "organization": "Pennsylvania State University; Georgia Institute of Technology",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2025JB032355",
    "source_type": "paper",
    "publication_date": "2026-02-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025JB032355",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A fully convolutional network segmented subduction zones in two-dimensional mantle-convection simulations using temperature, vertical velocity and fineness fields.",
      "unknown": "The physical inference comes from two-dimensional numerical mantle-convection models and their selected parameter space, not empirical reconstruction of Earth's early subduction history. The AI performs segmentation/tracking; the tectonic contrast is a statistical result from model outputs.",
      "ai_role": "Provided threshold-free, spatially contextual tracking of evolving subduction zones in simulation outputs; the resulting tracked histories enabled statistical comparison of lifetimes and sensitivity to mantle convective vigor by tectonic setting.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2025JB032355"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-04-machine-learning-maps-modeled-fault-system-co2-flux-within-active-tectonic-regions-worldwide",
    "title": "Machine learning maps modeled fault-system CO₂ flux within active tectonic regions worldwide",
    "summary": "A stacked ensemble trained on 867 compiled degassing records and geospatial raster predictors generated a continuous 0.10° map of model-estimated CO₂ flux within an active-tectonics applicability mask. The authors report spatially coherent high-flux estimates along the Andes, Pacific Ring of Fire and other active orogens. This is a new model-derived spatial estimate for fault-system flux, rather than a direct global survey or measurement.",
    "field": "climate",
    "ai_technology": "Stacked regression ensemble combining support-vector, k-nearest-neighbor, decision-tree, random-forest and gradient-boosting models",
    "organization": "Universidad Nacional de Colombia; Academia Colombiana de Ciencias Exactas",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2025JH000715",
    "source_type": "paper",
    "publication_date": "2026-04-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025JH000715",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A stacked ensemble trained on 867 compiled degassing records and geospatial raster predictors generated a continuous 0.10° map of model-estimated CO₂ flux within an active-tectonics applicability mask.",
      "unknown": "Predictions are model estimates based on 867 compiled records with heterogeneous dates and measurement techniques, not direct measurements at every pixel. The mask is limited to areas within 40 km of mapped active faults or with PGA ≥0.10 g; the 0.10° raster smooths local geology. The abstract reports R²=0.677, while the full text gives stacking test R²=0.693 and validation R²=0.552; fit is moderate. Sampling gaps include central Africa, eastern Asia and North America. High-flux belts overlap known tectonic/geothermal regions, and SHAP associations do not establish causal controls.",
      "ai_role": "Five regression learners were tuned and combined in a stacking regressor; that model predicted flux pixel by pixel from global raster inputs and generated the paper’s continuous map of estimated fault-system flux. The mask excluded areas outside the model’s stated tectonically active applicability domain.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2025JH000715"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-04-deep-learning-finds-comet-c-2020-t2-in-a-single-ztf-image",
    "title": "Deep learning finds comet C/2020 T2 in a single ZTF image",
    "summary": "The Tails EfficientDet model identified the extended object C/2020 T2 in one Zwicky Transient Facility Twilight Survey image on 2020 October 7. Follow-up orbital work confirmed it as a long-period comet. The study reports this as the first comet discovery assisted by deep learning; it also recovered a previously known comet, which is not counted as a new discovery here.",
    "field": "physics",
    "ai_technology": "Tails, an EfficientDet-based deep-learning comet detector",
    "organization": "California Institute of Technology; IPAC; Zwicky Transient Facility collaboration",
    "source_url": "https://arxiv.org/abs/2102.13352",
    "source_type": "paper",
    "publication_date": "2021-04-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-3881/abea7b",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The Tails EfficientDet model identified the extended object C/2020 T2 in one Zwicky Transient Facility Twilight Survey image on 2020 October 7.",
      "unknown": "The model surfaced the object, while human/standard follow-up established its orbit and comet classification. The entry counts only the newly discovered C/2020 T2, not recovery of P/2016 J3/P/2021 A3 or the detector’s performance metrics.",
      "ai_role": "Detected the comet’s extended coma/tail morphology in a single survey image and surfaced C/2020 T2 for discovery/follow-up; this enabled detection without requiring multi-epoch orbit linking at the initial search stage.",
      "evidence_urls": [
        "https://arxiv.org/abs/2102.13352"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-umap-analysis-of-desi-spectra-surfaces-newly-recognized-inhomogeneous-white-dwarfs",
    "title": "UMAP analysis of DESI spectra surfaces newly recognized inhomogeneous white dwarfs",
    "summary": "A machine-learning-guided analysis of DESI multi-epoch spectra used UMAP to identify three systems with strongly varying spectral signatures that were proposed as new ‘double-faced’ white dwarfs with inhomogeneous surface composition. Follow-up phase-resolved time-series spectra verified the interpretation for one system; the other two remain candidates. This is one paper-level result and only the verified system is counted as confirmed.",
    "field": "physics",
    "ai_technology": "UMAP dimensionality reduction applied to multi-epoch white-dwarf spectroscopy, alongside a neural-network spectral classifier",
    "organization": "University of Warwick; University of Leicester; University of Barcelona; Instituto de Astrofísica de Canarias; DESI collaboration",
    "source_url": "https://academic.oup.com/mnras/article/546/4/stag246/8466315",
    "source_type": "paper",
    "publication_date": "2026-02-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stag246",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning-guided analysis of DESI multi-epoch spectra used UMAP to identify three systems with strongly varying spectral signatures that were proposed as new ‘double-faced’ white dwarfs with inhomogeneous surface composition.",
      "unknown": "The paper describes three new candidates, but only one has validating phase-resolved time-series spectra; the remaining two have limited/unequal signal-to-noise and are not included as confirmed objects. Existing classification of the first source as a DA+DB binary was challenged, not assumed disproven without the authors’ supporting phase-dependent evidence.",
      "ai_role": "Mapped spectral/photometric features and flagged white dwarfs with strongly changing multi-epoch spectra for expert review; this identified the systems subsequently investigated with phase-resolved spectroscopy.",
      "evidence_urls": [
        "https://academic.oup.com/mnras/article/546/4/stag246/8466315"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-claude-assisted-analytic-proof-establishes-a-b-1-for-mean-field-jamming-exponents",
    "title": "Claude-assisted analytic proof establishes a+b=1 for mean-field jamming exponents",
    "summary": "Giorgio Parisi and Francesco Zamponi provide an analytic proof that the critical exponents a and b in the full replica-symmetry-breaking scaling solution for infinite-dimensional dense hard spheres obey a+b=1. The identity links gap, contact-force and overlap scaling relations and had previously been known numerically but not analytically. The authors report that Claude Opus 4.7 derived the core proof idea with minimal supervision, with Sonnet 4.6 helping refine it; the authors found and corrected errors in an early draft, then verified and edited the final proof.",
    "field": "mathematics",
    "ai_technology": "Anthropic Claude Opus 4.7 and Claude Sonnet 4.6",
    "organization": "Sapienza University of Rome; CNR-Nanotec Rome; INFN Sezione di Roma 1; Beihang University",
    "source_url": "https://doi.org/10.1088/1742-5468/ae7bd7",
    "source_type": "paper",
    "publication_date": "2026-07-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Giorgio Parisi and Francesco Zamponi provide an analytic proof that the critical exponents a and b in the full replica-symmetry-breaking scaling solution for infinite-dimensional dense hard spheres obey a+b=1.",
      "unknown": "The theorem is scoped to the fullRSB scaling equations for the infinite-dimensional dense hard-sphere jamming solution; it is not a general proof about every finite-dimensional jamming system. The result is peer-reviewed and published, but proof validation and correction are attributed to the human authors.",
      "ai_role": "After numerical investigation, Opus generated the analytic proof strategy and proof draft for the previously unproven scaling identity; Sonnet refined minor steps. The authors report 40 prompts, corrected early inconsistencies and verified/edited the published proof.",
      "evidence_urls": [
        "https://doi.org/10.1088/1742-5468/ae7bd7"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-09-ai-assisted-transfer-matrix-analysis-yields-an-exact-solution-for-the-1d-frustrated-q-state-potts-model",
    "title": "AI-assisted transfer-matrix analysis yields an exact solution for the 1D frustrated q-state Potts model",
    "summary": "The study exactly solves the one-dimensional J₁–J₂ q-state Potts model for arbitrary q by reducing its q²×q² transfer matrix to a 2×2 maximally symmetric subspace and mapping it to an effective q-state Potts model in a field. OpenAI’s o3-mini-high reasoning model helped derive the q=3 case; the author then generalized the result to arbitrary q and checked the formulas numerically. The solution nontrivially extends an older mapping known for q=2/Ising and yields q-dependent phase and residual-entropy behavior.",
    "field": "physics",
    "ai_technology": "OpenAI o3-mini-high reasoning model",
    "organization": "Brookhaven National Laboratory",
    "source_url": "https://journals.aps.org/prb/abstract/10.1103/y5vc-3t6q",
    "source_type": "paper",
    "publication_date": "2025-09-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/y5vc-3t6q",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study exactly solves the one-dimensional J₁–J₂ q-state Potts model for arbitrary q by reducing its q²×q² transfer matrix to a 2×2 maximally symmetric subspace and mapping it to an effective q-state Potts model in a field.",
      "unknown": "The AI contribution was to the q=3 transfer-matrix solution; the general-q theorem and full paper are the human author’s derivation, so this is an AI-assisted paper-level theoretical discovery, not an autonomous AI solution for arbitrary q. The exact result concerns the one-dimensional model and does not itself demonstrate a material realization.",
      "ai_role": "Proposed a transfer-matrix construction that enabled an exact solution of the q=3 frustrated Potts model; the author used that result as a starting point to generalize the exact mapping to arbitrary q, prove the subspace correspondence and numerically check the resulting formulas.",
      "evidence_urls": [
        "https://journals.aps.org/prb/abstract/10.1103/y5vc-3t6q"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-gemini-derives-an-exact-analytic-solution-for-the-cosmic-string-gravitational-radiation-power-integral",
    "title": "Gemini derives an exact analytic solution for the cosmic-string gravitational-radiation power integral",
    "summary": "A Gemini Deep Think system, guided by tree search and numerical feedback, derived closed-form spectral coefficients for the integral determining gravitational radiation power from cosmic-string loops, resolving the paper’s stated open calculation. The authors compare the expressions against direct numerical integration over ranges of loop harmonic and geometry parameters, finding close agreement. This is a theoretical/computational result, not a new astronomical detection.",
    "field": "physics",
    "ai_technology": "Google Gemini Deep Think integrated with tree search, executable Python evaluation and high-precision numerical feedback",
    "organization": "Affiliations are not recorded as an organization claim because the consulted arXiv page did not expose them in the checked abstract.",
    "source_url": "https://arxiv.org/abs/2603.04735",
    "source_type": "paper",
    "publication_date": "2026-03-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2603.04735",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A Gemini Deep Think system, guided by tree search and numerical feedback, derived closed-form spectral coefficients for the integral determining gravitational radiation power from cosmic-string loops, resolving the paper’s stated open calculation.",
      "unknown": "The source is an unrefereed arXiv preprint. The paper provides derivations and numerical cross-checks, not proof-assistant verification or independent replication. The first recurrence attempt contained an error that the later Gemini-assisted review corrected. The claim is limited to the stated integral for the cosmic-string radiation model and its exact analytic evaluation.",
      "ai_role": "Gemini generated and refined analytical approaches inside a tree-search workflow whose Python harness compared candidate expressions with high-precision numerical integration. A later human-initiated session with a more advanced Gemini model identified a recurrence error and telescoped the series into a finite analytic expression; this was a human–AI handoff, not a fully autonomous end-to-end result.",
      "evidence_urls": [
        "https://arxiv.org/abs/2603.04735"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-06-machine-learning-identifies-disordered-twist-angle-sequences-with-strongly-reduced-thermal-conductivity-in-multilayer-graphene",
    "title": "Machine learning identifies disordered twist-angle sequences with strongly reduced thermal conductivity in multilayer graphene",
    "summary": "Bayesian optimization and a CNN, coupled to non-equilibrium molecular dynamics, selected two-angle disordered twisted multilayer graphene structures with substantially lower simulated cross-plane thermal conductivity than aligned and periodic stacks. The optimized 14-layer sequence reached 0.095 W m−1 K−1 versus 0.512 W m−1 K−1 in aligned graphene in the reported model; phonon participation ratios and dynamical structure factors support strong partial phonon localization.",
    "field": "materials",
    "ai_technology": "Bayesian optimization (COMBO) and convolutional neural network coupled to molecular-dynamics simulations",
    "organization": "Sun Yat-sen University",
    "source_url": "https://www.nature.com/articles/s41524-025-01678-3",
    "source_type": "paper",
    "publication_date": "2025-06-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-025-01678-3",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Bayesian optimization and a CNN, coupled to non-equilibrium molecular dynamics, selected two-angle disordered twisted multilayer graphene structures with substantially lower simulated cross-plane thermal conductivity than aligned and periodic stacks.",
      "unknown": "This is a computational prediction supported by molecular-dynamics calculations, not a measured thermal-conductivity result from fabricated multilayer samples. The paper reports 80% reduction relative to aligned graphene for an optimized sequence and interprets phonon participation/structure-factor analyses as evidence of partial localization.",
      "ai_role": "Bayesian optimization selected low-conductivity twist sequences; the CNN screened the larger configuration space and proposed low-conductivity structures for subsequent molecular-dynamics evaluation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01678-3"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-09-machine-learning-analysis-attributes-the-apparent-pre-norcia-earthquake-b-value-rise-to-changing-activity-across-two-fault-structures",
    "title": "Machine-learning analysis attributes the apparent pre-Norcia earthquake b-value rise to changing activity across two fault structures",
    "summary": "For the 2016–2017 central Italy sequence, a combination of machine-learning techniques resolved two seismicity volumes with distinct fault orientations and Gutenberg–Richter b-values. Each structure retained its own b-value through the sequence; the apparent rise before the Mw 6.5 Norcia earthquake emerged when both were combined as their relative activity changed, showing that spatial heterogeneity can create an apparent temporal b-value signal.",
    "field": "physics",
    "ai_technology": "Combination of machine-learning techniques for seismicity structure and b-value analysis (paper abstract does not name a single model)",
    "organization": "University of Naples Federico II; Scuola Superiore Meridionale",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL116118",
    "source_type": "paper",
    "publication_date": "2025-09-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025GL116118",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "For the 2016–2017 central Italy sequence, a combination of machine-learning techniques resolved two seismicity volumes with distinct fault orientations and Gutenberg–Richter b-values.",
      "unknown": "The analysis is retrospective and specific to the 2016/2017 central Italy sequence; the authors state that b-value changes may reflect spatial redistribution as well as other processes and do not claim a usable prospective earthquake precursor.",
      "ai_role": "Machine-learning analysis separated seismicity into fault-associated volumes and enabled comparison of their distinct b-values over time, supporting the reported explanation of the aggregate temporal change.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL116118"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-machine-learning-phase-picks-constrain-the-2025-drake-passage-earthquake-depth-and-activated-fault-geometry",
    "title": "Machine-learning phase picks constrain the 2025 Drake Passage earthquake depth and activated fault geometry",
    "summary": "For the Mw 7.4 Drake Passage earthquake of 2 May 2025, a trained classifier identified depth-sensitive pP–sP seismic phases used to refine focal depths under sparse station coverage. The authors infer a mainshock depth of about 25 km, consistent with the small tsunami, and report aftershocks on west-dipping faults southeast of the epicenter near the Shackleton Fracture Zone; the paper integrates this with seismic-reflection and previously collected imaging data.",
    "field": "physics",
    "ai_technology": "Trained machine-learning classifier for depth-sensitive seismic phase identification",
    "organization": "Istituto Nazionale di Oceanografia e di Geofisica Sperimentale (OGS); CNR-ISMAR; University of Urbino; CONICET–University of Buenos Aires",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2026GL122178",
    "source_type": "paper",
    "publication_date": "2026-10-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2026GL122178",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "For the Mw 7.4 Drake Passage earthquake of 2 May 2025, a trained classifier identified depth-sensitive pP–sP seismic phases used to refine focal depths under sparse station coverage.",
      "unknown": "The result is a retrospective analysis of one earthquake sequence. The depth is inferred from phase picks and is consistent with tsunami observations; tectonic interpretation also relies on aftershock mechanisms and independent seismic imaging. The machine-learning component contributes phase identification rather than discovering the earthquake itself.",
      "ai_role": "Identified pP–sP phases in regional records; these phases were incorporated into event relocation and focal-depth determination, helping constrain the mainshock depth where station coverage is limited.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2026GL122178"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-02-optpde-identifies-new-candidate-integrable-pde-families-with-nontrivial-conserved-quantities",
    "title": "OptPDE identifies new candidate integrable PDE families with nontrivial conserved quantities",
    "summary": "Automatic differentiation optimized partial-differential-equation coefficients to increase the number of conserved quantities. The published study reports four PDE families with at least one nontrivial conserved quantity and analytically studies one new candidate, u_t = u_x^3; the work establishes candidate systems and conserved quantities, not that every generated family is fully integrable.",
    "field": "physics",
    "ai_technology": "OptPDE coefficient optimization with the CQFinder conserved-quantity search",
    "organization": "Massachusetts Institute of Technology",
    "source_url": "https://journals.aps.org/pre/abstract/10.1103/PhysRevE.111.025303",
    "source_type": "paper",
    "publication_date": "2025-02-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevE.111.025303",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Automatic differentiation optimized partial-differential-equation coefficients to increase the number of conserved quantities.",
      "unknown": "The peer-reviewed source reports four PDE families with at least one nontrivial conserved quantity and analytically examines one new candidate, u_t = u_x^3. The authors explicitly state that maximizing conserved quantities does not guarantee integrability; the result is a concrete candidate system and its studied properties, not proof that every output family is integrable.",
      "ai_role": "The authors explicitly describe OptPDE as a machine-learning approach. CQFinder computes the conserved-quantity objective from a symbolic PDE basis; OptPDE uses a sigmoid-smoothed objective and backpropagation to optimize 34 trainable PDE coefficients. They ran 5,000 random initializations and used dimensionality reduction to group the outputs into four families. This is a learned, trainable symbolic search, not simply use of an automatic-differentiation library.",
      "evidence_urls": [
        "https://journals.aps.org/pre/abstract/10.1103/PhysRevE.111.025303"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-07-machine-learning-identifies-sediment-derived-granites-in-the-hadean-zircon-record-supporting-early-subduction",
    "title": "Machine learning identifies sediment-derived granites in the Hadean zircon record, supporting early subduction",
    "summary": "A machine-learning classifier trained on trace-element fingerprints identified detrital zircon signatures from sediment-derived S-type granites as old as 4.24 billion years in the Jack Hills record. Across global detrital zircon data, the authors report long-term cyclic changes in S-type granite proportions and infer that exposed, weathered continental crust and subduction operated by about 4.2 Ga. The result bears on early Earth geodynamics, not a direct fossil or biological discovery.",
    "field": "physics",
    "ai_technology": "Trace-element Support Vector Machine (TSVM) classifier",
    "organization": "Chinese Academy of Sciences; University of Chinese Academy of Sciences; Macau University of Science and Technology; University of California, Davis; Queen’s University",
    "source_url": "https://www.pnas.org/doi/10.1073/pnas.2405160121",
    "source_type": "paper",
    "publication_date": "2024-07-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2405160121",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning classifier trained on trace-element fingerprints identified detrital zircon signatures from sediment-derived S-type granites as old as 4.24 billion years in the Jack Hills record.",
      "unknown": "The early-subduction interpretation is inferred from zircon trace elements and remains one interpretation of sparse Hadean geological evidence. The article reports that only the combination of early subaerial weathering and subduction explains the prevalence of the inferred S-type granites; it does not directly observe tectonics at 4.2 Ga.",
      "ai_role": "Classified zircon as S-type or non-S-type from multielement geochemical fingerprints, allowing the authors to identify the oldest sediment-derived granite signal and quantify its temporal distribution across detrital-zircon records.",
      "evidence_urls": [
        "https://www.pnas.org/doi/10.1073/pnas.2405160121"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-02-interpretable-machine-learning-finds-a-perpendicular-spin-signature-consistent-with-a-spinon-fermi-surface",
    "title": "Interpretable machine learning finds a perpendicular-spin signature consistent with a spinon Fermi surface",
    "summary": "A correlator convolutional neural network trained on projective snapshots from numerical simulations of the Kitaev–Heisenberg model identified a characteristic correlation pattern in the intermediate gapless phase when analyzing spin snapshots perpendicular to the applied field. The authors interpret the learned pattern as Friedel oscillations of gapless spinons forming a Fermi surface, supporting earlier theoretical proposals for this phase and suggesting measurable correlations for future experiments.",
    "field": "physics",
    "ai_technology": "Quantum-classical QuCl approach using a correlator convolutional neural network and regularization-path analysis",
    "organization": "Cornell University; The Ohio State University; Google Research; Harvard University; Ewha Womans University",
    "source_url": "https://www.nature.com/articles/s42005-024-01542-8",
    "source_type": "paper",
    "publication_date": "2024-02-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42005-024-01542-8",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A correlator convolutional neural network trained on projective snapshots from numerical simulations of the Kitaev–Heisenberg model identified a characteristic correlation pattern in the intermediate gapless phase when analyzing spin snapshots perpendicular to the applied field.",
      "unknown": "This is a numerical/theoretical result based on DMRG wavefunctions and simulated snapshots, not an experimental observation of a quantum spin liquid. The authors interpret the learned motif as evidence consistent with a spinon Fermi surface and offer it as a prediction to guide experiments; they do not claim it is already experimentally confirmed.",
      "ai_role": "Analyzed simulated projective-measurement snapshots and surfaced the phase signature: spin correlations in the direction perpendicular to the magnetic field, interpreted as the spatial pattern of Friedel oscillations associated with gapless spinons.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42005-024-01542-8"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-11-high-resolution-machine-learning-catalog-outlines-a-potential-unmapped-fault-in-the-weiyuan-shale-gas-region",
    "title": "High-resolution machine-learning catalog outlines a potential unmapped fault in the Weiyuan shale-gas region",
    "summary": "A machine-learning event detector combined with waveform cross-correlation produced a high-resolution catalog for injection-induced seismicity in the Weiyuan area, identifying one-third more earthquakes than the routine catalog and improving magnitude completeness. Spatial migration patterns show a pore-pressure diffusion front; a cluster outlines a potential unmapped fault that may have hosted the region’s largest recorded event, the September 2019 Mw 5.2 earthquake.",
    "field": "physics",
    "ai_technology": "Machine-learning earthquake detector combined with waveform cross-correlation (the abstract does not name a particular detector architecture)",
    "organization": "The Chinese University of Hong Kong; Sichuan Earthquake Agency",
    "source_url": "https://eppcgs.org/article/doi/10.26464/epp2021055?viewType=HTML",
    "source_type": "paper",
    "publication_date": "2021-11-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.26464/epp2021055",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning event detector combined with waveform cross-correlation produced a high-resolution catalog for injection-induced seismicity in the Weiyuan area, identifying one-third more earthquakes than the routine catalog and improving magnitude completeness.",
      "unknown": "The authors describe the fault as potential and state that it may have led to the September 2019 Mw 5.2 event; this is an interpretation, not a directly imaged fault or established causal proof. Results are from one field area and are limited by the station coverage.",
      "ai_role": "Detected additional earthquakes that enabled higher-resolution relocation and spatiotemporal mapping; authors use the denser catalog to infer a migrating pore-pressure front and identify a cluster outlining a potential fault trace.",
      "evidence_urls": [
        "https://eppcgs.org/article/doi/10.26464/epp2021055?viewType=HTML"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-06-eqtransformer-based-catalog-reveals-a-50-km-deep-outer-rise-fault-at-the-southern-mariana-trench",
    "title": "EQTransformer-based catalog reveals a 50-km-deep outer-rise fault at the Southern Mariana trench",
    "summary": "Applying EQTransformer phase detection to near-field ocean-bottom seismometer recordings collected near the Challenger Deep yielded more than 1,975 earthquakes. Relocation and waveform-similarity clustering outlined a southern Mariana outer-rise fault extending approximately 20–50 km below the seafloor (about 50 km maximum depth), inferred to be a reactivated normal fault; the depth suggests more incoming-plate water input than prior estimates.",
    "field": "physics",
    "ai_technology": "EQTransformer machine-learning seismic phase detector",
    "organization": "The Chinese University of Hong Kong; Chinese Academy of Sciences; Qingdao National Laboratory for Marine Science and Technology; Southern University of Science and Technology; Woods Hole Oceanographic Institution",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022GL097779",
    "source_type": "paper",
    "publication_date": "2022-06-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2022GL097779",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Applying EQTransformer phase detection to near-field ocean-bottom seismometer recordings collected near the Challenger Deep yielded more than 1,975 earthquakes.",
      "unknown": "The fault depth, fault type and hydration implications are inferred from a localized seismic cluster, waveform similarity, and regional tomography; focal mechanisms could not be robustly resolved, and the authors state that limited station coverage does not establish whether such deep faults are widespread.",
      "ai_role": "Detected and picked P- and S-wave phases in near-field ocean-bottom seismometer data, producing a denser earthquake catalog. Researchers manually reviewed selected phases, relocated the events and used waveform-similarity clustering to delineate the fault.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022GL097779"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-02-deep-learning-identifies-216-previously-unrecognized-green-pea-galaxies-in-desi-legacy-survey-imaging",
    "title": "Deep learning identifies 216 previously unrecognized Green Pea galaxies in DESI Legacy Survey imaging",
    "summary": "A MobileMamba-ECA image classifier searched DESI Legacy Imaging Surveys data and surfaced 775 Green Pea galaxy candidates. Cross-matching with DESI DR1 spectroscopy established 216 as previously unreported Green Pea galaxies: although spectra existed, the standard DESI pipeline had classified them only as generic galaxies. The study’s remaining 511 photometric-only objects are follow-up candidates and are not included in the confirmed count.",
    "field": "physics",
    "ai_technology": "MobileMamba-ECA deep-learning classifier (Mamba state-space backbone with efficient channel attention)",
    "organization": "Shandong University; National Astronomical Observatories, Chinese Academy of Sciences",
    "source_url": "https://academic.oup.com/mnras/article/547/1/stag278/8475333",
    "source_type": "paper",
    "publication_date": "2026-02-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stag278",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A MobileMamba-ECA image classifier searched DESI Legacy Imaging Surveys data and surfaced 775 Green Pea galaxy candidates.",
      "unknown": "The 216 newly identified galaxies have archival spectra and were classified with emission-line and BPT diagnostics; they were not all newly observed. The 511 sources without archival spectral coverage remain candidates and are excluded from this finding. The result is a newly recognized population/sample, not a claim that the galaxies themselves formed recently.",
      "ai_role": "Classified multiband DESI Legacy Survey images to prioritize compact Green Pea galaxy sources from an 843,592-object parent sample; this enabled spectral cross-identification and recognition of 216 previously unreported examples.",
      "evidence_urls": [
        "https://academic.oup.com/mnras/article/547/1/stag278/8475333"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-04-cnn-guided-spectroscopy-confirms-28-new-extremely-metal-poor-galaxies",
    "title": "CNN-guided spectroscopy confirms 28 new extremely metal-poor galaxies",
    "summary": "A three-stage convolutional-neural-network pipeline screened more than 7.7 million SDSS DR17 imaging sources lacking SDSS spectra, then selected targets for observation. Follow-up spectra of 45 candidates confirmed all as metal-poor galaxies, including 28 newly discovered extremely metal-poor systems; 18 have weak or undetected [N II] lines and oxygen-abundance upper limits.",
    "field": "physics",
    "ai_technology": "Three-stage CNN characterization pipeline (metal-poor galaxy classifier, XMP classifier, and N2-index predictor)",
    "organization": "University of California, Davis; Durham University; Isaac Newton Group of Telescopes; Southern Astrophysical Research Telescope",
    "source_url": "https://academic.oup.com/mnras/article/540/1/128/8121403",
    "source_type": "paper",
    "publication_date": "2025-04-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2502.01638",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A three-stage convolutional-neural-network pipeline screened more than 7.7 million SDSS DR17 imaging sources lacking SDSS spectra, then selected targets for observation.",
      "unknown": "The 28 new XMP classifications derive from follow-up of 45 targeted candidates, not the full 7.7-million-source imaging search. For 18 galaxies with weak or absent [N II], oxygen abundance is reported as a 2-sigma upper limit; the paper also reports four possible low-metallicity AGN candidates, which are not counted as confirmed here.",
      "ai_role": "The CNN pipeline ranked imaging-only sources using predicted spectral line ratios to nominate telescope targets; subsequent INT and SOAR spectroscopy established their metallicity classifications and confirmed 28 as new XMP galaxies.",
      "evidence_urls": [
        "https://academic.oup.com/mnras/article/540/1/128/8121403"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-msfa-net-and-balmer-line-fitting-identify-3-583-new-blue-horizontal-branch-stars",
    "title": "MSFA-Net and Balmer-line fitting identify 3,583 new blue horizontal-branch stars",
    "summary": "A two-stage deep-learning pipeline searched LAMOST DR12 spectra and retained 27,853 BHB candidate spectra. After removal of previously known objects, stringent Balmer-line-profile fitting, and deduplication, the authors report 3,583 unique newly identified blue horizontal-branch stars. Gaia photometry provides an additional consistency check.",
    "field": "physics",
    "ai_technology": "MSFA-Net, a multi-scale and frequency-attention deep neural network, followed by a binary refinement classifier",
    "organization": "National Astronomical Observatories, Chinese Academy of Sciences; University of Chinese Academy of Sciences",
    "source_url": "https://arxiv.org/abs/2606.29918",
    "source_type": "paper",
    "publication_date": "2026-06-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A two-stage deep-learning pipeline searched LAMOST DR12 spectra and retained 27,853 BHB candidate spectra.",
      "unknown": "This is an arXiv preprint. The paper says high-confidence objects were identified by Balmer-line profile fitting and Gaia color-magnitude consistency, while acknowledging that a fraction could still be contaminants; some atmospheric-parameter outliers are attributed to inference systematics. Counts are unique objects after deduplication, not 3,923 spectra.",
      "ai_role": "MSFA-Net searched the LAMOST DR12 hot-star spectral domain, used staged classification to reduce the spectra to candidate sets, and aided construction of the new sample; Balmer-line profile criteria and duplicate removal yielded the reported 3,583 unique objects. A separate SLAM model estimated atmospheric parameters rather than serving as the confirmation step.",
      "evidence_urls": [
        "https://arxiv.org/abs/2606.29918"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-wdnet-guided-desi-follow-up-confirms-75-new-white-dwarfs-and-7-cataclysmic-variables",
    "title": "WDNet-guided DESI follow-up confirms 75 new white dwarfs and 7 cataclysmic variables",
    "summary": "WDNet detected candidates in DESI Legacy Surveys DR9 images. The researchers cross-matched 18,242 photometric candidates with DESI DR1 spectra, obtaining 719 candidate spectra; manual visual inspection of those spectra identified 75 new white dwarfs and 7 new cataclysmic variables. These spectroscopic classifications, not the much larger image-candidate list, are the reported verified sample.",
    "field": "physics",
    "ai_technology": "WDNet, a deep-learning object-detection network based on YOLOv8 with multi-scale feature fusion",
    "organization": "Shandong University; National Astronomical Observatories, Chinese Academy of Sciences; Yunding Technology",
    "source_url": "https://www.mdpi.com/2218-1997/12/10/295",
    "source_type": "paper",
    "publication_date": "2026-09-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3390/universe12100295",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "WDNet detected candidates in DESI Legacy Surveys DR9 images.",
      "unknown": "The authors confirmed 75 new WDs and 7 new CVs among the 719 candidates with DESI spectra, and explicitly note substantial contamination among remaining candidates and selection effects in the spectroscopic subsample. No atmosphere-model fitting was done for all identifications; the authors recommend further high-S/N spectroscopic modeling. The article's abstract/results/conclusion give inconsistent counts for the full detected/candidate catalog (18,242, 61,624, and 63,214 are all reported in different sections); this entry relies only on the consistent 719 spectral matches and 75+7 inspected confirmations. The full photometric catalog is not counted as confirmed discoveries.",
      "ai_role": "WDNet detected and localized white-dwarf candidates in DESI Legacy Surveys photometric images, thereby prioritizing objects for DESI spectral cross-match and manual spectral inspection. Human inspection of DESI spectra supplied the classification confirmation; the neural network did not itself confirm the new stellar classes.",
      "evidence_urls": [
        "https://www.mdpi.com/2218-1997/12/10/295"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-03-deep-learning-enhanced-sparta-earthquake-catalog-resolves-a-shallow-multifault-rupture",
    "title": "Deep-learning-enhanced Sparta earthquake catalog resolves a shallow multifault rupture",
    "summary": "For the 2020 Mw 5.1 Sparta, North Carolina sequence, an EQTransformer-based catalog combined with matched-filter detections and differential-time relocations contains 1,761 high-precision events, about 5.8 times the original NEIC count. The denser locations resolve several shallow clusters and support the authors’ interpretation that the mainshock nucleated near the intersection of a blind strike-slip fault and a reverse fault associated with the surface rupture; moment-tensor inversion places its centroid at about 1.3 km depth.",
    "field": "physics",
    "ai_technology": "Earthquake Transformer (EQTransformer) deep-learning phase/event detector, followed by matched-filter detection and differential-time relocation",
    "organization": "Georgia Institute of Technology; University of Science and Technology of China; North Carolina State University; Chinese Academy of Sciences",
    "source_url": "https://www.nature.com/articles/s43247-024-01316-8",
    "source_type": "paper",
    "publication_date": "2024-03-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43247-024-01316-8",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "For the 2020 Mw 5.1 Sparta, North Carolina sequence, an EQTransformer-based catalog combined with matched-filter detections and differential-time relocations contains 1,761 high-precision events, about 5.8 times the original NEIC count.",
      "unknown": "This is a retrospective study of one 2020 earthquake sequence, using a local temporary-network deployment. The authors describe the inferred fault intersection/flower-structure interpretation as possible; the fit for the smallest cluster is poor and its geometry uncertain. The result is not a forecast or general fault model. The AI contribution was event detection within a hybrid workflow, not independent geological validation.",
      "ai_role": "EQTransformer detections expanded the template catalog, which seeded matched-filter detection; the resulting event set was relocated to improve temporal and spatial resolution used to infer the sequence’s shallow fault geometry. The fault interpretation is author analysis from the denser catalog plus waveform and moment-tensor methods, not a model-generated classification alone.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43247-024-01316-8"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-10-crystal-graph-learning-and-dft-identify-19-512-additional-thermodynamically-stable-materials",
    "title": "Crystal-graph learning and DFT identify 19,512 additional thermodynamically stable materials",
    "summary": "The authors trained crystal-graph attention networks on an expanded, composition- and symmetry-balanced dataset and used them to search 2,500 binary and ternary structure prototypes covering about one billion inorganic compounds. Density-functional-theory validation identified 19,512 additional compounds on the calculated thermodynamic convex hull and about 150,000 more within 50 meV/atom of it; the paper also analyzed predicted superconducting and superhard materials and compounds with large gap deformation potentials.",
    "field": "materials",
    "ai_technology": "Crystal-graph attention neural networks trained on expanded density-functional-theory materials data",
    "organization": "Martin Luther University Halle-Wittenberg; University of Coimbra; Friedrich Schiller University Jena",
    "source_url": "https://arxiv.org/abs/2210.00579",
    "source_type": "paper",
    "publication_date": "2022-10-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2210.00579",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors trained crystal-graph attention networks on an expanded, composition- and symmetry-balanced dataset and used them to search 2,500 binary and ternary structure prototypes covering about one billion inorganic compounds.",
      "unknown": "This is an arXiv preprint and the reported stable compounds are computationally predicted, with DFT convex-hull validation; the study does not report their experimental synthesis. Stability depends on the calculated structures and thermodynamic reference data. The additional ~150,000 compounds are near-hull candidates, not counted as thermodynamically stable discoveries in the title or summary.",
      "ai_role": "The graph networks accelerated screening of roughly one billion prototype-generated candidates, after which density-functional-theory calculations validated stability for identified candidates; machine learning plus ab initio calculations also ranked materials for selected functional properties.",
      "evidence_urls": [
        "https://arxiv.org/abs/2210.00579"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-03-unsupervised-clustering-links-deep-earthquake-scaling-to-hydration-differences-among-northwest-pacific-slabs",
    "title": "Unsupervised clustering links deep-earthquake scaling to hydration differences among northwest Pacific slabs",
    "summary": "K-means clustering of Japan Meteorological Agency deep-earthquake catalogs separated the northwest Pacific slab segments; Gutenberg–Richter analysis found a magnitude 3.7–3.8 kink in Honshu and Izu seismicity, with b values falling from about 1.4–1.7 to 0.6–0.7. The authors infer that small-event excess and an approximately kilometre-scale metastable-olivine-wedge rim may reflect deep hydration and enhanced transformational faulting.",
    "field": "physics",
    "ai_technology": "Unsupervised machine-learning clustering (K-means, checked against spectral clustering and Gaussian mixture models)",
    "organization": "Michigan State University; Freie Universität Berlin; Université d’Orléans–CNRS; University of California, Los Angeles; Yale University",
    "source_url": "https://www.nature.com/articles/s43247-022-00377-x",
    "source_type": "paper",
    "publication_date": "2022-03-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43247-022-00377-x",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "K-means clustering of Japan Meteorological Agency deep-earthquake catalogs separated the northwest Pacific slab segments; Gutenberg–Richter analysis found a magnitude 3.7–3.8 kink in Honshu and Izu seismicity, with b values falling from about 1.4–1.7 to 0.6–0.7.",
      "unknown": "The observed b-value kink is a statistical result. Hydrous defects and an approximately 1 km unstable wedge rim are the authors’ geodynamic interpretation, not direct water-content measurements; the authors discuss magnitude-detection and model assumptions.",
      "ai_role": "Clustered deep-earthquake distributions into four slab segments, enabling region-specific b-value analyses that exposed the Honshu/Izu magnitude kink and its proposed relation to slab hydration.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43247-022-00377-x"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-12-machine-learning-denoising-exposes-strong-size-scaling-in-deep-earthquake-rupture-parameters",
    "title": "Machine-learning denoising exposes strong size scaling in deep-earthquake rupture parameters",
    "summary": "A multitask neural denoiser recovered teleseismic P-wave pulses from deep Mw 5.0+ earthquakes that had previously been too noisy for source analysis. The study found moment–duration scaling M₀ ∼ τ⁴, stress-drop scaling Δσ ∼ M₀⁰·²¹ and radiated-energy scaling Eᵣ ∼ M₀¹·²⁴; deep events had a median radiation efficiency near 5%, weak directivity and few subevents. After accounting for rigidity, no systematic depth trend in duration, stress drop or radiated energy remained across 100–700 km.",
    "field": "physics",
    "ai_technology": "DenoTe, a transfer-learned multitask convolutional encoder–decoder neural network",
    "organization": "University of Washington, Department of Earth and Space Sciences",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023JB027334",
    "source_type": "paper",
    "publication_date": "2023-12-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2023JB027334",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A multitask neural denoiser recovered teleseismic P-wave pulses from deep Mw 5.0+ earthquakes that had previously been too noisy for source analysis.",
      "unknown": "The study is a source-parameter reanalysis enabled by neural denoising, not prospective earthquake prediction. It reports scaling relations and interpretations from the available global event sample; stress and energy estimates remain subject to waveform, path-correction and source-model uncertainties.",
      "ai_role": "Separated low-SNR teleseismic P-wave signals from overlapping noise, enabling source-parameter estimation from 920 deep-earthquake records, including moderate events ordinarily omitted from these analyses.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023JB027334"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-05-machine-learning-classification-shows-rigidity-corrected-moment-release-is-similar-across-earthquake-depths",
    "title": "Machine-learning classification shows rigidity-corrected moment release is similar across earthquake depths",
    "summary": "Random-forest classification of source-time-function characteristics for 3,675 global Mw > 5.5 earthquakes showed that depth-dependent rigidity explains much of the differences in rupture duration, length and stress drop between shallow, intermediate and deep earthquakes. Once medium rigidity is corrected, the analyzed events exhibit similar moment-release processes, supporting a constant-strain-drop interpretation across depths and host-rock types.",
    "field": "physics",
    "ai_technology": "Random-forest machine-learning classification and feature-importance analysis",
    "organization": "University of Science and Technology of China; Mengcheng National Geophysical Observatory",
    "source_url": "https://www.nature.com/articles/s41561-023-01176-5",
    "source_type": "paper",
    "publication_date": "2023-05-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41561-023-01176-5",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Random-forest classification of source-time-function characteristics for 3,675 global Mw > 5.5 earthquakes showed that depth-dependent rigidity explains much of the differences in rupture duration, length and stress drop between shallow, intermediate and deep earthquakes.",
      "unknown": "This is a global observational analysis of 3,675 moderate-to-large earthquakes. Its constant-strain-drop conclusion is a supported scaling interpretation, not proof that every earthquake rupture mechanism is identical; depth class and source-time-function estimates have sampling limitations.",
      "ai_role": "Compared source-time-function features across depth classes and identified surrounding-medium rigidity as the principal explanatory feature, enabling the rigidity-corrected cross-depth comparison.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41561-023-01176-5"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-01-cnn-augmented-focal-mechanisms-reveal-stress-linked-faulting-changes-around-major-southern-california-earthquakes",
    "title": "CNN-augmented focal mechanisms reveal stress-linked faulting changes around major southern California earthquakes",
    "summary": "CNN phase and first-motion pickers supplied additional waveform polarities and arrivals for a 297,478-event southern California focal-mechanism catalog, about 40% larger than the prior catalog. The authors used the expanded catalog to observe systematic post-mainshock faulting-style changes spatially correlated with estimated static stress perturbations from the 2010 El Mayor–Cucapah and 2019 Ridgecrest earthquakes, including changes more than 100 km from the epicentres.",
    "field": "physics",
    "ai_technology": "Convolutional neural networks for seismic phase arrival and first-motion polarity picking",
    "organization": "University of Southern California; California Institute of Technology; University of Southern California Southern California Earthquake Center",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022JB025975",
    "source_type": "paper",
    "publication_date": "2023-01-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2022JB025975",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "CNN phase and first-motion pickers supplied additional waveform polarities and arrivals for a 297,478-event southern California focal-mechanism catalog, about 40% larger than the prior catalog.",
      "unknown": "Catalog quality varies with location and station coverage; the authors report uncertainty in some solutions and frame stress processes as associations/inferences. The catalog provides improved observations, not a claim of earthquake predictability.",
      "ai_role": "Extracted additional phase arrivals and P-wave first-motion polarities from noisy seismograms; combined with existing observations in HASH focal-mechanism solutions to increase catalog coverage and resolution for small earthquakes.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022JB025975"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-05-random-forest-selection-and-spectroscopy-identify-20-new-high-redshift-quasars",
    "title": "Random-forest selection and spectroscopy identify 20 new high-redshift quasars",
    "summary": "A supervised random-forest search of Pan-STARRS1 and WISE photometry identified follow-up targets in the difficult redshift range around z≈5.5, where quasars resemble red and brown dwarfs. Spectroscopic observations of 37 candidates confirmed 20 previously unknown high-redshift quasars (18 at z=4.6–5.5 and two around z≈5.7), including a source outside common color-cut selections.",
    "field": "physics",
    "ai_technology": "Supervised random-forest classification and regression applied to Pan-STARRS1 and WISE data",
    "organization": "Max Planck Institute for Astronomy; Steward Observatory, University of Arizona; Max Planck Institute for Extraterrestrial Physics; European Southern Observatory; other international observatories",
    "source_url": "https://arxiv.org/abs/2105.09171",
    "source_type": "paper",
    "publication_date": "2021-05-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-3881/ac0254",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A supervised random-forest search of Pan-STARRS1 and WISE photometry identified follow-up targets in the difficult redshift range around z≈5.5, where quasars resemble red and brown dwarfs.",
      "unknown": "Only the 20 spectroscopically observed confirmations are counted. The much larger candidate catalog and predicted unobserved quasars are not treated as discoveries; the source reports a survey-restriction-dependent completeness and efficiency.",
      "ai_role": "Ranked high-redshift quasar candidates from multiband survey photometry; the resulting sample guided telescope follow-up, and spectroscopy established the 20 reported new quasars.",
      "evidence_urls": [
        "https://arxiv.org/abs/2105.09171"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-12-cnn-tracking-reveals-irradiation-driven-nanovoid-size-fluctuations-and-shrinkage-in-copper",
    "title": "CNN tracking reveals irradiation-driven nanovoid size fluctuations and shrinkage in copper",
    "summary": "A convolutional neural network combined with greedy matching to detect, track and measure pre-existing nanovoids in in-situ TEM videos of irradiated copper. The analysis revealed non-monotonic, random-walk-like size fluctuations and shrinkage during irradiation; phase-field simulations were used to investigate the mechanism. Follow-up manual measurements and additional experiments confirmed the size-fluctuation observation.",
    "field": "materials",
    "ai_technology": "Convolutional neural network for void detection, with greedy matching for tracking and migration analysis",
    "organization": "Purdue University; University of Hong Kong; Harbin Institute of Technology, Shenzhen",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S0022311522006687",
    "source_type": "paper",
    "publication_date": "2022-12-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.jnucmat.2022.154189",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A convolutional neural network combined with greedy matching to detect, track and measure pre-existing nanovoids in in-situ TEM videos of irradiated copper.",
      "unknown": "The source reports the finding for pre-existing nanovoids in irradiated copper and uses phase-field simulations to explain it. The observed fluctuation was additionally checked by manual measurements and experiments, but generalization to other materials or irradiation regimes is not established by this study.",
      "ai_role": "Automated defect recognition and tracking across high-volume in-situ microscopy video, making it possible to measure nanovoid evolution at scale; phase-field simulations then analyzed the observed dynamics.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S0022311522006687"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-10-physics-guided-machine-learning-links-southern-ocean-regime-shifts-to-a-northward-shifting-antarctic-circumpolar-current",
    "title": "Physics-guided machine learning links Southern Ocean regime shifts to a northward-shifting Antarctic Circumpolar Current",
    "summary": "Applying THOR to one eddy-permitting CM4 climate simulation, researchers detected a projected dynamical-regime shift where the Antarctic Circumpolar Current crosses the Pacific–Antarctic Ridge under the SSP5-8.5 scenario. Neural-network explanations and transport analysis attributed the modeled change to stronger wind stress shifting the current northward into less topographically constrained waters, where the flow intensifies. This is a model-based scenario result, not an observed future change.",
    "field": "climate",
    "ai_technology": "THOR (Tracking global Heating with Ocean Regimes), using NEMI clustering, an ensemble of 50 multilayer perceptrons, SHAP and layer-wise relevance propagation",
    "organization": "University of California, Davis; NOAA Geophysical Fluid Dynamics Laboratory; European Centre for Medium-Range Weather Forecasts; Institut Pierre-Simon Laplace",
    "source_url": "https://arxiv.org/abs/2310.13916",
    "source_type": "paper",
    "publication_date": "2023-10-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Applying THOR to one eddy-permitting CM4 climate simulation, researchers detected a projected dynamical-regime shift where the Antarctic Circumpolar Current crosses the Pacific–Antarctic Ridge under the SSP5-8.5 scenario.",
      "unknown": "This is an arXiv preprint and computational climate-model analysis using CM4/MOM6, one model and the SSP5-8.5 scenario. The paper reports neural-network validation accuracy around 65–70%; it describes the model output and projected shift, not an observed real-world regime transition.",
      "ai_role": "The clustering and neural-network ensemble identified and tracked local ocean dynamical regimes across model scenarios; SHAP and layer-wise relevance propagation highlighted wind-stress curl and bathymetry for follow-up physical analysis.",
      "evidence_urls": [
        "https://arxiv.org/abs/2310.13916"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-02-machine-learning-identifies-a-projected-shift-from-dic-to-alkalinity-dominated-ocean-co2-flux-variability",
    "title": "Machine learning identifies a projected shift from DIC- to alkalinity-dominated ocean CO₂-flux variability",
    "summary": "Kernel ridge regression reconstructed present and future air–sea CO₂-flux variability in five Earth-system models. The analysis found dissolved inorganic carbon (DIC) and alkalinity to be the main drivers, while DIC’s relative importance generally declines in warming scenarios—especially in the equatorial Pacific, Southern Ocean and North Atlantic—as anthropogenic carbon uptake flattens vertical DIC gradients. This result is conditional on the models and scenarios analyzed.",
    "field": "climate",
    "ai_technology": "Kernel ridge regression and machine-learning clustering/classification",
    "organization": "NORCE Norwegian Research Centre; Bjerknes Centre for Climate Research",
    "source_url": "https://www.nature.com/articles/s43247-024-01257-2",
    "source_type": "paper",
    "publication_date": "2024-02-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43247-024-01257-2",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Kernel ridge regression reconstructed present and future air–sea CO₂-flux variability in five Earth-system models.",
      "unknown": "This is an analysis of outputs from five Earth-system models and four historical/future scenarios, not a measurement of future ocean conditions. The interpretation that anthropogenic-carbon-driven gradient flattening contributes to reduced DIC importance is the authors’ proposed mechanism; relative predictor importance does not alone establish causal attribution.",
      "ai_role": "Kernel ridge regression reconstructed CO₂-flux variability from Earth-system-model variables; machine-learning analysis compared predictor importance among spatial clusters and climate scenarios to identify the changing roles of DIC and alkalinity.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43247-024-01257-2"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-machine-learning-classification-dates-a-tectonic-transition-and-probable-slab-detachment-in-central-tibet",
    "title": "Machine-learning classification dates a tectonic transition and probable slab detachment in central Tibet",
    "summary": "A random-forest classifier trained on global basalt trace-element data distinguished destructive from constructive plate-boundary signatures, with the paper reporting greater than 98% accuracy on its labeled evaluation set. Applied to basalts from Tibet’s Bangong–Nujiang suture zone, it found mixed signatures before about 108–107 Ma and only destructive-boundary signatures afterward; the authors interpret the transition as likely oceanic-slab detachment, consistent with other geological evidence.",
    "field": "physics",
    "ai_technology": "Random-forest classifier trained on global basalt trace-element concentrations",
    "organization": "China Geological Survey; Chengdu University of Technology; University of Canterbury; China University of Mining and Technology; Southern Marine Science and Engineering Guangdong Laboratory; Hunan University of Science and Technology; Nanjing University; China University of Geosciences (Beijing); Chinese Academy of Sciences",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2024JH000400",
    "source_type": "paper",
    "publication_date": "2024-12-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2024JH000400",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A random-forest classifier trained on global basalt trace-element data distinguished destructive from constructive plate-boundary signatures, with the paper reporting greater than 98% accuracy on its labeled evaluation set.",
      "unknown": "The specific age and slab-detachment interpretation are geological inferences from classified basalt geochemistry, not direct observation of a detached slab. The authors describe the detachment explanation as likely and note consistency with existing evidence; model accuracy is not independent confirmation of the tectonic interpretation.",
      "ai_role": "The classifier learned trace-element signatures associated with destructive and constructive plate boundaries and applied them to Tibetan suture-zone basalts, revealing a time-ordered change in inferred tectonic setting that supported the authors’ slab-detachment interpretation.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2024JH000400"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-04-machine-learning-analysis-identifies-kinematic-evidence-for-a-protoplanet-in-the-hd-142666-disk",
    "title": "Machine-learning analysis identifies kinematic evidence for a protoplanet in the HD 142666 disk",
    "summary": "A machine-learning model identified a strong, localized non-Keplerian gas-motion feature in the protoplanetary disk around HD 142666. Hydrodynamic simulations of a system containing a 5-Jupiter-mass planet at 75 au reproduced the observed kinematic structure; the authors conclude that HD 142666 hosts an embedded planet by current field criteria. The planet is inferred from disk kinematics rather than directly imaged.",
    "field": "physics",
    "ai_technology": "EfficientNetV2 and RegNet image classifiers (six channel-count variants), trained on synthetic planet/no-planet disk observations",
    "organization": "University of Oxford (J. P. Terry’s repository-listed affiliation); University of Georgia (resource support acknowledged in the paper)",
    "source_url": "https://iopscience.iop.org/article/10.3847/1538-4357/acc737",
    "source_type": "paper",
    "publication_date": "2023-04-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-4357/acc737",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning model identified a strong, localized non-Keplerian gas-motion feature in the protoplanetary disk around HD 142666.",
      "unknown": "The paper’s planet conclusion is an inference from gas kinematics supported by simulations, not a direct image or independent dynamical mass measurement. The models were classifiers, not trained to localize planets, and the 5-Jupiter-mass, 75-au configuration is a simulation that reproduces the observed feature, not a separately measured planet parameter.",
      "ai_role": "The classifiers applied to ALMA data assigned high planet-presence scores and their activation maps highlighted the previously unreported non-Keplerian velocity kink; the authors then used standard hydrodynamic simulations to test whether an embedded planet reproduced that feature.",
      "evidence_urls": [
        "https://iopscience.iop.org/article/10.3847/1538-4357/acc737"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-ai-assisted-proof-establishes-a-dimension-independent-kls-isoperimetric-bound",
    "title": "AI-assisted proof establishes a dimension-independent KLS isoperimetric bound",
    "summary": "Zhao Song and Xinzhi Zhang prove a universal, dimension-independent bound for the Kannan–Lovász–Simonovits constant of isotropic log-concave probability measures, resolving the KLS conjecture. Their proof repeatedly refines bounds on polynomial variance and curvature; the authors report that AI tools helped explore more than 100 approaches, obtain the earlier log-star bound, and develop and verify the final proof. Independent researchers also posted proofs, and the paper’s streamlined proof is being formalized in Lean. The 1 October preprint gave a slowly dimension-dependent bound; the 4 October version established the dimension-independent theorem described here.",
    "field": "mathematics",
    "ai_technology": "GPT-6 Astra; GPT-5.6 Sol; Claude Fable 5; Fable (as named in the paper’s AI disclosure)",
    "organization": "Microsoft Research; independent research collaboration",
    "source_url": "https://arxiv.org/pdf/2610.01447v2",
    "source_type": "paper",
    "publication_date": "2026-10-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2610.01447",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Zhao Song and Xinzhi Zhang prove a universal, dimension-independent bound for the Kannan–Lovász–Simonovits constant of isotropic log-concave probability measures, resolving the KLS conjecture.",
      "unknown": "The proof is a recent arXiv preprint, not yet journal peer-reviewed in the source reviewed. Independent proofs were posted, and the Microsoft Research author account reports later Lean formalization of a streamlined proof; the paper version and posted revisions should be checked as the work develops. This entry counts the theorem result once, not the intermediate log-star bound separately.",
      "ai_role": "The authors state that they explored candidate approaches collaboratively with AI tools; the tools helped obtain the earlier log-star bound and develop detailed proof steps, while the researchers selected the direction and completed the repeated refinement argument. The authors state they checked the proofs with AI tools and take responsibility for correctness.",
      "evidence_urls": [
        "https://arxiv.org/pdf/2610.01447v2",
        "https://arxiv.org/abs/2610.01447",
        "https://www.microsoft.com/en-us/research/blog/solving-a-31-year-old-grand-challenge-in-mathematics-the-kls-conjecture/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-09-ai-discovered-proof-resolves-the-komlos-vector-balancing-conjecture",
    "title": "AI-discovered proof resolves the Komlós vector-balancing conjecture",
    "summary": "The paper proves that for every finite family of Euclidean-norm-at-most-one vectors, signs can be chosen so the signed sum has infinity norm below 3√(2π), independent of vector count and dimension. It derives a corresponding square-root-in-sets bound for Beck–Fiala discrepancy. The authors explicitly credit the Odin Automatic AI Research Agent with discovering the proof; a separate elementary proof was posted independently. The algorithmic construction of such signings is a separate follow-up result and is not claimed here.",
    "field": "mathematics",
    "ai_technology": "Odin Automatic AI Research Agent",
    "organization": "Duke University; University of California, Irvine; independent research collaboration",
    "source_url": "https://arxiv.org/abs/2609.11189",
    "source_type": "paper",
    "publication_date": "2026-09-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The paper proves that for every finite family of Euclidean-norm-at-most-one vectors, signs can be chosen so the signed sum has infinity norm below 3√(2π), independent of vector count and dimension.",
      "unknown": "The result is reported in an arXiv preprint, not a peer-reviewed journal source in this review. A separate elementary proof of the conjecture appeared later, providing an independent proof route; the paper’s AI attribution is author-reported. The result establishes existence, not a polynomial-time signing algorithm.",
      "ai_role": "The authors credit Odin with discovering the proof of their vector-balancing theorem; Duke’s official account says the system discovered the proof and human authors scrutinized and organized the argument into the paper.",
      "evidence_urls": [
        "https://arxiv.org/abs/2609.11189"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-09-ai-discovered-construction-improves-the-lower-bound-for-the-complex-grothendieck-constant",
    "title": "AI-discovered construction improves the lower bound for the complex Grothendieck constant",
    "summary": "The authors prove the classical complex Grothendieck constant exceeds 1.35584631827168, closing more than one quarter of the gap between Davie’s lower bound and Haagerup’s upper bound reported in the paper. The numerical search used Hermite projections and a radial weight; interval arithmetic on exact rational inputs rigorously verifies the stated bound. The authors credit the Odin Automatic AI Research Agent with discovering both the lower bound and its proof.",
    "field": "mathematics",
    "ai_technology": "Odin Automatic AI Research Agent",
    "organization": "Independent research collaboration",
    "source_url": "https://arxiv.org/abs/2609.07000",
    "source_type": "paper",
    "publication_date": "2026-09-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors prove the classical complex Grothendieck constant exceeds 1.35584631827168, closing more than one quarter of the gap between Davie’s lower bound and Haagerup’s upper bound reported in the paper.",
      "unknown": "The source is an arXiv preprint. Its stated lower bound is computer-assisted and the proof includes interval arithmetic with exact inputs; the paper does not determine the exact complex Grothendieck constant. The authors’ AI attribution is not independent evidence about the system’s internal process.",
      "ai_role": "The authors explicitly state that Odin discovered the improved lower bound and its proof; the preprint describes numerical optimization that found a candidate, followed by exact rational inputs and interval-arithmetic verification by the authors’ supplied code.",
      "evidence_urls": [
        "https://arxiv.org/abs/2609.07000"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-10-machine-learning-predicts-timing-and-fault-state-evolution-in-meter-scale-laboratory-earthquakes",
    "title": "Machine learning predicts timing and fault-state evolution in meter-scale laboratory earthquakes",
    "summary": "A Random Forest using a network representation of acoustic-emission event catalogs predicted the timing of meter-scale laboratory mainshocks from tens of seconds to milliseconds ahead (test-set log R²=0.84). Feature analysis found systematic changes in event intervals and seismic moments as failure approached; comparison with dynamic simulations led the authors to infer that acoustic emissions indirectly encode stress evolution on creeping fault patches. This is a laboratory-scale result, not a validated method for forecasting natural earthquakes.",
    "field": "physics",
    "ai_technology": "Random Forest machine-learning model; deep-learning comparison model",
    "organization": "Kyoto University; Disaster Prevention Research Institute, Kyoto University",
    "source_url": "https://www.nature.com/articles/s41467-025-64542-4",
    "source_type": "paper",
    "publication_date": "2025-10-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-64542-4",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Random Forest using a network representation of acoustic-emission event catalogs predicted the timing of meter-scale laboratory mainshocks from tens of seconds to milliseconds ahead (test-set log R²=0.84).",
      "unknown": "This is one meter-scale rock-friction experiment (experiment LB12-011) and a corresponding numerical model; forecast skill varies with foreshock density and cannot be generalized to natural earthquakes. The article was corrected on 2026-02-02 to replace an uncorrected proof PDF; the correction says it did not describe a scientific-result change.",
      "ai_role": "The Random Forest model made the time-to-failure predictions from the event catalog; the authors analyzed predictive features and compared the results to dynamic simulations to infer that catalog signals encode evolving fault state. The machine-learning analysis is the core result, not merely automated event detection.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-64542-4"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-ai-assisted-seismic-catalog-links-the-2018-baige-landslide-dammed-lakes-to-increased-local-earthquakes",
    "title": "AI-assisted seismic catalog links the 2018 Baige landslide-dammed lakes to increased local earthquakes",
    "summary": "An earthquake catalog built with the PhaseNet machine-learning phase detector and cross-correlation/template matching revealed a statistically significant (over 99%) increase in small earthquakes as the second Baige landslide-dammed lake approached peak water level; rates returned toward background after breach. Coulomb-stress modeling found about 90% of the earthquakes in regions of positive modeled stress change, supporting likely triggering through direct water loading and pore-pressure diffusion. The paper presents this as evidence that earthquake–surface-hazard triggering can be bidirectional.",
    "field": "physics",
    "ai_technology": "PhaseNet machine-learning phase detector, combined with PyOcto association and cross-correlation/template-matching methods",
    "organization": "The Chinese University of Hong Kong; WSL Swiss Federal Institute for Forest, Snow and Landscape Research; Institute of Mountain Hazards and Environment, Chinese Academy of Sciences; Sichuan Earthquake Administration; Géoazur",
    "source_url": "https://www.nature.com/articles/s41467-024-47130-w",
    "source_type": "paper",
    "publication_date": "2024-04-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-47130-w",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An earthquake catalog built with the PhaseNet machine-learning phase detector and cross-correlation/template matching revealed a statistically significant (over 99%) increase in small earthquakes as the second Baige landslide-dammed lake approached peak water level; rates returned toward background after breach.",
      "unknown": "This is a case study of one pair of 2018 landslide-dammed lakes and nearby faults. Event-location uncertainties are several kilometers; fault geometry and fluid-diffusion parameters require assumptions. The authors conclude the lakes likely triggered the sequence, rather than directly observing causality, and the global implication is modeled potential.",
      "ai_role": "PhaseNet detected P- and S-wave arrivals in continuous seismic records; these detections supported a new 2014–2023 earthquake catalog used to identify and statistically test the local seismicity increase. Coulomb-stress modeling then assessed a possible physical mechanism.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-47130-w"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-08-machine-learning-detects-volcanic-fingerprints-and-suggests-a-northern-extratropical-signal-for-the-unidentified-1809-eruption",
    "title": "Machine learning detects volcanic fingerprints and suggests a northern extratropical signal for the unidentified 1809 eruption",
    "summary": "A convolutional neural network trained on idealized Earth-system-model simulations classified eruption occurrence and broad latitude band from global surface-temperature anomalies in the first post-eruption summer. Applied to reanalysis and observational climate records, the method identified major historic events and assigned a northern-extratropical footprint to the unidentified 1809 eruption. The same analysis found that some smaller events leave weaker, less distinguishable temperature signals.",
    "field": "climate",
    "ai_technology": "Convolutional neural network with explainable-AI relevance mapping",
    "organization": "German Climate Computing Center (DKRZ); Max Planck Institute for Meteorology; Center for Climate Physics, Institute for Basic Science; Pusan National University",
    "source_url": "https://www.nature.com/articles/s43247-024-01617-y",
    "source_type": "paper",
    "publication_date": "2024-08-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43247-024-01617-y",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A convolutional neural network trained on idealized Earth-system-model simulations classified eruption occurrence and broad latitude band from global surface-temperature anomalies in the first post-eruption summer.",
      "unknown": "The model is trained on idealized simulations with only three broad eruption latitude classes and four sulfur-injection strengths. The 1809 geographic inference is an AI classification of a climate fingerprint, not direct observation of the eruption source; predictions show sensitivity to normalization and differing observational datasets. It should be interpreted as a clue, not a definitive eruption-location determination.",
      "ai_role": "The CNN classified volcanic presence/location using post-eruption temperature patterns; explainability analysis localized climate regions driving those classifications and helped infer the 1809 event’s northern-extratropical footprint from observational/reanalysis data.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43247-024-01617-y"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-10-self-supervised-learning-identifies-two-previously-unreported-seismic-sequences-during-the-mayotte-submarine-eruption",
    "title": "Self-supervised learning identifies two previously unreported seismic sequences during the Mayotte submarine eruption",
    "summary": "A SimSiam self-supervised representation model with t-SNE and DBSCAN clustering explored two months of continuous seismological records from near the Fani Maoré submarine volcano. Researchers identified two previously undocumented multi-day sequences in October–November 2019, each beginning with increased volcano-tectonic events and followed by tremor-like, hydroacoustic or mixed signals. The paper describes the sequences as potentially associated with ongoing eruptive activity.",
    "field": "physics",
    "ai_technology": "SimSiam self-supervised learning with t-SNE dimensionality reduction and DBSCAN clustering",
    "organization": "University of Strasbourg; Institut de Physique du Globe de Paris; NORSAR; University of Haute-Alsace; IFREMER",
    "source_url": "https://academic.oup.com/gji/article/240/1/1/7817994",
    "source_type": "paper",
    "publication_date": "2024-10-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/gji/ggae361",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A SimSiam self-supervised representation model with t-SNE and DBSCAN clustering explored two months of continuous seismological records from near the Fani Maoré submarine volcano.",
      "unknown": "The event data are from two months in 2019 and a nearby ocean-bottom seismometer; the study does not establish exact event locations or independently prove that each sequence was eruptive. Authors state that instrumental noise cannot be ruled out for some high-energy signals, and describe the volcanic association as potential. It is a peer-reviewed 2025 issue paper published online 2024-10-10.",
      "ai_role": "The self-supervised model constructed signal representations from full continuous seismograms; clustering surfaced two coherent, previously undocumented temporal event sequences that the researchers then interpreted using expert signal review. This avoids treating the result as a detector-only tool release.",
      "evidence_urls": [
        "https://academic.oup.com/gji/article/240/1/1/7817994"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-09-gpu-phase-folding-and-deep-learning-identify-five-statistically-validated-ultra-short-period-sub-earth-planets-in-kepler-data",
    "title": "GPU phase folding and deep learning identify five statistically validated ultra-short-period sub-Earth planets in Kepler data",
    "summary": "A GPU-based transit search paired with a convolutional neural network identified five sub-Earth-size ultra-short-period transit signals in Kepler light curves: Kepler-158d, Kepler-963c, Kepler-879c, Kepler-1489c, and KOI-4978.02. The authors applied transit vetting, false-alarm tests, and statistical false-positive probabilities below 1% to each, and statistically validated the five signals as planets. The validation is photometric/statistical rather than independent radial-velocity or spectroscopic confirmation; KOI-4978.02 remains designated as a KOI object in the source.",
    "field": "physics",
    "ai_technology": "GPFC algorithm: GPU phase folding with convolutional neural network transit scoring",
    "organization": "Shanghai Astronomical Observatory; Stanford University; Princeton University; University of Geneva",
    "source_url": "https://academic.oup.com/mnras/article/534/3/1913/7762975",
    "source_type": "paper",
    "publication_date": "2024-09-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stae2155",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A GPU-based transit search paired with a convolutional neural network identified five sub-Earth-size ultra-short-period transit signals in Kepler light curves: Kepler-158d, Kepler-963c, Kepler-879c, Kepler-1489c, and KOI-4978.02.",
      "unknown": "The paper calls the five objects statistically validated, not all independently confirmed by spectroscopy or radial velocities. Validation relies on Kepler photometry, transit vetting, estimated false-positive probabilities and false-alarm tests. KOI-4978.02 retains its KOI designation. One paper-level result; individual planets are not split into separate entries.",
      "ai_role": "GPU phase folding searched Kepler light curves and a CNN scored/identified transit-like signals; follow-up statistical and transit-shape tests vetted the model-discovered signals. Authors report FPP below 1% for all five.",
      "evidence_urls": [
        "https://academic.oup.com/mnras/article/534/3/1913/7762975"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-06-generative-machine-learning-guides-experimental-synthesis-of-lizn2pt-and-nipt2ga-heusler-phases",
    "title": "Generative machine learning guides experimental synthesis of LiZn₂Pt and NiPt₂Ga Heusler phases",
    "summary": "Researchers combined a generative probabilistic crystal-graph model (PGCGM) with an ALIGNN stability predictor to propose and screen ternary structures, then used the model outputs to guide synthesis. They experimentally synthesized phase-pure LiZn₂Pt and NiPt₂Ga; the measured NiPt₂Ga adopts a tetragonal Heusler structure rather than the model’s cubic proposal, with DFT comparisons supporting its lower energy. They also synthesized LiZn₂Pd as a human-selected follow-up that was not among PGCGM’s candidate outputs. This is one paper-level AI-enabled synthesis workflow and three related compounds, not three separate records.",
    "field": "materials",
    "ai_technology": "PGCGM generative Wasserstein GAN for ternary crystal structures; ALIGNN graph neural network for decomposition-enthalpy stability screening",
    "organization": "Johns Hopkins University; Johns Hopkins University Applied Physics Laboratory; Gdansk University of Technology",
    "source_url": "https://pubs.rsc.org/en/content/articlehtml/2025/ra/d5ra00427f",
    "source_type": "paper",
    "publication_date": "2025-06-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/d5ra00427f",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers combined a generative probabilistic crystal-graph model (PGCGM) with an ALIGNN stability predictor to propose and screen ternary structures, then used the model outputs to guide synthesis.",
      "unknown": "The reported synthesis outcome is experimentally characterized in the paper, but the AI was a candidate-generation and stability-screening aid rather than an autonomous discoverer: expert chemistry selected candidates and performed synthesis. LiZn₂Pt’s composition and Fm3̄m structure were already represented in Materials Project/OQMD training data, so its novelty is experimental realization rather than structure invention. The NiPt₂Ga experimental structure is tetragonal P4/mmm, differing from the cubic PGCGM proposal; the paper gives a DFT energy comparison. No superconductivity was observed, and magnetic ordering temperature for NiPt₂Ga was unresolved.",
      "ai_role": "PGCGM generated candidate crystal structures and ALIGNN screened their predicted stability. Researchers used these model outputs to down-select and choose Heusler-family candidates for synthesis. The authors note that expert judgment remained necessary and that LiZn₂Pt’s composition/structure appeared in the model training databases; NiPt₂Ga’s experimentally realized tetragonal structure was not the exact cubic structure proposed by PGCGM.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlehtml/2025/ra/d5ra00427f"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-ai-and-dft-guided-synthesis-identifies-a-previously-unreported-calcium-ruthenate-phase",
    "title": "AI- and DFT-guided synthesis identifies a previously unreported calcium-ruthenate phase",
    "summary": "An autonomous materials-discovery workflow used AI-feedback density-functional-theory simulations to select six previously well-explored ternary oxide systems for synthesis. In the Ca–Ru–O campaign, experiments yielded two previously unreported phases; room-temperature diffraction and follow-up characterization most clearly support an isolated low-temperature calcium-rich phase, approximately Ca1+xRuO3+y, distinct from conventional CaRuO3. The authors state the realized phases were not the specific crystal structures predicted by the simulation.",
    "field": "materials",
    "ai_technology": "CAMD (computer-aided materials discovery), an AI-feedback density-functional-theory phase-search workflow",
    "organization": "Toyota Research Institute; Lawrence Berkeley National Laboratory; University of California, Berkeley; North Carolina State University; Lucideon",
    "source_url": "https://pubs.rsc.org/en/content/articlehtml/2024/sc/d3sc04823c",
    "source_type": "paper",
    "publication_date": "2024-03-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/d3sc04823c",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An autonomous materials-discovery workflow used AI-feedback density-functional-theory simulations to select six previously well-explored ternary oxide systems for synthesis.",
      "unknown": "One low-temperature phase was isolated and extensively characterized; the second high-temperature phase remained multiphase/metastable and structurally unresolved, so this record rests on the better-supported low-temperature result. Exact composition and location of excess Ca remain under investigation. AI/DFT guided chemical-system selection but did not correctly predict the realized phase structure; authors report only one successful system among six and do not claim AI accelerated the overall process.",
      "ai_role": "CAMD mapped candidate phases across chemical systems and informed expert down-selection. Human researchers designed and ran synthesis and characterization. The successful calcium-ruthenate phases emerged experimentally in the AI-prioritized system but did not match the exact predicted crystal structures; authors describe one success among six systems and stress current AI-guided synthesis limitations.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlehtml/2024/sc/d3sc04823c"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-12-xlumina-computationally-discovers-a-super-resolution-microscopy-blueprint-combining-two-optical-principles",
    "title": "XLuminA computationally discovers a super-resolution microscopy blueprint combining two optical principles",
    "summary": "The XLuminA differentiable optical-design system rediscovered established STED beam shaping and optical-vortex approaches, then identified a previously unreported experimental blueprint combining principles from both to produce sub-diffraction imaging in the study’s computational model. This is a specific theoretical/optical-design result, not a report of an independently built and experimentally validated microscope implementing the new blueprint.",
    "field": "physics",
    "ai_technology": "XLuminA auto-differentiating optical experiment-design framework; transformer-based code-generation model used in the discovery workflow",
    "organization": "Max Planck Institute for the Science of Light; Friedrich-Alexander-Universität Erlangen-Nürnberg",
    "source_url": "https://www.nature.com/articles/s41467-024-54696-y",
    "source_type": "paper",
    "publication_date": "2024-12-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-54696-y",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The XLuminA differentiable optical-design system rediscovered established STED beam shaping and optical-vortex approaches, then identified a previously unreported experimental blueprint combining principles from both to produce sub-diffraction imaging in the study’s computational model.",
      "unknown": "The novel design is established through computational optical modeling and simulation; the paper does not report physical construction or experimental validation of this specific blueprint. The three rediscovered designs are prior work and are not counted as new. Interpret as a computationally proposed optical design, not a demonstrated imaging instrument or biological discovery.",
      "ai_role": "AI-guided computational search generated and evaluated candidate optical experimental designs; numerical simulations identified a previously unreported combination of optical principles with predicted sub-diffraction capability. Researchers analyzed and interpreted the design.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-54696-y"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-05-autonomous-btsbot-nearby-follow-up-captures-an-infant-supernovas-flash-spectrum-and-constrains-progenitor-mass-loss",
    "title": "Autonomous BTSbot-nearby follow-up captures an infant supernova’s flash spectrum and constrains progenitor mass loss",
    "summary": "The BTSbot-nearby system used the BTSbot machine-learning classifier to identify SN 2024jlf in ZTF data and autonomously trigger a spectroscopic follow-up request, enabling spectroscopy about 0.7 days after first light. The early spectra showed short-lived Hα, He II and C IV flash-ionization features; comparison with two radiative-hydrodynamics model grids constrained the red-supergiant progenitor’s enhanced pre-explosion mass-loss rate to roughly 10⁻⁴–10⁻³ solar masses per year.",
    "field": "physics",
    "ai_technology": "BTSbot machine-learning transient classifier integrated with the BTSbot-nearby autonomous follow-up program",
    "organization": "Northwestern University; California Institute of Technology; Stockholm University; Liverpool John Moores University; NSF-Simons AI Institute for the Sky; Zwicky Transient Facility collaboration",
    "source_url": "https://iopscience.iop.org/article/10.3847/1538-4357/adcf1e",
    "source_type": "paper",
    "publication_date": "2025-05-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-4357/adcf1e",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The BTSbot-nearby system used the BTSbot machine-learning classifier to identify SN 2024jlf in ZTF data and autonomously trigger a spectroscopic follow-up request, enabling spectroscopy about 0.7 days after first light.",
      "unknown": "The AI contribution is event identification and automated prioritization, not inference of the stellar mass-loss physics. The rate is model-dependent and derived from two radiative-hydrodynamics code grids under adopted wind assumptions. The paper reports an individual nearby Type IIP event; this is not a population-level claim. The transient was originally reported to the Transient Name Server by a human observer, while the BTSbot-nearby program supplied rapid automated follow-up.",
      "ai_role": "BTSbot identified the transient in ZTF alert data; host-distance and youth filters in BTSbot-nearby automatically triggered target-of-opportunity spectroscopy. This shortened latency to approximately seven minutes from alert-filter pass to first spectrum (spectroscopy was obtained about 0.7 days after first light), enabling observation of ephemeral flash features. Human and instrument teams obtained/interpreted observations; radiation-hydrodynamics models supported the mass-loss inference.",
      "evidence_urls": [
        "https://iopscience.iop.org/article/10.3847/1538-4357/adcf1e"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-physics-informed-retrieval-augmented-language-model-guides-synthesis-of-five-previously-unreported-inorganic-crystals",
    "title": "Physics-informed retrieval-augmented language model guides synthesis of five previously unreported inorganic crystals",
    "summary": "Researchers built PIRAG-LM, which combines a structured knowledge base of experimental synthesis routes, physics-based chemical/structural/thermodynamic similarity retrieval, and LLM route reasoning. In prospective tests, it guided synthesis of five compounds previously unreported in experimental literature: BaMo0.3In0.7O2.95, BaNb0.4In0.6O2.9, Hg[B(CN)4]2, CoCo(CN)6 and SrNb2Fe2(PO4)6. Powder X-ray diffraction patterns were consistent with target phases. This is one paper-level synthesis-planning result, not five separate entries.",
    "field": "materials",
    "ai_technology": "PIRAG-LM, physics-informed retrieval-augmented language model, with DeepSeek-V4 used in the structured synthesis-knowledge-base extraction workflow",
    "organization": "Soochow University; Zhengzhou University; Fudan University; Hefei National Laboratory",
    "source_url": "https://arxiv.org/abs/2608.25392",
    "source_type": "paper",
    "publication_date": "2026-08-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Researchers built PIRAG-LM, which combines a structured knowledge base of experimental synthesis routes, physics-based chemical/structural/thermodynamic similarity retrieval, and LLM route reasoning.",
      "unknown": "This is a source-checked preprint (not treated as peer-reviewed here). The reported new-compound claims refer to absence from experimental literature and prospective synthesis; PXRD matches support target phase formation, but the paper does not establish broad application performance or independent replication. Experts selected materials and adjusted operational conditions; the system did not autonomously run experiments. Five compounds remain bundled as one paper-level finding.",
      "ai_role": "PIRAG-LM retrieved synthesis precedents by chemical, structural and thermodynamic similarity and proposed synthesis methods, precursors and processing conditions. Experts evaluated routes and refined practical parameters; researchers synthesized and characterized the five target phases. The final route class, precursor logic and main processing sequence generally followed model recommendations.",
      "evidence_urls": [
        "https://arxiv.org/abs/2608.25392"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-llm-guided-iterative-optimization-yields-a-previously-unreported-rb3bii6-perovskite-inspired-thin-film",
    "title": "LLM-guided iterative optimization yields a previously unreported Rb3BiI6 perovskite-inspired thin film",
    "summary": "A closed-loop materials workflow combined an LLM that distilled synthesis knowledge from papers, high-throughput hyperspectral imaging, and multi-objective Bayesian optimization with experimental feedback. In a demonstration, the team synthesized previously unreported Rb3BiI6 thin films and validated optimized films with optical bandgap analysis and X-ray diffraction. LLM-assisted initialization produced more Pareto-optimal samples and higher hypervolume than matched Latin-hypercube initialization in the reported paired campaign.",
    "field": "materials",
    "ai_technology": "LLM-assisted synthesis-knowledge extraction and initialization with multi-objective Bayesian optimization and high-throughput hyperspectral imaging",
    "organization": "Massachusetts Institute of Technology; Toyota Research Institute; University of Tokyo",
    "source_url": "https://arxiv.org/abs/2608.15928",
    "source_type": "paper",
    "publication_date": "2026-08-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A closed-loop materials workflow combined an LLM that distilled synthesis knowledge from papers, high-throughput hyperspectral imaging, and multi-objective Bayesian optimization with experimental feedback.",
      "unknown": "This is a preprint. The specific Rb3BiI6 result is reported as a previously unreported compound experimentally synthesized as thin films and checked by XRD and optical bandgap analysis; the study does not establish independent replication, long-term stability or device performance. Do not treat the accompanying optimization benchmark as an independent discovery.",
      "ai_role": "The LLM distilled literature synthesis expertise and initialized candidate experimental choices; Bayesian optimization iteratively selected subsequent conditions based on measured hyperspectral outcomes. Human operators synthesized thin films and performed optical/XRD validation.",
      "evidence_urls": [
        "https://arxiv.org/abs/2608.15928"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-ai-guided-proof-search-solves-five-open-problems-in-banach-space-theory",
    "title": "AI-guided proof search solves five open problems in Banach space theory",
    "summary": "A research project used ChatGPT 5.5 Pro and Codex agents to develop proof ideas and proofs for five Banach-space problems. The resulting paper gives affirmative solutions to questions on toroidal Elton–Odell separation, non-Calkin unital Banach algebras, strict cosingularity and adjoints, weakly compact basis factorization, and primariness of Lp(L1). The authors report that models generated the core mathematics; human mathematicians checked and refined the proofs. The results are presented in an unrefereed preprint, and the authors take responsibility for correctness.",
    "field": "mathematics",
    "ai_technology": "ChatGPT 5.5 Pro and Codex agents using an automated open-problem search pipeline",
    "organization": "Lancaster University; University of Bern",
    "source_url": "https://arxiv.org/abs/2607.17388",
    "source_type": "paper",
    "publication_date": "2026-07-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A research project used ChatGPT 5.5 Pro and Codex agents to develop proof ideas and proofs for five Banach-space problems.",
      "unknown": "Unrefereed arXiv preprint. Authors state that humans verified and refined the proofs and are responsible for correctness; the proofs are not formally machine-checked. The five problems were selected by mathematicians, while a separate pipeline component automatically found other candidate problems. The five results remain bundled as one paper-level report.",
      "ai_role": "In the pipeline component, agents searched arXiv literature for open problems and attempted solutions. For the five selected Banach-space problems, the authors say model attempts generated the proof ideas and essentially complete arguments before human verification, editing and integration; models were not limited to exposition.",
      "evidence_urls": [
        "https://arxiv.org/abs/2607.17388"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-03-deep-learning-seismic-catalog-reveals-a-potentially-reactivated-blind-fault-in-taiwans-2025-dapu-earthquake-sequence",
    "title": "Deep-learning seismic catalog reveals a potentially reactivated blind fault in Taiwan’s 2025 Dapu earthquake sequence",
    "summary": "A RED-PAN deep-learning phase picker enabled a catalog of 6,805 relocated events for the 11-day sequence following the M6.4 Dapu earthquake. Analysis of those events, focal mechanisms and velocity structure identified an east-dipping plane at about 5–10 km and a west-dipping plane at about 10–15 km; the authors infer the mainshock involved a previously unrecognized blind west-dipping fault and interactions with the known Muchiliao–Liuchia system. This is a localized structural interpretation, and the authors describe some fault associations as potential.",
    "field": "physics",
    "ai_technology": "RED-PAN deep-learning seismic phase picker",
    "organization": "Academia Sinica; National Central University; National Taiwan University; Central Weather Administration of Taiwan",
    "source_url": "https://link.springer.com/article/10.1007/s44195-025-00093-x",
    "source_type": "paper",
    "publication_date": "2025-03-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1007/s44195-025-00093-x",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A RED-PAN deep-learning phase picker enabled a catalog of 6,805 relocated events for the 11-day sequence following the M6.4 Dapu earthquake.",
      "unknown": "Peer-reviewed open-access short report. The principal west-dipping plane is interpreted as a potentially reactivated, previously unmapped blind fault; the paper calls some extended activity potential and recommends further monitoring. It is one earthquake sequence and the AI created the event catalog rather than independently deriving the tectonic model. A companion WTMA paper (DOI 10.1007/s44195-025-00095-9) analyzes the same Dapu sequence and similar two-plane geometry; it is a related, partly overlapping report and is not counted separately.",
      "ai_role": "RED-PAN detected P- and S-wave arrivals from continuous seismic recordings and supported event association and catalog construction; this catalog provided the relocated aftershock distribution used with focal mechanisms and velocity profiles to infer fault geometry. The fault interpretation itself involved human analysis, not an autonomous AI conclusion.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1007/s44195-025-00093-x"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-07-euclid-and-machine-learning-selection-identify-31-spectroscopically-confirmed-quasars-at-redshift-6-67-8",
    "title": "Euclid and machine-learning selection identify 31 spectroscopically confirmed quasars at redshift 6.6–7.8",
    "summary": "Using Euclid Wide Survey photometry over about 3,000 square degrees, the team applied several photometric selection methods, including XGBoost classifiers and probabilistic modeling, then obtained telescope spectra for candidates. Spectroscopy confirmed 31 new quasars at redshifts 6.6–7.8; 12 are at z≥7, and the paper reports EUCL J172902.75+641018.1 at z≈7.77 as the highest-redshift quasar known at publication. The sample extends quasar studies to fainter luminosities at z>7.",
    "field": "physics",
    "ai_technology": "XGBoost machine-learning classifiers and Extreme-Deconvolution high-redshift quasar selection, alongside probabilistic spectral-energy-distribution methods",
    "organization": "Euclid Consortium; Leiden University; University of Michigan; University of Arizona; Max Planck Institute for Astronomy; University of California, Santa Barbara; international telescope partners",
    "source_url": "https://doi.org/10.1051/0004-6361/202658883",
    "source_type": "paper",
    "publication_date": "2026-07-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using Euclid Wide Survey photometry over about 3,000 square degrees, the team applied several photometric selection methods, including XGBoost classifiers and probabilistic modeling, then obtained telescope spectra for candidates.",
      "unknown": "Peer-reviewed Euclid Consortium study. Candidate ranking was AI-assisted, but human review and spectroscopic follow-up were essential to confirmation. The redshift record is the authors’ claim at the article’s publication date, not a claim about the present-day record. This is one paper-level sample of 31 confirmed quasars; candidate lists and the 31 individual objects are not separate entries.",
      "ai_role": "Multiple photometric selection methods ranked high-redshift quasar candidates from Euclid colors; XGBoost estimated quasar probabilities and redshifts. Humans visually removed artifacts and prioritized follow-up, and Keck, Magellan and LBT spectra established the 31 quasar identifications.",
      "evidence_urls": [
        "https://doi.org/10.1051/0004-6361/202658883"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-earthlink-identifies-an-eight-month-predictor-set-for-atlantic-nino-hindcasts",
    "title": "EarthLink identifies an eight-month predictor set for Atlantic Niño hindcasts",
    "summary": "On an open-ended climate-analysis task, the EarthLink agent autonomously selected two Atlantic-internal predictors for the summer ATL3 index at an eight-month lead: the SON-to-November tendency of the equatorial east–west sea-level-pressure gradient and November western-equatorial-Atlantic sea level as a thermocline-tilt proxy. Rolling-origin hindcasts for 1995–2023 achieved a temporal correlation of about 0.46, above persistence (~0.13) and ENSO-only baselines. The paper proposes a Kelvin-wave/Bjerknes-feedback explanation for the association.",
    "field": "climate",
    "ai_technology": "EarthLink multi-agent AI research system based on foundation language models and climate data/tools",
    "organization": "Shanghai Artificial Intelligence Laboratory; Fudan University; Institute of Atmospheric Physics, Chinese Academy of Sciences; Nanjing University of Information Science and Technology; Ocean University of China; Columbia University",
    "source_url": "https://arxiv.org/abs/2507.17311",
    "source_type": "paper",
    "publication_date": "2025-07-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "On an open-ended climate-analysis task, the EarthLink agent autonomously selected two Atlantic-internal predictors for the summer ATL3 index at an eight-month lead: the SON-to-November tendency of the equatorial east–west sea-level-pressure gradient and November western-equatorial-Atlantic sea level as a thermocline-tilt proxy.",
      "unknown": "This is an unrefereed arXiv preprint and a retrospective rolling-origin hindcast from one observational/reanalysis data setup. The physical pathway is an AI-generated, testable hypothesis that the authors explicitly say requires further validation; the source does not establish a causal mechanism or demonstrated prospective operational forecast. Skill is ~0.46 and falls short of the authors’ 0.5 target.",
      "ai_role": "Given an open-ended Atlantic Niño precursor problem, EarthLink autonomously planned analyses, generated code, tested predictors using regression models, reviewed literature and synthesized a physical mechanism. The paper states the proposed AI-generated hypothesis needs further validation.",
      "evidence_urls": [
        "https://arxiv.org/abs/2507.17311"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-05-unsupervised-seismicity-categories-distinguish-documented-preparatory-phases-before-three-large-earthquakes",
    "title": "Unsupervised seismicity categories distinguish documented preparatory phases before three large earthquakes",
    "summary": "A K-means framework grouped earthquake families by catalog-derived spatial, temporal, magnitude and seismo-mechanical features. Retrospective analyses found highly localized families with higher interaction and strain release before the 2023 Kahramanmaraş, 2009 L’Aquila and 2014 Iquique earthquakes; similar critical categories were absent before the 2016 Amatrice and 2024 Noto events, where no catalog-visible preparatory phase was detected. The work describes a conditional diagnostic, not a general earthquake prediction system.",
    "field": "physics",
    "ai_technology": "Unsupervised K-means categorization of catalog-derived seismic-family features",
    "organization": "GFZ Helmholtz Centre for Geosciences; University of Potsdam; University of Naples Federico II; University of Genoa; Stanford University; international collaborators",
    "source_url": "https://www.nature.com/articles/s41467-026-72279-x",
    "source_type": "paper",
    "publication_date": "2026-05-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-72279-x",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A K-means framework grouped earthquake families by catalog-derived spatial, temporal, magnitude and seismo-mechanical features.",
      "unknown": "Peer-reviewed retrospective analysis of five large-earthquake sequences. Preparatory phases in Kahramanmaraş, L’Aquila and Iquique had been documented previously; this paper contributes an unsupervised family-level categorization and comparative feature characterization. The authors state that the approach works only when detectable seismic transients occur, and its use for operational forecasting remains prospective; it is not evidence of reliable prospective prediction.",
      "ai_role": "Unsupervised machine learning grouped spatiotemporal event families from multiple seismicity features into categories without predefining the categories; the authors identified a category with highly localized, interacting seismicity and increased strain release in cases with documented preparatory phases.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-72279-x"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-02-language-model-discovers-general-construction-rules-for-two-previously-unknown-families-of-quantum-experiments",
    "title": "Language model discovers general construction rules for two previously unknown families of quantum experiments",
    "summary": "A sequence-to-sequence transformer generated readable Python meta-programs that construct photonic experimental blueprints for entire families of quantum states, rather than only individual target states. In the final paper, two successful generalizations were previously unknown: a photonic construction for general spin-1/2 states with no adjacent spin-ups, and one for states of the Majumdar–Ghosh model. The generated programs matched states beyond the three examples used as input, providing computational evidence of the inferred rules.",
    "field": "physics",
    "ai_technology": "Transformer-based sequence-to-sequence language model trained on synthetic quantum-state and experimental-blueprint pairs",
    "organization": "University of Tübingen; Max Planck Institute for the Science of Light; University of Toronto; University of California, Berkeley; Stanford University; xAI",
    "source_url": "https://www.nature.com/articles/s42256-025-01153-0",
    "source_type": "paper",
    "publication_date": "2026-02-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42256-025-01153-0",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A sequence-to-sequence transformer generated readable Python meta-programs that construct photonic experimental blueprints for entire families of quantum states, rather than only individual target states.",
      "unknown": "This is a computational design result, not a laboratory realization of the proposed photonic experiments. Generalization was tested by checking generated states at finite larger sizes; the authors state that arbitrary code equivalence for all system sizes is not straightforward to prove. The two constructions are counted as one paper-level discovery.",
      "ai_role": "Generated human-readable Python meta-programs from short sequences of target quantum states; the model’s programs proposed reusable experimental construction rules for whole state families, which the authors checked by simulating output states at larger system sizes.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42256-025-01153-0"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-gpt-assisted-proof-reduces-zieves-hurwitz-realizability-conjecture-to-the-three-branch-point-case",
    "title": "GPT-assisted proof reduces Zieve’s Hurwitz realizability conjecture to the three-branch-point case",
    "summary": "The authors prove that, assuming Zieve-admissible branch data are realizable for three branch points, all such data with four or more branch points are realizable. This establishes that no additional obstruction arises solely from adding branch points; the underlying three-branch-point conjecture remains open. The paper discloses GPT-5.6 Sol and GPT-6 Astra assistance in exploring reduction strategies and in proof development and checking.",
    "field": "mathematics",
    "ai_technology": "OpenAI GPT-5.6 Sol and GPT-6 Astra",
    "organization": "School of Mathematical Sciences, University of Science and Technology of China; National Engineering Laboratory for Brain-inspired Intelligence Technology and Applications, University of Science and Technology of China",
    "source_url": "https://arxiv.org/abs/2609.23663",
    "source_type": "paper",
    "publication_date": "2026-09-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors prove that, assuming Zieve-admissible branch data are realizable for three branch points, all such data with four or more branch points are realizable.",
      "unknown": "This is a new conditional mathematical reduction theorem, not a resolution of Zieve’s full conjecture: the three-branch-point case remains open. The source is an arXiv preprint; human authors report reviewing AI-assisted work and accept responsibility, but no independent verification of the proof is claimed here.",
      "ai_role": "Authors explicitly report that the models assisted throughout the research, including exploration of reduction strategies and proof development/checking, as well as notation consistency and manuscript work. Humans reviewed all AI-assisted content and take responsibility for the proof.",
      "evidence_urls": [
        "https://arxiv.org/abs/2609.23663"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-gpt-6-astra-proves-a-linear-coloring-bound-for-graphs-excluding-clique-minors",
    "title": "GPT-6 Astra proves a linear coloring bound for graphs excluding clique minors",
    "summary": "The authors present a proof of the Linear Hadwiger conjecture: there is a constant C such that every graph with no K_t minor is colorable with at most Ct colors. GPT-6 Astra developed the proof after the authors directed it to the dense case and then asked it to bridge existing bounds; the resulting proof combines several known frameworks with a new bootstrap argument. The result is an asymptotic linear bound, not the stronger Hadwiger conjecture χ(G)≤h(G).",
    "field": "mathematics",
    "ai_technology": "OpenAI GPT-6 Astra (described in the paper as ChatGPT model Astra 6)",
    "organization": "University of British Columbia; University of Graz",
    "source_url": "https://arxiv.org/abs/2610.05291",
    "source_type": "paper",
    "publication_date": "2026-10-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors present a proof of the Linear Hadwiger conjecture: there is a constant C such that every graph with no K_t minor is colorable with at most Ct colors.",
      "unknown": "The result is an arXiv preprint and is not the full Hadwiger conjecture. It proves existence of an absolute linear-coloring constant for K_t-minor-free graphs. The authors take responsibility for the proof; Independent formal verification was not assessed in the source review.",
      "ai_role": "Authors disclose that the proof was developed by the model following their research directions and feedback; the model supplied a new bootstrap argument after being prompted first on a dense case and then to bridge existing bounds. Authors independently wrote the exposition, reviewed the result and take responsibility for correctness.",
      "evidence_urls": [
        "https://arxiv.org/abs/2610.05291"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-gpt-6-pro-proves-an-optimal-anticoncentration-bound-for-polynomial-littlewoodofford-inequalities",
    "title": "GPT-6 Pro proves an optimal anticoncentration bound for polynomial Littlewood–Offord inequalities",
    "summary": "The source presents a proof discovered autonomously by GPT-6 Pro establishing an O_d(r^{-1/2}) zero-probability bound for degree-d multilinear polynomials with r disjoint degree-d monomials, improving the prior bound by removing a polylogarithmic factor and resolving the Nguyen–Vu conjecture. The key argument also proves a total-influence estimate for bounded-degree rational functions, resolving a recent conjecture of Kothari, Kovacs-Deak, Wang and Yang; these are one connected paper-level result, not separate entries.",
    "field": "mathematics",
    "ai_technology": "OpenAI GPT-6 Pro",
    "organization": "Alexandr Grebennikov (affiliation not stated in arXiv record)",
    "source_url": "https://arxiv.org/abs/2610.08708",
    "source_type": "paper",
    "publication_date": "2026-10-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The source presents a proof discovered autonomously by GPT-6 Pro establishing an O_d(r^{-1/2}) zero-probability bound for degree-d multilinear polynomials with r disjoint degree-d monomials, improving the prior bound by removing a polylogarithmic factor and resolving the Nguyen–Vu conjecture.",
      "unknown": "This is an arXiv preprint. The author reports personally verifying and rewriting the proof; this review has not independently formalized it. Bound and related rational-function influence result are counted once because the paper derives them as a connected argument.",
      "ai_role": "The author states that GPT-6 Pro autonomously discovered the proof and that the author claims no credit for the proof ideas; the author subsequently verified, digested and rewrote the argument.",
      "evidence_urls": [
        "https://arxiv.org/abs/2610.08708"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-gpt-6-pro-proves-a-quadratic-size-threshold-suffices-for-random-permutations-to-contain-every-length-n-pattern",
    "title": "GPT-6 Pro proves a quadratic-size threshold suffices for random permutations to contain every length-n pattern",
    "summary": "The paper proves that a uniformly random permutation of length 32n² contains every permutation of length n as a pattern with probability at least 1−exp(−c n/log n), for an absolute c>0 and sufficiently large n. This removes the prior log-log factor from the sufficient-length bound, resolves Alon’s open 1000n² problem, and establishes the correct Θ(n²) order.",
    "field": "mathematics",
    "ai_technology": "OpenAI GPT-6 Pro",
    "organization": "Yu Chen; and Yiting Wang (affiliations not specified in arXiv record)",
    "source_url": "https://arxiv.org/abs/2610.13530",
    "source_type": "paper",
    "publication_date": "2026-10-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The paper proves that a uniformly random permutation of length 32n² contains every permutation of length n as a pattern with probability at least 1−exp(−c n/log n), for an absolute c>0 and sufficiently large n.",
      "unknown": "ArXiv preprint. Authors report checking and digesting the proof; no separate independent validation is claimed here. The result is an asymptotic sufficient bound with constant 32 (the paper explains the constant can be reduced to any value greater than 1), not a sharp threshold constant.",
      "ai_role": "The authors state that GPT-6 Pro found the proof; they do not claim credit for its proof ideas, and say they verified and digested it and take responsibility for the manuscript.",
      "evidence_urls": [
        "https://arxiv.org/abs/2610.13530"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-gpt-6-astra-resolves-the-spectral-isometry-conjecture-for-arbitrary-unital-c-algebras",
    "title": "GPT-6 Astra resolves the spectral isometry conjecture for arbitrary unital C*-algebras",
    "summary": "The paper proves that every unital surjective complex-linear spectral isometry between unital C*-algebras is a Jordan isomorphism; without unitality, such a map is a central-unitary multiple of a Jordan isomorphism. The proof establishes that entire one-parameter groups of unital spectral isometries are algebra automorphisms, then reduces the general case to the commutative case. The authors report that the result was formally verified in Lean.",
    "field": "mathematics",
    "ai_technology": "OpenAI GPT-6 Astra",
    "organization": "University of Cambridge; California Institute of Technology",
    "source_url": "https://arxiv.org/abs/2610.40470",
    "source_type": "paper",
    "publication_date": "2026-10-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The paper proves that every unital surjective complex-linear spectral isometry between unital C*-algebras is a Jordan isomorphism; without unitality, such a map is a central-unitary multiple of a Jordan isomorphism.",
      "unknown": "ArXiv preprint. The authors explicitly describe the article as a human exposition/digestion of an AI-found proof. Their manuscript says formal verification is in Lean; The reported formalization was not independently audited in the source review.",
      "ai_role": "Authors state GPT-6 Astra found the proof after a sequence of simple prompts. They then discussed it with ChatGPT 6 to understand the mechanisms, wrote the paper themselves, and used ChatGPT 6 only for minor language editing; the Lean formalization is evidence for proof checking.",
      "evidence_urls": [
        "https://arxiv.org/abs/2610.40470"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-11-gpt-6-astra-constructs-mutually-orthogonal-latin-squares-that-improve-lower-bounds-at-33-orders",
    "title": "GPT-6 Astra constructs mutually orthogonal Latin squares that improve lower bounds at 33 orders",
    "summary": "GPT-6 Astra produced explicit mutually orthogonal Latin-square constructions that improve known lower bounds for N(n) at 33 values of n from 22 through 494. The standout construction gives four MOLS of order 22, completing the last open case needed to show N(n)≥4 for every n>10. The same paper reports resulting bounds at other orders and derives a strongly regular graph with parameters (484,126,40,30) from the order-22 construction; these connected consequences are counted once as one paper-level result.",
    "field": "mathematics",
    "ai_technology": "OpenAI GPT-6 Astra, used through ChatGPT Pro and the OpenAI Codex harness",
    "organization": "Christopher D. Rosin (Constructive Codes; independent research project)",
    "source_url": "https://arxiv.org/html/2611.16299",
    "source_type": "paper",
    "publication_date": "2026-11-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2611.16299",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "GPT-6 Astra produced explicit mutually orthogonal Latin-square constructions that improve known lower bounds for N(n) at 33 values of n from 22 through 494.",
      "unknown": "This is an arXiv preprint. The constructions are finite combinatorial objects, checked by code written by the author and separately reconstructed/checked by Claude Opus 5.5 as reported in the paper. Counted once despite 33 parameter values and multiple corollaries.",
      "ai_role": "The author states GPT-6 Astra found all listed MOLS constructions and wrote their descriptions/code. The author wrote independent verification code for each construction; the paper also reports an independent Claude Opus 5.5 recreation and verification from the descriptions without the supplied code.",
      "evidence_urls": [
        "https://arxiv.org/html/2611.16299"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-gpt-6-astra-finds-a-counting-proof-of-the-erdossos-tree-embedding-conjecture",
    "title": "GPT-6 Astra finds a counting proof of the Erdős–Sós tree-embedding conjecture",
    "summary": "A counting argument proves that every finite graph with average degree greater than t−2 contains every tree on t vertices, resolving the Erdős–Sós conjecture. The proof counts adjacency-marked vertex orderings and uses an induction on the tree to bound those not supporting a rooted copy. The exposition states the sharp classical theorem and records its proof was discovered by GPT-6 Astra; a Lean repository checks the benchmark formalization of the result. Related expository papers of the same proof are aliases, not separate discoveries.",
    "field": "mathematics",
    "ai_technology": "Pre-release OpenAI GPT-6 Astra in the FrontierMath Erdős benchmark",
    "organization": "OpenAI FrontierMath Erdős project; exposition by David R. Wood (affiliation not stated in arXiv record)",
    "source_url": "https://arxiv.org/abs/2609.17877",
    "source_type": "paper",
    "publication_date": "2026-09-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A counting argument proves that every finite graph with average degree greater than t−2 contains every tree on t vertices, resolving the Erdős–Sós conjecture.",
      "unknown": "The arXiv paper is an exposition of the proof, not the original benchmark output. The linked Lean proof is machine-checked against the benchmark formalization; the registry notes a minor parity-case difference in that compared statement, while the paper’s counting lemma derives the sharp classical average-degree threshold. This remains a preprint and the formalization is not independently audited here.",
      "ai_role": "GPT-6 Astra autonomously found the counting proof while solving Erdős problem 548 in the FrontierMath Erdős benchmark; the exposition author reports the discovery and lays out the proof. A Lean 4 repository states the proof was machine-checked by a Comparator against the benchmark’s formal statement.",
      "evidence_urls": [
        "https://arxiv.org/abs/2609.17877"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2025-02-transformer-assisted-survey-identifies-87-previously-unreported-low-surface-brightness-galaxies-in-abell-194",
    "title": "Transformer-assisted survey identifies 87 previously unreported low-surface-brightness galaxies in Abell 194",
    "summary": "Transfer-learning transformer ensembles applied to deeper Hyper Suprime-Cam images of the Abell 194 cluster contributed to a vetted sample of 171 low-surface-brightness galaxies, including 87 reported as new discoveries. The models identified 159 sample members; Sérsic profile fitting and visual inspection were used to remove likely false positives, and the analysis characterized a subset of 28 ultra-diffuse galaxies. This is one paper-level survey result, not 87 separate catalog entries.",
    "field": "physics",
    "ai_technology": "Ensembles of LSBG Detection Transformers (LSBG DETR) and LSBG Vision Transformers (LSBG ViT)",
    "organization": "National Centre for Nuclear Research, Poland; Instituto de Astrofísica de Canarias; Stony Brook University; National Astronomical Observatory of Japan; collaborating institutions",
    "source_url": "https://arxiv.org/abs/2502.03142",
    "source_type": "paper",
    "publication_date": "2025-02-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202452934",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Transfer-learning transformer ensembles applied to deeper Hyper Suprime-Cam images of the Abell 194 cluster contributed to a vetted sample of 171 low-surface-brightness galaxies, including 87 reported as new discoveries.",
      "unknown": "The paper reports photometrically defined LSBGs and does not provide spectroscopic confirmation for every new object. The 87 are a sample-level count, not individually confirmed galaxy memberships; 12 of the 171 sample objects were found by visual inspection rather than the transformer models, so AI is not attributed to every sample detection.",
      "ai_role": "Applied models trained on Dark Energy Survey images to find low-surface-brightness structures in deeper HSC images from Abell 194 without fine-tuning; candidate masks were then assessed with Sérsic fits and human visual inspection.",
      "evidence_urls": [
        "https://arxiv.org/abs/2502.03142"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-01-random-forest-analysis-of-continuous-seismic-waveforms-identifies-five-previously-unreported-slow-slip-events-in-western-shikoku",
    "title": "Random-forest analysis of continuous seismic waveforms identifies five previously unreported slow-slip events in western Shikoku",
    "summary": "A random-forest regression model trained on seismic waveform features predicted GNSS-derived displacement-rate changes over roughly six years in western Shikoku, Japan. Comparing detections with earlier catalogs, the authors report five previously unreported short-term slow-slip events among 23 detected episodes; GNSS deformation patterns and fitted fault models support interpreting the transients as slow slip. The paper-level result is a retrospective detection method and event set, not prospective earthquake prediction.",
    "field": "physics",
    "ai_technology": "Random-forest regression model",
    "organization": "University of Tokyo; Tohoku University; National Research Institute for Earth Science and Disaster Resilience, Japan",
    "source_url": "https://link.springer.com/article/10.1186/s40623-025-02331-5",
    "source_type": "paper",
    "publication_date": "2026-01-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1186/s40623-025-02331-5",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A random-forest regression model trained on seismic waveform features predicted GNSS-derived displacement-rate changes over roughly six years in western Shikoku, Japan.",
      "unknown": "The study is retrospective and geographically limited to western Shikoku. Model-predicted rates can differ from observations; smoothing can make events appear longer, a practical threshold affects counts (raising it leaves two newly identified events), and the authors state that events without associated tremor may be missed. It does not establish a prospective earthquake-warning capability or a causal trigger for the event that followed the 2016 Kumamoto earthquake.",
      "ai_role": "Mapped statistical features from continuous seismic waveforms to GNSS-derived displacement rates and flagged sharp predicted increases as possible short-term slow-slip events; authors then fitted fault models and compared deformation with existing event catalogs.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s40623-025-02331-5"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-04-regional-deep-learning-phase-picking-reveals-24-additional-earthquakes-in-a-gorda-ridge-swarm",
    "title": "Regional deep-learning phase picking reveals 24 additional earthquakes in a Gorda Ridge swarm",
    "summary": "A long-window regional phase picker and phase-association workflow detected 34 earthquakes in a 29 February 2024 swarm near the Gorda Ridge, offshore northern California: all 10 reported by the USGS plus 24 additional events. The detections were possible from stations roughly 300 km from the swarm, beyond the distance range represented by standard local-event pickers. This is one paper-level catalog result, not 24 separate discoveries.",
    "field": "physics",
    "ai_technology": "SKYNET deep-learning P/S phase picker with GENIE graph-neural-network phase association",
    "organization": "Stanford University",
    "source_url": "https://seismica.library.mcgill.ca/article/view/1431",
    "source_type": "paper",
    "publication_date": "2025-04-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.26443/seismica.v4i1.1431",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A long-window regional phase picker and phase-association workflow detected 34 earthquakes in a 29 February 2024 swarm near the Gorda Ridge, offshore northern California: all 10 reported by the USGS plus 24 additional events.",
      "unknown": "The application covers one swarm day and a specific offshore region. The paper reports 24 additional algorithmically detected events and estimates local magnitudes, but does not claim every event had independent analyst confirmation. SKYNET is designed to complement local pickers; its near-source performance is weaker than PhaseNet, and the result is not a new tectonic mechanism.",
      "ai_role": "SKYNET picked long-duration regional seismic waveforms and GENIE associated the picks into events; their combined workflow supplied the new event detections in sparsely instrumented offshore seismicity.",
      "evidence_urls": [
        "https://seismica.library.mcgill.ca/article/view/1431"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-deep-learning-reveals-migrating-foreshocks-during-laboratory-fault-nucleation",
    "title": "Deep learning reveals migrating foreshocks during laboratory fault nucleation",
    "summary": "In biaxial stick-slip experiments on a 0.85 m saw-cut granodiorite fault, the AEbagging ensemble deep-learning picker generated acoustic-emission catalogs for eight laboratory earthquake cycles. The model-derived catalogs revealed a foreshock sequence migrating from a lower-right fault segment toward the upper-left region where the mainshock began, alongside a precursory decrease in the b-value. This is a paper-level laboratory seismology result.",
    "field": "physics",
    "ai_technology": "AEbagging ensemble deep-learning acoustic-emission picker",
    "organization": "Institute of Geophysics, China Earthquake Administration; MIT; University of Strasbourg/CNRS",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2024JB029934",
    "source_type": "paper",
    "publication_date": "2025-07-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2024JB029934",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In biaxial stick-slip experiments on a 0.85 m saw-cut granodiorite fault, the AEbagging ensemble deep-learning picker generated acoustic-emission catalogs for eight laboratory earthquake cycles.",
      "unknown": "This is an interpretation of eight controlled stick-slip events on one laboratory fault, not evidence that the same migration pattern is a reliable precursor to natural earthquakes. The authors suggest a possible relation to fault-surface roughness; that causal explanation is not established. The result depends on the model-generated event catalog.",
      "ai_role": "A supervised ensemble of deep neural networks detected acoustic emissions and picked their primary arrivals from continuous multichannel laboratory recordings; its generated event catalogs enabled the authors to resolve the temporal and spatial foreshock migration pattern during nucleation.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2024JB029934"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-08-symbolic-regression-turbulence-analysis-separates-transferable-flow-physics-from-dataset-specific-closure-terms",
    "title": "Symbolic-regression turbulence analysis separates transferable flow physics from dataset-specific closure terms",
    "summary": "Using high-fidelity data for plane channels, zero-pressure-gradient turbulent boundary layers and adverse-pressure-gradient boundary layers, the authors used symbolic regression to decompose learned turbulence-closure corrections into inner-layer, outer-layer and pressure-gradient components. They report that retaining the generalisable inner-layer term and adding flow-specific corrections improved predictions on held-out boundary-layer and aerofoil configurations, including unseen geometries and Reynolds numbers, outperforming baseline RANS closures in a priori and a posteriori tests. The authors identify conflation of transferable and training-specific flow behavior as a source of poor model generalization in the tested setting.",
    "field": "physics",
    "ai_technology": "Symbolic regression applied to high-fidelity turbulence simulation data",
    "organization": "Northwestern Polytechnical University; Pennsylvania State University",
    "source_url": "https://www.cambridge.org/core/journals/journal-of-fluid-mechanics/article/abs/dataenabled-discovery-of-specific-and-generalisable-turbulence-closures/4077B2AAC80E5386F5A941ED13FBB451",
    "source_type": "paper",
    "publication_date": "2025-08-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1017/jfm.2025.10332",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using high-fidelity data for plane channels, zero-pressure-gradient turbulent boundary layers and adverse-pressure-gradient boundary layers, the authors used symbolic regression to decompose learned turbulence-closure corrections into inner-layer, outer-layer and pressure-gradient components.",
      "unknown": "The finding concerns model-derived closure components and validation on selected numerical flow configurations, not a new direct physical measurement. The claimed root cause of poor generalization is the authors' conclusion about data-driven turbulence closures in the studied context, not a universal causal theorem about all turbulence models. Out-of-sample tests and comparative RANS simulations support the stated transfer result.",
      "ai_role": "Symbolic regression generated interpretable closure expressions from progressively expanded flow datasets. Decomposition of the learned terms let the researchers distinguish inner-layer components that transferred across flow configurations from outer-layer and pressure-gradient corrections that were specific to flow classes, enabling improved out-of-training-flow predictions.",
      "evidence_urls": [
        "https://www.cambridge.org/core/journals/journal-of-fluid-mechanics/article/abs/dataenabled-discovery-of-specific-and-generalisable-turbulence-closures/4077B2AAC80E5386F5A941ED13FBB451"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-claude-discovers-algorithms-that-refute-the-3sum-and-apsp-time-complexity-hypotheses",
    "title": "Claude discovers algorithms that refute the 3SUM and APSP time-complexity hypotheses",
    "summary": "Alman and Vassilevska Williams present deterministic algorithms for 3SUM on polynomial-size integers in O(n^1.9992) time and all-pairs shortest paths on directed graphs with polynomially bounded integer weights in O(n^2.9995) time, the first polynomial improvements over the textbook exponents. The single underlying technique computes selected entries of thin matrix products more efficiently, and known reductions yield the main algorithmic consequences. The paper says an internal Anthropic Claude research model discovered the algorithm autonomously; the human authors then understood, simplified, strengthened and extended it. One paper-level theoretical-computer-science result is counted.",
    "field": "computer science",
    "ai_technology": "Internal Anthropic Claude research model (specific model version undisclosed)",
    "organization": "Anthropic; Columbia University; Massachusetts Institute of Technology",
    "source_url": "https://arxiv.org/abs/2610.06783",
    "source_type": "paper",
    "publication_date": "2026-10-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Alman and Vassilevska Williams present deterministic algorithms for 3SUM on polynomial-size integers in O(n^1.9992) time and all-pairs shortest paths on directed graphs with polynomially bounded integer weights in O(n^2.9995) time, the first polynomial improvements over the textbook exponents.",
      "unknown": "This is an arXiv preprint as of the checked record. The stated time bounds are for polynomially bounded integer inputs; real-valued variants use Las Vegas randomization and different bounds. Lean certification covers the explicitly enumerated theorems, not every statement or presentation detail in the 76-page paper. The authors take responsibility for the paper and developed important extensions/consequences after the initial AI-discovered algorithm.",
      "ai_role": "The paper states that Claude was tasked with checking and improving cryptographic constructions but instead developed the new thin-matrix-product algorithm, first for average-case and then worst-case problems, in a 16-million-output-token session with no human input. The authors subsequently worked to understand, simplify, strengthen and extend the algorithm. Anthropic later used an internal model to Lean-certify specified main results.",
      "evidence_urls": [
        "https://arxiv.org/abs/2610.06783"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2022-10-semi-supervised-machine-learning-identifies-two-strong-gravitational-lenses-in-the-deep-lens-survey",
    "title": "Semi-supervised machine learning identifies two strong gravitational lenses in the Deep Lens Survey",
    "summary": "A ResNetV2-based semi-supervised workflow ranked images from the 20-square-degree Deep Lens Survey and surfaced two Grade-A lens systems for follow-up. Keck spectroscopy secured a source redshift of z=1.81 for DLS212072337, while DLS432021848 has only a tentative single-line redshift but supporting high-resolution Hubble morphology; the paper counts both as lensing systems, with the latter less secure. The broader visually inspected list contains additional candidates, not confirmed lenses.",
    "field": "physics",
    "ai_technology": "ResNetV2 neural network with semi-supervised MixMatch or Pi-Model and GAN-based image augmentation",
    "organization": "University of California, Davis; University of California, Santa Cruz; University of California, Berkeley",
    "source_url": "https://arxiv.org/abs/2211.00047",
    "source_type": "paper",
    "publication_date": "2022-10-31",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A ResNetV2-based semi-supervised workflow ranked images from the 20-square-degree Deep Lens Survey and surfaced two Grade-A lens systems for follow-up.",
      "unknown": "This is a paper-level result about a survey search and validated lens identification, not a claim that all 22 visually graded systems are confirmed. DLS212072337 has secure spectroscopic source and deflector redshifts; DLS432021848 has an ambiguous single emission line and is judged likely from HST morphology rather than conclusive spectroscopy. arXiv version 1 is a preprint; current full text was updated in 2026.",
      "ai_role": "Ranked galaxy-survey images for strong-lensing morphology; the top-ranked systems were then human-inspected and followed up spectroscopically/imaged to validate lensing.",
      "evidence_urls": [
        "https://arxiv.org/abs/2211.00047"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-11-machine-learning-guided-search-discovers-and-synthesizes-the-rare-earth-free-magnet-fe3cob2",
    "title": "Machine-learning-guided search discovers and synthesizes the rare-earth-free magnet Fe₃CoB₂",
    "summary": "A feedback workflow combining machine learning, an adaptive genetic algorithm, first-principles calculations, and experiments identified and synthesized Fe₃CoB₂, a rare-earth-free magnetic compound. Magnetic measurements reported saturation polarization Js=1.39 T and anisotropy K1=1.2 MJ/m³, properties the authors describe as suitable for rare-earth-free permanent-magnet applications. This is a discovery and synthesis of a material compound; commercial performance or device readiness was not demonstrated.",
    "field": "materials",
    "ai_technology": "Machine-learning-guided adaptive feedback with an adaptive genetic algorithm and first-principles calculations",
    "organization": "Ames Laboratory; Iowa State University; University of Nebraska–Lincoln; University of Texas at Austin; University of Tokyo; Yantai University; Zhejiang Agriculture and Forestry University",
    "source_url": "https://doi.org/10.1073/pnas.2204485119",
    "source_type": "paper",
    "publication_date": "2022-11-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A feedback workflow combining machine learning, an adaptive genetic algorithm, first-principles calculations, and experiments identified and synthesized Fe₃CoB₂, a rare-earth-free magnetic compound.",
      "unknown": "The paper reports an experimentally synthesized compound and magnetic measurements, but suitability is for potential rare-earth-free permanent-magnet applications; no commercial device demonstration is reported. The workflow combines ML with adaptive genetic search, DFT and human synthesis/measurement.",
      "ai_role": "Machine learning screened and informed the material search, while adaptive genetic structure search and first-principles calculations guided candidates; researchers synthesized the selected compound and measured its magnetic properties. The discovery is attributed to the integrated workflow, not the ML model alone.",
      "evidence_urls": [
        "https://doi.org/10.1073/pnas.2204485119"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-claude-code-produces-the-first-primary-lund-jet-plane-density-measurement-in-electronpositron-collisions",
    "title": "Claude Code produces the first primary Lund jet-plane density measurement in electron–positron collisions",
    "summary": "A Claude Code agent pipeline analyzed archived 1994 ALEPH hadronic-Z data and reported the first primary Lund jet-plane density measurement in e⁺e⁻ collisions. Its unfolded 57-bin fiducial result shows the expected triangular QCD radiation structure and a shallow low-transverse-momentum turnover; the integrated mean primary-emission count is 4.751 ± 0.224 per hemisphere. This is one new measurement, not a claim of discovering new particles or physics beyond QCD.",
    "field": "physics",
    "ai_technology": "Claude Code agents orchestrated by the Just Furnish Context (JFC) framework",
    "organization": "Massachusetts Institute of Technology; NSF AI Institute for Artificial Intelligence and Fundamental Interactions; CERN; ALEPH Collaboration archival data",
    "source_url": "https://arxiv.org/html/2603.20179v3",
    "source_type": "paper",
    "publication_date": "2026-03-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2603.20179",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A Claude Code agent pipeline analyzed archived 1994 ALEPH hadronic-Z data and reported the first primary Lund jet-plane density measurement in e⁺e⁻ collisions.",
      "unknown": "This is an arXiv preprint, and the authors explicitly qualify the novel measurement as subject to independent verification. It uses 1994 ALEPH data and one available full-simulation sample; the corrected density is prior/model dependent, with a reported 6.0% density-weighted total uncertainty and 4.7% uncertainty on the integrated count. The shallow residual shape is within current generator-theory uncertainty, so the paper frames it as a benchmark rather than evidence excluding generators.",
      "ai_role": "Agents planned and executed the event selection, jet declustering, unfolding, uncertainty propagation, statistical checks and report drafting using archived ALEPH data and literature retrieval. The paper attributes the new Lund-plane measurement to an autonomous AI-agent analysis pipeline; human authors supplied the broad task and review framework.",
      "evidence_urls": [
        "https://arxiv.org/html/2603.20179v3"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-03-time-delay-neural-networks-expose-three-fault-development-phases-across-laboratory-pressure-regimes",
    "title": "Time-delay neural networks expose three fault-development phases across laboratory pressure regimes",
    "summary": "A time-delay neural network trained on multi-parameter acoustic-emission records from Alzo-granite triaxial tests at 5–40 MPa resolved three recurring phases: microcrack nucleation marked by waveform-scattering changes, fault growth with evolving spatial and magnitude distributions, and coalescence marked by peak-delay increases and b-value changes. The model also predicted stress-drop timing and size across axial-splitting and shear-localization failures. This is a laboratory result, not a natural-earthquake precursor claim.",
    "field": "physics",
    "ai_technology": "Genetic-algorithm-optimized time-delay neural network (TDNN)",
    "organization": "University of Birmingham; University of Cambridge",
    "source_url": "https://www.nature.com/articles/s43247-025-03003-8",
    "source_type": "paper",
    "publication_date": "2026-03-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43247-025-03003-8",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A time-delay neural network trained on multi-parameter acoustic-emission records from Alzo-granite triaxial tests at 5–40 MPa resolved three recurring phases: microcrack nucleation marked by waveform-scattering changes, fault growth with evolving spatial and magnitude distributions, and coalescence marked by peak-delay increases and b-value changes.",
      "unknown": "The analysis uses four laboratory experiments in Alzo granite, so results are sample- and apparatus-limited and do not establish natural-earthquake predictability. Training-data smoothing and the small number of experiments constrain generalization; stress estimates show more variability and modelled subtle phase transitions are less accurate. The observed phase interpretation is based on model-assisted analysis of acoustic emissions and measured mechanical data.",
      "ai_role": "The TDNN integrated time-dependent acoustic-emission waveform features (peak delay, coda Q/scattering) and occurrence measures (event rate, Gutenberg–Richter b-value, spatial fractal dimension), dynamically weighting them to model stress/strain evolution and identify repeatable phases across pressure conditions. Leave-one-experiment-out validation tested cross-condition generalization.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43247-025-03003-8"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-06-machine-learning-seismicity-maps-a-250-km-eastern-edge-of-the-subducted-yakutat-microplate",
    "title": "Machine-learning seismicity maps a 250-km eastern edge of the subducted Yakutat microplate",
    "summary": "An expanded machine-learning earthquake catalog and complementary receiver-function and ambient-noise images reveal a previously unrecognized, roughly 250-km linear cluster of about 1,750 earthquakes that the authors interpret as tracing the northeastern/eastern edge of the subducted Yakutat microplate beneath south-central Alaska. The inferred slab edge lies beneath the curved Denali fault; its possible role in the 2002 Mw 7.9 earthquake’s nucleation is a hypothesis, not an established cause.",
    "field": "physics",
    "ai_technology": "Modified EQTransformer machine-learning seismic phase picker, REAL phase associator, and NonLinLoc locator",
    "organization": "Australian National University; South Dakota School of Mines and Technology; University of California, Davis",
    "source_url": "https://pubs.geoscienceworld.org/ssa/tsr/article/6/2/230/731510/Razor-Sharp-Edge-The-Yakutat-Slab-Dissecting-South",
    "source_type": "paper",
    "publication_date": "2026-06-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1785/0320250055",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An expanded machine-learning earthquake catalog and complementary receiver-function and ambient-noise images reveal a previously unrecognized, roughly 250-km linear cluster of about 1,750 earthquakes that the authors interpret as tracing the northeastern/eastern edge of the subducted Yakutat microplate beneath south-central Alaska.",
      "unknown": "This is a localized tectonic interpretation from data collected between 2018 and 2021, not prospective earthquake prediction. The 64,870-event catalog reports average location uncertainties of about 5.0 km latitude, 5.2 km longitude and 8.6 km depth; the authors support the slab-edge interpretation with seismicity, receiver functions, tremor distribution and ambient-noise tomography. They propose—but do not establish—that the slab edge influenced nucleation of the 2002 Denali earthquake and the position of volcanic fields.",
      "ai_role": "The ML phase picker helped construct a denser earthquake catalog from permanent and temporary station records. The 64,870-event catalog exposed the previously unseen long linear earthquake cluster; receiver-function and ambient-noise tomography supplied independent complementary images used to interpret its relationship to the Yakutat slab edge.",
      "evidence_urls": [
        "https://pubs.geoscienceworld.org/ssa/tsr/article/6/2/230/731510/Razor-Sharp-Edge-The-Yakutat-Slab-Dissecting-South"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-gpt-5-6-constructs-a-counterexample-to-talagrands-operator-cotype-problem",
    "title": "GPT-5.6 constructs a counterexample to Talagrand’s operator cotype problem",
    "summary": "The paper gives a negative answer, already for q=2, to Talagrand’s question whether an operator’s Rademacher cotype is universally bounded by its Gaussian cotype and its (q,1)-summing norm. For every sufficiently large power-of-two dimension it constructs a Banach space and operator violating such a universal bound. The same construction also yields a negative answer to Kwapień’s decomposition problem.",
    "field": "mathematics",
    "ai_technology": "ChatGPT GPT-5.6",
    "organization": "Xinglong Wu (author affiliation not stated in the arXiv record)",
    "source_url": "https://arxiv.org/abs/2609.19731",
    "source_type": "paper",
    "publication_date": "2026-09-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2609.19731",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The paper gives a negative answer, already for q=2, to Talagrand’s question whether an operator’s Rademacher cotype is universally bounded by its Gaussian cotype and its (q,1)-summing norm.",
      "unknown": "This is a single-author arXiv preprint, not a peer-reviewed publication. GPT-5.6 is credited with discovering the counterexample during discussions; the author says they independently verified all arguments. No Lean or other proof-assistant formalization is reported. The mathematical scope is the stated q=2 counterexample and its stated corollary, not a complete characterization of operator cotype.",
      "ai_role": "The author states that the counterexample was discovered during discussions with GPT-5.6. The paper presents an explicit Walsh–Hadamard-based construction and proof; the author independently verified the arguments.",
      "evidence_urls": [
        "https://arxiv.org/abs/2609.19731"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-a-dense-nodal-array-and-machine-learning-reveal-a-blind-fault-zone-beneath-the-san-fernando-valley",
    "title": "A dense nodal array and machine learning reveal a blind fault zone beneath the San Fernando Valley",
    "summary": "A machine-learning workflow analyzed one month of recordings from a 140-station nodal array and located 62 earthquakes, including 36 not in the regional catalog. Their distribution exposes a previously unidentified 5.3 × 4 km blind fault zone about 5 km beneath the southern San Fernando Valley; the small-event cluster is rare in the regional catalog despite a 2014 Mb 4.4 earthquake in the zone.",
    "field": "physics",
    "ai_technology": "Previously developed machine-learning seismic phase picker, fine-tuned on similar waveforms, with two-step event association",
    "organization": "University of Texas at Austin; University of Houston",
    "source_url": "https://arxiv.org/abs/2505.03496",
    "source_type": "paper",
    "publication_date": "2025-05-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2505.03496",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A machine-learning workflow analyzed one month of recordings from a 140-station nodal array and located 62 earthquakes, including 36 not in the regional catalog.",
      "unknown": "This is a one-month temporary-array study reported as an arXiv preprint. The fault-zone dimensions and location are inferred from a small cluster of newly detected microearthquakes, not from direct imaging or a long-term catalog; it informs hazard mapping but does not predict future earthquakes.",
      "ai_role": "The model detected weak earthquake signals and picked seismic phases in high-noise recordings, enabling the expanded local event catalog from which the blind fault-zone cluster was interpreted.",
      "evidence_urls": [
        "https://arxiv.org/abs/2505.03496"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-machine-learning-attributes-post-2015-emissions-to-intensification-of-europes-recent-heat-extremes",
    "title": "Machine learning attributes post-2015 emissions to intensification of Europe’s recent heat extremes",
    "summary": "A diffusion model trained on ten CMIP6 climate models estimated event-specific temperature changes conditional on observed meteorology. It found >95% probability that anthropogenic emissions since the 2015 Paris Agreement intensified Europe’s summer temperature extremes in 2021–2025; for the June 2025 heatwave, regional mean temperature is estimated to have been 0.34 °C lower under 2015 cumulative emissions, with >99% probability.",
    "field": "climate",
    "ai_technology": "Generative diffusion model for spatially explicit temperature fields",
    "organization": "Stanford University; Boston University; University of Auckland; Colorado State University",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2026GL125079",
    "source_type": "paper",
    "publication_date": "2026-09-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2026GL125079",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A diffusion model trained on ten CMIP6 climate models estimated event-specific temperature changes conditional on observed meteorology.",
      "unknown": "This event-specific attribution is conditioned on observed large-scale meteorology and trained on ten CMIP6 models; it is not a direct experiment that removed emissions. The paper emphasizes uncertainty from internal variability/model spread and reports robust regional-mean effects, while grid-cell-level changes are more limited. The June 2025 value is a median modeled counterfactual estimate.",
      "ai_role": "The model learned climate-model relationships between meteorological conditions, cumulative CO₂ emissions and daily temperature fields, then generated conditional counterfactual temperature distributions to attribute the incremental post-2015 emissions effect to recent European heat extremes.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2026GL125079"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-06-observation-trained-machine-learning-halves-estimated-arctic-winter-sea-ice-surface-cooling",
    "title": "Observation-trained machine learning halves estimated Arctic winter sea-ice surface cooling",
    "summary": "A neural network trained on observations from five Arctic field campaigns reduced ERA5 errors in the Arctic sea-ice surface energy budget and removed ERA5’s wintertime near-surface warm bias of about 4 K. The corrected estimates imply about 50% weaker winter surface cooling and 80–120 cm seasonal thermodynamic ice growth, closer to SMOS–CryoSat thickness increases than ERA5’s 150–200 cm estimate.",
    "field": "climate",
    "ai_technology": "Feed-forward neural network product SEBai",
    "organization": "Alfred Wegener Institute; Helmholtz Centre Hereon; Helmholtz AI; Deutsches Klimarechenzentrum; NOAA Physical Sciences Laboratory; University of Leeds; Norwegian Polar Institute; Karlsruhe Institute of Technology; Stockholm University; Washington State University",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL121289",
    "source_type": "paper",
    "publication_date": "2026-06-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025GL121289",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A neural network trained on observations from five Arctic field campaigns reduced ERA5 errors in the Arctic sea-ice surface energy budget and removed ERA5’s wintertime near-surface warm bias of about 4 K.",
      "unknown": "Validation uses leave-one-campaign-out cross-validation and multiple field campaigns, but direct flux observations remain sparse in space and time. The climatology applies to ice-covered Arctic regions north of 70°N with sea-ice concentration above 80%; authors do not recommend extrapolation to marginal ice zones or coastal regions. The ice-growth values are thermodynamic estimates rather than direct pan-Arctic measurements.",
      "ai_role": "The network learned the relationship between ERA5 atmospheric states and observed fluxes from MOSAiC, SHEBA, AO2018, N-ICE2015 and ARTofMELT campaigns, then bias-corrected hourly Arctic sea-ice surface fluxes for 1994–2024 to derive a revised regional climatology and ice-growth estimates.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL121289"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-transformer-guided-genetic-search-finds-six-new-champion-linear-error-correcting-codes-over-f8",
    "title": "Transformer-guided genetic search finds six new champion linear error-correcting codes over F₈",
    "summary": "A transformer trained to predict minimum Hamming distance, paired with a genetic algorithm and exact Brouwer–Zimmermann distance checks, found at least six previously unknown champion generalized toric codes over F₈ at dimensions 18, 22 and 38 (length 49). The article reports more than 500 champion codes in its F₈ search and up to a twofold reduction in exact distance evaluations versus random search. The six records at the three newly improved parameter settings are one paper-level coding-theory result.",
    "field": "computer science",
    "ai_technology": "Transformer model adapted from GPT-2, coupled with a genetic algorithm",
    "organization": "University of London; University of Warwick; University of Oxford; London Institute for Mathematical Sciences",
    "source_url": "https://www.nature.com/articles/s44387-026-00077-3",
    "source_type": "paper",
    "publication_date": "2026-03-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44387-026-00077-3",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A transformer trained to predict minimum Hamming distance, paired with a genetic algorithm and exact Brouwer–Zimmermann distance checks, found at least six previously unknown champion generalized toric codes over F₈ at dimensions 18, 22 and 38 (length 49).",
      "unknown": "The codes are newly discovered mathematical constructions in the generalized toric family, with minimum distances verified computationally using established exact algorithms and compared to known records. The reported >500 total includes known/previous champion codes, while six at dimensions 18, 22 and 38 are identified as novel; these are not six separate catalog entries. The method searches a restricted family and does not establish optimal codes over all linear codes or quantum codes.",
      "ai_role": "The transformer predicts generalized toric codes’ minimum Hamming distances and guides genetic search toward promising generator matrices. Candidate champions are then checked with the Brouwer–Zimmermann algorithm and compared with known bounds; the paper reports six previously unknown champion codes at F₈ dimensions 18, 22 and 38.",
      "evidence_urls": [
        "https://www.nature.com/articles/s44387-026-00077-3"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-gpt-5-6-helps-prove-an-exact-mapping-between-frustrated-and-field-driven-spin-chains",
    "title": "GPT-5.6 helps prove an exact mapping between frustrated and field-driven spin chains",
    "summary": "A single-author preprint proves that every finite one-dimensional open O(n) spin chain with arbitrary inhomogeneous nearest- and next-nearest-neighbor interactions has an exact, temperature-independent Hamiltonian mapping to a simpler nearest-neighbor chain with axial single-spin potentials. A parallel result for q-state Potts chains yields a closed-form solution for every q ≥ 2 and finite chain length. The theorem generalizes a 1990 Heisenberg-chain challenge and connects spontaneous collective behavior from competing interactions to field-induced behavior.",
    "field": "physics",
    "ai_technology": "OpenAI GPT-5.6 Sol; GPT-5.6 Sol Pro for final adversarial manuscript review",
    "organization": "Brookhaven National Laboratory",
    "source_url": "https://arxiv.org/abs/2609.00322",
    "source_type": "paper",
    "publication_date": "2026-08-31",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2609.00322",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A single-author preprint proves that every finite one-dimensional open O(n) spin chain with arbitrary inhomogeneous nearest- and next-nearest-neighbor interactions has an exact, temperature-independent Hamiltonian mapping to a simpler nearest-neighbor chain with axial single-spin potentials.",
      "unknown": "This is a single-author arXiv preprint, not yet peer reviewed. The theorem concerns classical one-dimensional open chains and exact Hamiltonian mappings; the paper discusses possible experimental relevance but reports no experimental test. The author assumes responsibility for reviewing and verifying AI-generated content; Mathematica 14.3 was also used to verify results.",
      "ai_role": "The author credits GPT-5.6 Sol with scientific discovery, including generating the recursive Householder moving-frame proof and recognizing broad generalizations. The author identified the hidden reciprocity, extended the theorem to arbitrary spatial inhomogeneity, and states that all AI-generated scientific content was reviewed and verified. GPT-5.6 Sol Pro conducted a final adversarial review.",
      "evidence_urls": [
        "https://arxiv.org/abs/2609.00322"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-generative-ai-designs-and-experimentally-validates-an-ultrathin-broadband-enz-absorber",
    "title": "Generative AI designs and experimentally validates an ultrathin broadband ENZ absorber",
    "summary": "A generative deep neural network searched multilayer indium-tin-oxide/aluminum-doped-zinc-oxide designs and selected an epsilon-near-zero absorber stack. After the researchers merged two similar adjacent layers for fabrication, the approximately 159-nm device experimentally maintained more than 98% absorption over a 1,077-nm near-infrared bandwidth at 49° p-polarized illumination. The modeled design predicted 1,688 nm bandwidth; the experiment’s detector cutoff limited measurements above 2,600 nm.",
    "field": "materials",
    "ai_technology": "GLONet generative deep neural network, trained in simulation loop with differentiable transfer-matrix solver TMM-Fast",
    "organization": "University of California, Irvine; Los Alamos National Laboratory; Center for Integrated Nanotechnologies",
    "source_url": "https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.77966",
    "source_type": "paper",
    "publication_date": "2026-09-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.77966",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A generative deep neural network searched multilayer indium-tin-oxide/aluminum-doped-zinc-oxide designs and selected an epsilon-near-zero absorber stack.",
      "unknown": "The demonstrated result is a fabricated device under p-polarized illumination at a 49° incidence angle using a prism-coupled ENZ mode; it is not an angle-independent or application-qualified absorber. Measurement above 2,600 nm was limited by the detector, so the 1,688-nm bandwidth is a simulation result and the measured bandwidth is 1,077 nm. The network-produced four-layer design was manually simplified by merging two similar layers before fabrication. Earlier work by the same group used neural networks to design homogeneous ITO stacks; this paper reports a distinct heterogeneous ITO/AZO design with experimental validation. No commercial application is demonstrated.",
      "ai_role": "The GLONet maps random latent vectors to multilayer designs, optimizing per-layer refractive/material parameters and thickness using differentiable optical simulations to maximize broadband absorption. The authors generated 10,000 candidate stacks and selected a high-performing design for experimental fabrication and measurement. A manual layer-merging step simplified the four-layer design to three layers before fabrication.",
      "evidence_urls": [
        "https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.77966"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-ai-assisted-hubble-asteroid-sample-extends-the-shallow-size-distribution-slope-to-h20",
    "title": "AI-assisted Hubble asteroid sample extends the shallow size-distribution slope to H≈20",
    "summary": "Using 632 serendipitously imaged Hubble Space Telescope asteroids found by the prior AI-assisted Hubble Asteroid Hunter survey, the authors fit trail parallax to estimate distances, absolute magnitudes, and approximate sizes. Their cumulative magnitude distribution has a shallower slope (α≈0.26) from H=15 to about H=20, extending the previously reported trend by roughly two magnitudes; the analyzed sample is dominated by potential main-belt objects smaller than 1 km.",
    "field": "physics",
    "ai_technology": "Google Cloud deep-learning asteroid-trail detector trained on Hubble Asteroid Hunter citizen-science classifications (upstream survey)",
    "organization": "European Space Agency; European Space Astronomy Centre; Autonomous University of Madrid; Astronomical Institute of the Romanian Academy; University of Craiova; Jet Propulsion Laboratory; Université Côte d’Azur",
    "source_url": "https://www.aanda.org/articles/aa/pdf/2024/03/aa46771-23.pdf",
    "source_type": "paper",
    "publication_date": "2024-03-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202346771",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using 632 serendipitously imaged Hubble Space Telescope asteroids found by the prior AI-assisted Hubble Asteroid Hunter survey, the authors fit trail parallax to estimate distances, absolute magnitudes, and approximate sizes.",
      "unknown": "The 632-object sample comprises 454 unmatched and 178 known objects; unmatched does not mean every object is independently confirmed or has a reliable orbit. Only about 40% of raw objects yielded meaningful distance estimates after filtering. Sizes assume spherical objects and a uniform albedo of 0.15. The authors state that the combined archive sample is not a dedicated or observationally unbiased survey, and the result is an extension of the size-distribution slope to H≈20 rather than a complete census.",
      "ai_role": "The primary study states that the input asteroid trails were detected in the preceding Hubble Asteroid Hunter work using a deep-learning algorithm trained on volunteer classifications. This AI-assisted search supplied the previously unidentified trail sample; the present paper then applies a parallax-fitting method and magnitude-distribution analysis to characterize it. The quantitative population result depends on this archive-scale AI-assisted detection pipeline and the present physical analysis.",
      "evidence_urls": [
        "https://www.aanda.org/articles/aa/pdf/2024/03/aa46771-23.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-ai-guided-corona-imagery-search-finds-four-archaeologically-verified-tells-in-abu-ghraib",
    "title": "AI-guided CORONA imagery search finds four archaeologically verified tells in Abu Ghraib",
    "summary": "A fine-tuned deep-learning model trained with 1960s CORONA satellite images flagged eight previously unrecognized tell candidates in Iraq's Abu Ghraib district. Archaeologists investigated all eight; four contained surface ceramic sherds confirming archaeological remains and enabling dating. The AI suggestions led the field team to inspect locations that routine surveys would not have visited.",
    "field": "humanities",
    "ai_technology": "MANet deep convolutional segmentation network with attention, transfer learning and two-stage fine-tuning",
    "organization": "University of Bologna",
    "source_url": "https://arxiv.org/pdf/2507.13420",
    "source_type": "paper",
    "publication_date": "2025-07-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2507.13420",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A fine-tuned deep-learning model trained with 1960s CORONA satellite images flagged eight previously unrecognized tell candidates in Iraq's Abu Ghraib district.",
      "unknown": "This is an arXiv preprint. The results are limited to mounded tell sites in Abu Ghraib and depend on a small training set (208 images) and archaeologist-selected field follow-up. Four of eight AI-suggested sites were confirmed; the other four should not be described as archaeological finds. The source reports that the four confirmed sites were destroyed or no longer visible on the ground, with sherds establishing their presence and dates.",
      "ai_role": "Researchers retrained the existing neural model using CORONA historical imagery and generated probability heatmaps that archaeologists used to select previously unrecognized locations for field reconnaissance. Eight AI-suggested sites were investigated; four were confirmed through collected and studied ceramic sherds. The authors state the four locations would not have been investigated in routine field surveys without the model's suggestions.",
      "evidence_urls": [
        "https://arxiv.org/pdf/2507.13420"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-deep-learning-compact-catalogue-adds-116-previously-unreported-galaxy-clusters",
    "title": "Deep-learning ComPACT catalogue adds 116 previously unreported galaxy clusters",
    "summary": "Analysis of the CNN-detected ComPACT Sunyaev–Zel’dovich catalogue identified 116 previously unreported galaxy clusters with new redshift estimates, including five massive systems (M500c > 6 × 10^14 solar masses) at z > 0.7. The authors report 158 new cluster-mass determinations and higher completeness than ACT DR5 at fixed purity, especially for massive high-redshift systems.",
    "field": "physics",
    "ai_technology": "ComPACT deep-learning convolutional neural network catalogue, followed by optical photometric redshift and SZ mass analysis",
    "organization": "Space Research Institute, Russian Academy of Sciences; HSE University; Lomonosov Moscow State University",
    "source_url": "https://www.aanda.org/articles/aa/full_html/2026/07/aa59201-26/aa59201-26.html",
    "source_type": "paper",
    "publication_date": "2026-07-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202659201",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Analysis of the CNN-detected ComPACT Sunyaev–Zel’dovich catalogue identified 116 previously unreported galaxy clusters with new redshift estimates, including five massive systems (M500c > 6 × 10^14 solar masses) at z > 0.7.",
      "unknown": "The ComPACT catalogue contains 2,962 SZ-selected candidates; the paper confirms approximately 60% as clusters, so the full candidate list must not be treated as confirmed. Its newly reported cluster count relies on cross-matches and optical photometric-redshift significance criteria with an estimated 5% false-positive rate; most are not presented as new spectroscopic confirmations. Masses are scaling-relation estimates. The five high-mass, high-redshift systems are identified with redshift algorithms and SZ measurements, and their interpretation remains subject to catalogue selection/completeness and mass-estimation uncertainties.",
      "ai_role": "The CNN detected SZ cluster candidates in combined ACT+Planck maps, creating the candidate catalogue analyzed here. The authors used existing catalogues and DESI Legacy Imaging Survey photometry to validate systems and estimate redshifts, and ACT+Planck/Planck Compton-y maps with SZ scaling relations to derive masses. The CNN catalogue has higher completeness than ACT DR5 at fixed purity, particularly at high mass and redshift; it is the entry point enabling these new cluster identifications and measurements.",
      "evidence_urls": [
        "https://www.aanda.org/articles/aa/full_html/2026/07/aa59201-26/aa59201-26.html"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-pcfnet-guided-spectroscopy-confirms-four-protoclusters-at-redshift-3-33-4",
    "title": "PCFNet-guided spectroscopy confirms four protoclusters at redshift 3.3–3.4",
    "summary": "Subaru/MOIRCS spectroscopy of four PCFNet-selected candidate regions confirmed four protoclusters at z = 3.316, 3.403, 3.408 and 3.425, across three regions, with galaxy-density peaks more than five times the uniform distribution. Twenty-three spectroscopically identified members are largely on the star-forming main sequence. Model-weighted estimates put star-formation-rate density above cosmic averages, while metallicity differences are marginal in two systems and consistent with the field in the others.",
    "field": "physics",
    "ai_technology": "PCFNet point-cloud deep-learning protocluster finder",
    "organization": "The University of Tokyo; National Astronomical Observatory of Japan; Tohoku University; Subaru Telescope",
    "source_url": "https://arxiv.org/html/2609.35471",
    "source_type": "paper",
    "publication_date": "2026-11-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stag2110",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Subaru/MOIRCS spectroscopy of four PCFNet-selected candidate regions confirmed four protoclusters at z = 3.316, 3.403, 3.408 and 3.425, across three regions, with galaxy-density peaks more than five times the uniform distribution.",
      "unknown": "This is a small follow-up sample: four of 121 PCFNet candidates were targeted, and the parent catalogue is trained on simulated galaxies. Four protoclusters are confirmed across three of four observed regions; a fourth observed candidate region had no significant redshift peak. One of the confirmed regions had previously been reported as a photometric candidate, so this is a spectroscopic confirmation/characterization result rather than four wholly unprecedented structures. The halo masses are inferred from member stellar masses, not direct lensing measurements. The elevated SFR densities depend on membership-probability weighting and are sensitive to assumed geometry, volume and methodology. Metallicity excesses are only 1.4–2 sigma in two systems; two others are statistically consistent with the field. The authors’ proposed gas-inflow/recycling explanation is interpretive.",
      "ai_role": "PCFNet used broad-band optical photometry and galaxy positions to assign protocluster-membership probabilities and identify candidate regions. The model outputs determined which regions received Subaru/MOIRCS follow-up and prioritized slit targets; their probabilities also informed member selection and the authors’ volume/SFR-density estimator. Spectroscopy then tested those predictions, finding a 40% member detection rate among high-probability galaxies and confirming three of four observed regions.",
      "evidence_urls": [
        "https://arxiv.org/html/2609.35471"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-ai-guided-experiments-uncover-a-new-polymer-polymorph-and-improve-organic-transistor-performance",
    "title": "AI-guided experiments uncover a new polymer polymorph and improve organic transistor performance",
    "summary": "An AI-powered advisor in an autonomous fabrication-and-test platform guided 64 trials of mixed ion–electron conducting polymers for organic electrochemical transistors. The study reports a mobility–volumetric-capacitance figure of merit up to 1,275 F cm⁻¹ V⁻¹ s⁻¹, a 150% improvement over conventional spin coating, identifies larger crystalline lamellar spacing and greater specific surface area as favorable structural factors, and reports a previously unknown polymer polymorph.",
    "field": "materials",
    "ai_technology": "Adaptive AI decision interface and AI advisor integrated with autonomous electronic-material fabrication and testing",
    "organization": "The University of Chicago; Argonne National Laboratory; Lawrence Berkeley National Laboratory; University of Southern Mississippi",
    "source_url": "https://www.nature.com/articles/s44286-025-00318-3",
    "source_type": "paper",
    "publication_date": "2025-12-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44286-025-00318-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An AI-powered advisor in an autonomous fabrication-and-test platform guided 64 trials of mixed ion–electron conducting polymers for organic electrochemical transistors.",
      "unknown": "The primary source describes a single mixed-ion/electron conducting polymer system and an organic electrochemical transistor testbed. The reported 150% figure-of-merit improvement is relative to conventional spin coating in the study. Structural-factor associations come from a small set of ten statistically selected samples; they are not general causal rules. The new polymorph is reported in the primary paper, but the available abstract does not establish broader stability or device lifetime. The platform includes human-AI interaction and laboratory personnel performed synthesis, fabrication and characterization.",
      "ai_role": "The AI advisor monitored experimental progress in real time, analyzed results and collaborated interactively with researchers to adapt the fabrication, testing and analysis steps across a data-limited campaign. The resulting AI/AE workflow guided experimental exploration over 64 autonomous trials; the model did not merely predict properties from a benchmark dataset. Human researchers synthesized polymers, fabricated devices and performed electrical and structural characterization.",
      "evidence_urls": [
        "https://www.nature.com/articles/s44286-025-00318-3"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-unsupervised-learning-separates-rockfall-and-glacier-sliding-precursors-before-the-blatten-collapse",
    "title": "Unsupervised learning separates rockfall and glacier-sliding precursors before the Blatten collapse",
    "summary": "Unsupervised clustering of 20 days of seismic data before the 28 May 2025 Blatten glacier collapse separated rockfall signals from events associated with basal glacier sliding. Sliding-related event counts accelerated approximately exponentially beginning about five days before collapse, with further acceleration about 1.5 days beforehand; the authors interpret this pattern as consistent with nucleation-like slip weakening.",
    "field": "physics",
    "ai_technology": "Deep scattering network, independent component analysis and hierarchical clustering (unsupervised machine learning)",
    "organization": "Université Grenoble Alpes; Institut de Physique du Globe de Paris; University of Tokyo; Service des dangers naturels, Valais",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL121175",
    "source_type": "paper",
    "publication_date": "2026-05-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025GL121175",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Unsupervised clustering of 20 days of seismic data before the 28 May 2025 Blatten glacier collapse separated rockfall signals from events associated with basal glacier sliding.",
      "unknown": "The finding is a retrospective analysis of one event, relying primarily on a single seismic station 5.3 km from the site. The authors describe the physical mechanism as consistent with nucleation and hypothesize that rockfalls contributed to instability; these interpretations require study of other events. The inferred slip-rate proxy has high absolute uncertainty, and the approach is not an operational early-warning system.",
      "ai_role": "The unsupervised representation and clustering explored continuous single-station seismic records without event templates, separating rockfall, glacier-sliding and cultural-noise signatures. This made the evolving sliding-related event population and its acceleration measurable in the 20-day record.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL121175"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-05-machine-learning-reconstructs-deep-time-crustal-thickness-and-links-thickening-episodes-with-porphyry-deposits",
    "title": "Machine learning reconstructs deep-time crustal thickness and links thickening episodes with porphyry deposits",
    "summary": "A machine-learning mohometry model estimated paleo-crustal thickness from 66 geochemical proxies across subduction, collision and intraplate settings, extending chemical reconstruction beyond arc-related rocks. Tested reconstructions for southern Tibet and South China agreed with independently constrained crustal histories (R² = 0.937; RMSE = 4.3 km). The authors report that reconstructed thickening episodes correlate with porphyry-ore formation in ancient orogens.",
    "field": "physics",
    "ai_technology": "Machine-learning mohometry regression model with feature-importance filtering",
    "organization": "Ocean University of China; University of Sydney; University of Tasmania; Laoshan Laboratory",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024JB030404",
    "source_type": "paper",
    "publication_date": "2025-05-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2024JB030404",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning mohometry model estimated paleo-crustal thickness from 66 geochemical proxies across subduction, collision and intraplate settings, extending chemical reconstruction beyond arc-related rocks.",
      "unknown": "The ancient thickness histories and porphyry association are model-derived reconstructions, not direct measurements of past Moho depth or proof that thickening caused ore formation. Validation is reported in southern Tibet and the South China Block; R²/RMSE summarize the model data and do not remove uncertainty from proxy assumptions, tectonic context or the geological record. The study reports a correlation that may inform exploration, not a validated mineral-deposit prediction system.",
      "ai_role": "The regression model learned relationships between whole-rock geochemical measurements and Moho depth from a global database spanning subduction, collision and intraplate tectonic settings. It estimated ancient crust thickness from records that may be missing elements and enabled reconstruction beyond arc magmas; feature-importance filtering highlighted geochemical proxies. Those model-derived reconstructions supported the paper’s comparison of crust-thickening episodes with porphyry-deposit timing.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024JB030404"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-08-mace-guided-screening-yields-three-corundum-high-entropy-oxides-and-a-cation-ordered-oxide",
    "title": "MACE-guided screening yields three corundum high-entropy oxides and a cation-ordered oxide",
    "summary": "MACE-MP-0b2 interatomic-potential calculations screened 462 four- and five-cation trivalent-oxide compositions and selected 16 for synthesis. Experiments produced three phase-pure corundum high-entropy oxides—(Al,Cr,Fe,Ga)₂O₃, (Al,Cr,Fe,Rh,Sc)₂O₃ and (Cr,Fe,Ga,Rh,Sc)₂O₃—and a separate phase-pure cation-ordered (Al,Fe,Ga,Sc)₂O₃ oxide. The ordered compound crystallized in orthorhombic Pba2; its precise cation-site occupancies were not resolved conclusively by powder diffraction.",
    "field": "materials",
    "ai_technology": "MACE-MP-0b2 machine-learning interatomic potential (MLIP)",
    "organization": "University of British Columbia",
    "source_url": "https://arxiv.org/abs/2608.20596",
    "source_type": "paper",
    "publication_date": "2026-08-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "MACE-MP-0b2 interatomic-potential calculations screened 462 four- and five-cation trivalent-oxide compositions and selected 16 for synthesis.",
      "unknown": "This is an unrefereed arXiv preprint. Only three of 16 tested compositions formed phase-pure corundum high-entropy oxides; most attempts produced competing phases, and rhodium-containing samples showed redox instability. Synthesis outcomes depended strongly on route: the three corundum HEOs were phase-pure only with one of solid-state or combustion synthesis. The ordered oxide’s structure is supported by powder diffraction/Rietveld refinement, but exact site preferences could not be conclusively determined from pXRD. The screening supports candidate ranking, not reliable prediction of phase purity or synthesis conditions for arbitrary HEOs.",
      "ai_role": "The MLIP relaxed approximately 1,000-atom disordered supercells and estimated mixing enthalpy and atom-energy-distribution descriptors for the 462-composition search. The authors combined these with a stability metric to rank 16 synthesis candidates. Experimental powder-X-ray diffraction and Rietveld refinement then assessed the predicted materials, linking model-guided selection to the new synthesized compositions.",
      "evidence_urls": [
        "https://arxiv.org/abs/2608.20596"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-llm-guided-experiments-identify-three-phases-unique-to-the-ternary-coalge-system",
    "title": "LLM-guided experiments identify three phases unique to the ternary Co–Al–Ge system",
    "summary": "Researchers used an LLM to plan iterative composition measurements and experimentally mapped the Co–Al–Ge phase diagram at 900 °C. The domain-specific aLLoyM model directed early measurements into compositionally complex interior regions, enabling the earliest discovery of all three novel phases reported to form only in the ternary system; the general-purpose LLM strategy found more phases in fewer cycles. This is an arXiv preprint and the finding is bounded to this experimental phase-diagram study.",
    "field": "materials",
    "ai_technology": "aLLoyM domain-specific large language model and general-purpose LLM experimental planner",
    "organization": "National Institute for Materials Science; Tohoku University; University of Tokyo",
    "source_url": "https://arxiv.org/abs/2604.20304",
    "source_type": "paper",
    "publication_date": "2026-04-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Researchers used an LLM to plan iterative composition measurements and experimentally mapped the Co–Al–Ge phase diagram at 900 °C.",
      "unknown": "This source is an arXiv preprint, not a peer-reviewed journal publication as of the checked record. The source abstract supports the reported ternary phase-diagram result and three phases described as novel and ternary-only, but does not provide their formulas or independent replication evidence in the abstract. The discovery is a bounded materials result, not a claim that every phase in the full ternary diagram has been exhaustively mapped.",
      "ai_role": "LLMs selected compositions for each cycle in a closed loop with high-throughput synthesis and X-ray diffraction phase identification. The original abstract specifically attributes earliest discovery of all three newly reported ternary-only phases to aLLoyM-guided initial measurements; a general-purpose LLM chose a textbook-like measurement strategy that efficiently found a larger number of phases in fewer cycles.",
      "evidence_urls": [
        "https://arxiv.org/abs/2604.20304"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-machine-learning-mapping-finds-broader-preservation-of-pre-glacial-landscapes-across-arctic-canada-and-greenland",
    "title": "Machine-learning mapping finds broader preservation of pre-glacial landscapes across Arctic Canada and Greenland",
    "summary": "A random-forest model using six terrain morphometrics from ArcticDEM classified glacial erosional styles across Arctic Canada and Greenland. The map indicates that regions with little or no glacial erosion, and thus potentially preserved pre-glacial landscape fragments, occur more widely than previous mapping recognized; it also finds distinct morphologies between Greenland and Canada. The authors infer that inherited topography, ice dynamics, latitude and geology help shape these patterns.",
    "field": "physics",
    "ai_technology": "Random-forest machine-learning classifier",
    "organization": "Durham University; Newcastle University",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2025JF008661",
    "source_type": "paper",
    "publication_date": "2026-05-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025JF008661",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A random-forest model using six terrain morphometrics from ArcticDEM classified glacial erosional styles across Arctic Canada and Greenland.",
      "unknown": "The paper reports validation accuracy of about 79.5% (±4.6%) and notes low classification probabilities in some areas, where training coverage or coarse classes may limit confidence. “Little or no glacial erosion” is a model classification interpreted as possible preservation of pre-glacial terrain, not direct dating or field confirmation of every mapped tile. Proposed controls such as ice dynamics and basal hydrology are interpretations, not demonstrated causal mechanisms.",
      "ai_role": "The random forest classified ArcticDEM tiles using six elevation-derived morphometric variables, assigning class probabilities at continental scale and producing model-derived feature-importance estimates. This enabled a quantitative revision of prior qualitative landscape maps and the reported finding that minimally glacially modified terrain is more widespread than previously mapped.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2025JF008661"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-07-transformer-based-sdss-search-identifies-two-previously-unreported-opticaluv-tidal-disruption-events",
    "title": "Transformer-based SDSS search identifies two previously unreported optical–UV tidal disruption events",
    "summary": "A PCA-enhanced Transformer classifier screened 298,195 SDSS DR7 spectra and selected 14 TDE-like cases for secondary review. The authors report two newly identified TDEs: SDSS J124225.39+642919.0, whose GALEX UV source was present by the spectrum date (before 11 February 2002), and SDSS J152459.70+045423.1, whose TDE-H+He spectrum coincided with a Catalina optical outburst in 2007–08. The first is reported as the earliest optical–UV TDE then known; one other model-selected source was a previously reported likely TDE.",
    "field": "physics",
    "ai_technology": "PCA-enhanced Transformer TDE spectrum classifier",
    "organization": "University of Science and Technology of China; Purple Mountain Observatory; Chinese Academy of Sciences",
    "source_url": "https://arxiv.org/abs/2607.11539",
    "source_type": "paper",
    "publication_date": "2026-07-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A PCA-enhanced Transformer classifier screened 298,195 SDSS DR7 spectra and selected 14 TDE-like cases for secondary review.",
      "unknown": "The source is an arXiv preprint whose arXiv record says accepted by The Astrophysical Journal; no journal DOI was supplied in the source checked. The TDE classifications are supported by spectral and contemporaneous archival photometric evidence, not new contemporaneous follow-up observations. The event dates predate 2020, but this record uses the 2026 source-publication date for the since-2020 catalog period. The paper also reports a third previously known likely TDE, which is not counted as a new discovery.",
      "ai_role": "The Transformer classifier screened preprocessed SDSS galaxy spectra and selected TDE-like events for follow-up assessment. From roughly 298,000 spectra it produced 14 cases above the selection threshold; after checking spectral features against archival GALEX or Catalina variability and other diagnostics, the authors reported two previously unreported TDEs and one previously reported likely TDE. Reported test-set precision and recall were 0.88 and 0.99, respectively.",
      "evidence_urls": [
        "https://arxiv.org/abs/2607.11539"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-08-physics-informed-generative-learning-yields-experimentally-verified-single-phase-b2-intermetallic-alloys",
    "title": "Physics-informed generative learning yields experimentally verified single-phase B2 intermetallic alloys",
    "summary": "A conditional variational autoencoder and artificial neural network, using physics-informed descriptors of B2 sublattice ordering, generated and ranked compositions in complex alloy systems. Experimental synthesis and XRD/TEM characterization verified previously unreported single-phase B2 alloys, including three Fe–Co–Ni–Ti–Zr compositions, Co19.9Ni30.5Ti38.1Zr11.5, and Cu4.7Co9.2Ni36.4Ti38.7Zr7.4Hf3.5. The study reports that random-sublattice descriptors separated B2 from multiphase alloys more effectively than conventional random-mixing descriptors in these datasets.",
    "field": "materials",
    "ai_technology": "Physics-informed conditional variational autoencoder (CVAE) and artificial neural network (ANN)",
    "organization": "Central South University; City University of Hong Kong; City University of Hong Kong (Dongguan); Shanghai University",
    "source_url": "https://www.nature.com/articles/s41524-025-01775-3",
    "source_type": "paper",
    "publication_date": "2025-08-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-025-01775-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A conditional variational autoencoder and artificial neural network, using physics-informed descriptors of B2 sublattice ordering, generated and ranked compositions in complex alloy systems.",
      "unknown": "The result is one paper-level experimental discovery program; individual alloys are not counted separately. Experiments validated selected compositions from a small subset of the generated search space, not all model proposals. The authors report several new single-phase B2 compositions and strong structural characterization for named samples; broader property superiority and generalization to arbitrary alloy systems are not inferred here. Synthesis and characterization were performed by researchers.",
      "ai_role": "The ANN learned B2-versus-non-B2 phase likelihoods from historical alloy-composition data using 18 physics-informed descriptors, while the conditional VAE generated candidate compositions. ANN scores filtered generated compositions for experimental validation. The paper reports successful B2 structure confirmation after synthesis, including XRD/SEM and TEM evidence for long-range B2 ordering in tested alloys.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01775-3"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-deep-learning-denoising-reveals-new-short-period-seismic-observations-of-known-ultra-low-velocity-zones",
    "title": "Deep-learning denoising reveals new short-period seismic observations of known ultra-low-velocity zones",
    "summary": "A supervised deep-learning denoising autoencoder exposed previously undocumented short-period S-diffracted wave observations associated with several known ultra-low-velocity zones (ULVZs). It also revealed four Hawaiian-ULVZ S-diff post-cursor records at Japan F-net stations from East Pacific Rise earthquakes, providing ray paths with mean azimuth about 78° from earlier North American paths. These new cross-cutting observations add constraints on ULVZ geometry and frequency/azimuth-dependent waveforms, but the paper does not present a new ULVZ detection or a completed structural inversion.",
    "field": "physics",
    "ai_technology": "Supervised deep-learning denoising autoencoder (DeepSeisDAE)",
    "organization": "University of Münster; James Cook University; Utrecht University; Australian National University; Geological Survey of Denmark and Greenland",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2026JB034067",
    "source_type": "paper",
    "publication_date": "2026-11-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2026JB034067",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A supervised deep-learning denoising autoencoder exposed previously undocumented short-period S-diffracted wave observations associated with several known ultra-low-velocity zones (ULVZs).",
      "unknown": "This paper reports new seismic observations sampling known ULVZs; it does not establish a new ULVZ structure. The Japanese post-cursors are difficult to separate from the main S-diff phases and require future 3D modelling to determine their interpretation and resolve trade-offs. Denoiser amplitude fidelity is a limitation, especially under training/application mismatch; the authors advise caution for single-station denoising. These are observational constraints, not a completed inversion of shape or composition.",
      "ai_role": "The denoiser was trained on synthetic seismic waveforms mixed with real global noise, then used to separate signal from noise in real S-diff record sections. Its output helped researchers identify short-period post-cursor observations not reported in earlier literature, including four events recorded in Japan with nearly perpendicular ray geometry to previous North American observations of the Hawaiian ULVZ.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2026JB034067"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-11-ai-guided-experiments-uncover-new-zeolitic-imidazolate-framework-crystals",
    "title": "AI-guided experiments uncover new zeolitic imidazolate framework crystals",
    "summary": "AIRES combined probabilistic machine-learning predictions with robotic synthesis, optical crystal screening and single-crystal X-ray diffraction to explore zinc zeolitic imidazolate frameworks. The campaign crystallized new structures using 10 of 48 previously untested linkers, doubling the discovery rate of random exploration (700 versus 1,400 experiments) and increasing the single-linker Zn-ZIF library by about one-third. The paper reports structure-resolved products, including eleven previously unreported crystals across its discovery campaign.",
    "field": "materials",
    "ai_technology": "Algorithmic Iterative Reticular Synthesis (AIRES)",
    "organization": "University of California, Berkeley; Monash University; VinUniversity; University of Haifa; Humboldt University of Berlin",
    "source_url": "https://www.nature.com/articles/s44160-025-00939-9",
    "source_type": "paper",
    "publication_date": "2025-11-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44160-025-00939-9",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "AIRES combined probabilistic machine-learning predictions with robotic synthesis, optical crystal screening and single-crystal X-ray diffraction to explore zinc zeolitic imidazolate frameworks.",
      "unknown": "This is one paper-level experimental discovery campaign. The reported efficiency comparison is against a random-exploration baseline within the study, not an independent replication. Structures were identified by single-crystal X-ray diffraction; the paper describes discovery of ZIF crystals but does not establish application performance. The campaign covered a defined single-linker Zn-ZIF chemical and reaction space, not all reticular materials.",
      "ai_role": "AIRES used probabilistic random-forest and Gaussian-process crystallization models to guide experiment selection across linker chemistries and reaction conditions. Its greedy and batch decision rules prioritized experiments predicted to yield distinct new crystals; automated synthesis, computer-vision crystal screening and subsequent single-crystal X-ray diffraction supplied experimental outcomes for model updates and structure determination. The AI-guided campaign found structures for 10 of 48 new linkers, with twice the distinct-crystal discovery rate of random exploration.",
      "evidence_urls": [
        "https://www.nature.com/articles/s44160-025-00939-9"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-llm-guided-alloy-design-yields-self-limiting-oxidation-behavior-at-900-c",
    "title": "LLM-guided alloy design yields self-limiting oxidation behavior at 900 °C",
    "summary": "A knowledge-enhanced AI workflow led to Co22Cr24Fe20Ni25Mo5Mn4, a non-equiatomic high-entropy alloy with experimentally observed single-phase FCC structure and self-limiting oxidation kinetics at 900 °C. After about 50 hours, the study reports a steady-state oxidation-rate constant of 1.41 × 10⁻⁸ mgⁿ cm⁻²ⁿ h⁻¹ (reaction order n = 45.35) and a Cr₂O₃/(Mn,Fe)Cr₂O₄ bilayer oxide scale. The proposed barrier–buffer explanation is the authors’ mechanistic interpretation.",
    "field": "materials",
    "ai_technology": "Physics-guided ML phase classifiers and LLM-assisted alloy-design workflow",
    "organization": "Shanghai University",
    "source_url": "https://www.nature.com/articles/s41524-026-02071-4",
    "source_type": "paper",
    "publication_date": "2026-04-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-026-02071-4",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A knowledge-enhanced AI workflow led to Co22Cr24Fe20Ni25Mo5Mn4, a non-equiatomic high-entropy alloy with experimentally observed single-phase FCC structure and self-limiting oxidation kinetics at 900 °C.",
      "unknown": "The specific oxidation result is experimentally measured for one designed alloy under the paper’s 900 °C test conditions and time interval; it is not evidence of service lifetime or performance in a component. The bilayer-oxide barrier–buffer mechanism is the authors’ interpretation from characterization and kinetic behavior. The two additional non-equiatomic FCC/BCC alloys in the study validate phase predictions and remain bundled with the principal material result. The LLM reasoning-agent model is not identified clearly in the paper; DeepSeek-R1-671B is explicitly named for processing-text parsing.",
      "ai_role": "The authors used random-forest and support-vector classifiers trained on physics-based descriptors to prioritize single-phase alloy compositions. DeepSeek-R1-671B parsed literature processing descriptions into structured features. Separately, their LLM-agent workflow synthesized oxidation literature and mechanism knowledge to propose composition changes, including increasing Cr and adding about 4 at.% Mn; this guided selection of Co22Cr24Fe20Ni25Mo5Mn4 for synthesis and oxidation experiments. The paper does not clearly identify the model used for the reasoning-agent stage, so that identity is left unspecified.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-026-02071-4"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-first-observation-of-single-top-quark-production-with-a-w-and-z-boson-at-cms",
    "title": "First observation of single-top-quark production with a W and Z boson at CMS",
    "summary": "CMS reports the first observation of single-top-quark production in association with a W and Z boson in proton-proton collisions. A Particle Transformer classifier and improved reconstruction separated the rare tWZ signal from major backgrounds in three- and four-lepton events. Combining 13 and 13.6 TeV data (200 fb⁻¹), the measured signal has 5.8σ observed significance (3.5σ expected) and cross sections of 248 ± 52 fb at 13 TeV and 242 ± 77 fb at 13.6 TeV. The result agrees with the Standard Model within 2.3σ.",
    "field": "physics",
    "ai_technology": "Particle Transformer (ParT) multiclass classifier",
    "organization": "CMS Collaboration; CERN Large Hadron Collider",
    "source_url": "https://cms-results.web.cern.ch/cms-results/public-results/publications/TOP-24-009/",
    "source_type": "paper",
    "publication_date": "2026-02-25",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/rk6w-1pcl",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "CMS reports the first observation of single-top-quark production in association with a W and Z boson in proton-proton collisions.",
      "unknown": "This is the first observation of the Standard-Model tWZ production process, not evidence for physics beyond the Standard Model. A 2024 CMS analysis had reported earlier evidence at 3.4σ; the 2026 paper adds 13.6 TeV Run 3 data and improved reconstruction/ML analysis to reach observation. The 5.8σ result combines 13 and 13.6 TeV datasets. Cross sections and signal strength are consistent with the Standard Model within the reported uncertainties.",
      "ai_role": "CMS trained four ParT models, split by three- versus four-lepton signal region and 13 versus 13.6 TeV samples. In the three-lepton region, a four-node classifier distinguished tWZ, ttZ and other backgrounds; in the four-lepton region, a binary classifier separated tWZ from backgrounds. Its classifier-score categories entered the statistical signal extraction. The paper reports a substantial sensitivity gain versus the previous analysis: expected significance on the reused 13 TeV data rose from 1.4σ to 3.0σ, contributing to the first observation when combined with 13.6 TeV data.",
      "evidence_urls": [
        "https://cms-results.web.cern.ch/cms-results/public-results/publications/TOP-24-009/",
        "https://arxiv.org/abs/2510.19080"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-12-omnifold-enables-an-unbinned-24-dimensional-z-jets-cross-section-measurement-by-atlas",
    "title": "OmniFold enables an unbinned 24-dimensional Z+jets cross-section measurement by ATLAS",
    "summary": "Using 139 fb⁻¹ of 13 TeV proton-proton data, ATLAS simultaneously unfolded 24 correlated Z+jets kinematic observables with the machine-learning method OmniFold and released the particle-level result as an unbinned event dataset. The measured total fiducial cross section is 1,808 ± 42 fb; most illustrative differential bins have 3–5% total uncertainty, rising to about 15% in tails. The measurement is more precise than the compared predictions for many distributions, especially Sherpa. The event-level release also supports derived observables, including an infrared-and-collinear-safe τ₂₁ measurement in a specified fiducial region.",
    "field": "physics",
    "ai_technology": "OmniFold iterative machine-learning unfolding",
    "organization": "ATLAS Collaboration; CERN Large Hadron Collider",
    "source_url": "https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.133.261803",
    "source_type": "paper",
    "publication_date": "2024-12-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevLett.133.261803",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using 139 fb⁻¹ of 13 TeV proton-proton data, ATLAS simultaneously unfolded 24 correlated Z+jets kinematic observables with the machine-learning method OmniFold and released the particle-level result as an unbinned event dataset.",
      "unknown": "This is a new high-dimensional precision measurement of known Z+jets production, not a new particle or beyond-Standard-Model discovery. The released measurement is unbinned in 24 measured variables; displayed plots may be binned for presentation. The authors report mostly 3–5% uncertainties in illustrative bins, up to about 15% in tails, with dominant unfolding-prior/hidden-variable uncertainties for some observables. Pseudo-data closure is generally good, with exceptions including a p-value of 0.038 for leading-jet transverse momentum; users are given phase-space and covariance-use guidance.",
      "ai_role": "OmniFold used neural-network classifiers to iteratively reweight simulated events so the detector-level simulation matched collision data, then propagated the learned correction to particle-level events. That multivariate unfolding handled 24 correlated observables simultaneously and produced event weights with uncertainty variations, enabling a high-dimensional unbinned measurement and event-level construction of additional derived observables. The paper validates closure on pseudo-data and reports its uncertainty components.",
      "evidence_urls": [
        "https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.133.261803"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-deep-learning-expands-the-utah-forge-das-microearthquake-catalog-and-resolves-stimulation-structure",
    "title": "Deep learning expands the Utah FORGE DAS microearthquake catalog and resolves stimulation structure",
    "summary": "A ResNet-based DASEventNet model applied to continuous distributed acoustic sensing from the April 2022 Utah FORGE Well 16A (78)-32 stimulation manually verified 5,749 microearthquakes absent from the Silixa STA/LTA catalog; 45 of 5,794 model-positive windows were false detections on manual review. The study reports events down to Mw −1.80 versus −1.14 for STA/LTA and a broader Stage 3 seismicity distribution consistent with fracture geometry seen in an independent deep-borehole geophone catalog. This is a local event-catalog and stimulation-geometry result, not a direct DAS-based event-location measurement.",
    "field": "physics",
    "ai_technology": "DASEventNet (ResNet-50 convolutional neural network)",
    "organization": "Pennsylvania State University; Utah FORGE",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024JB029102",
    "source_type": "paper",
    "publication_date": "2024-09-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2024JB029102",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A ResNet-based DASEventNet model applied to continuous distributed acoustic sensing from the April 2022 Utah FORGE Well 16A (78)-32 stimulation manually verified 5,749 microearthquakes absent from the Silixa STA/LTA catalog; 45 of 5,794 model-positive windows were false detections on manual review.",
      "unknown": "Original JGR: Solid Earth paper reports manual review of model-positive events and comparison with conventional Silixa detections. Model predicts event presence but does not estimate event locations or magnitudes from DAS alone; timing association and location/magnitude comparison use an independent deep-borehole geophone catalog with an assumed 0.2 s matching window. Fiber placement limits direct geometry inference. Detection of cataloged events is not independent proof of every physical interpretation.",
      "ai_role": "Trained on manually checked DAS event and noise windows, the network detected weak microearthquake waveforms in continuous seven-day DAS recordings. It expanded the catalog and enabled a denser spatiotemporal characterization of the hydraulic stimulation seismicity.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024JB029102"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-11-xrd-guided-learning-identifies-li6asse5i-as-a-room-temperature-superionic-conductor",
    "title": "XRD-guided learning identifies Li₆AsSe₅I as a room-temperature superionic conductor",
    "summary": "XRDMatch, a semi-supervised model using X-ray diffraction patterns as its descriptor, screened unlabeled crystal data and identified 38 lithium superionic-conductor predictions. The paper says 32 were supported by prior literature, six were new candidates assessed with ab initio molecular dynamics, and Li₆AsSe₅I was synthesized and experimentally confirmed as a superionic conductor. The source abstract does not establish that all 38 are new or experimentally validated; this record covers the single reported synthesized material.",
    "field": "materials",
    "ai_technology": "XRDMatch semi-supervised learning framework",
    "organization": "East China Normal University; Shanghai Jiao Tong University; Westlake University; Microsoft Research Asia; collaborators",
    "source_url": "https://pubs.rsc.org/en/content/articlelanding/2024/ee/d4ee02970d",
    "source_type": "paper",
    "publication_date": "2024-11-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D4EE02970D",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "XRDMatch, a semi-supervised model using X-ray diffraction patterns as its descriptor, screened unlabeled crystal data and identified 38 lithium superionic-conductor predictions.",
      "unknown": "The publisher abstract says six newly predicted candidates were quantified with ab initio molecular dynamics, but names only Li₆AsSe₅I as synthesized and experimentally confirmed. It reports 32 additional predictions matched recent literature. The full publisher article was inaccessible during the source review (HTTP 403); the annotation is limited to the primary publisher abstract and its article metadata.",
      "ai_role": "Consistency regularization and pseudo-labeling let the model learn from limited labeled conductivity data and unlabeled ICSD X-ray diffraction patterns. It nominated Li₆AsSe₅I for synthesis, where the researchers experimentally confirmed superionic conductivity.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlelanding/2024/ee/d4ee02970d"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-06-phasenet-tf-expands-tonga-earthquake-catalog-and-resolves-the-intermediate-depth-double-seismic-zone",
    "title": "PhaseNet-TF expands Tonga earthquake catalog and resolves the intermediate-depth double seismic zone",
    "summary": "A semi-supervised PhaseNet-TF workflow detected and relocated 13,616 events from a one-year Tonga OBS and land-station record, about 11 times the manually picked reference catalog. The denser catalog delineates the intermediate-depth double seismic zone to about 300 km and indicates that its lower plane is limited to roughly 19°–21°S, rather than continuous along the studied section. The study did not recover a deeper 350–460 km double zone; sparse events and location uncertainties limit inference at those depths.",
    "field": "physics",
    "ai_technology": "PhaseNet-TF deep-learning phase picker with semi-supervised iterative retraining",
    "organization": "Michigan State University; Stanford University; University of California, Berkeley",
    "source_url": "https://academic.oup.com/gji/article/238/2/1073/7689220",
    "source_type": "paper",
    "publication_date": "2024-06-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/gji/ggae200",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A semi-supervised PhaseNet-TF workflow detected and relocated 13,616 events from a one-year Tonga OBS and land-station record, about 11 times the manually picked reference catalog.",
      "unknown": "The catalog is based on one year of OBS and land data. About half of events are outside the array and have larger hypocenter uncertainty from azimuthal gaps; velocity-model choice adds absolute-location uncertainty. The authors say the absent 350–460 km double zone may reflect insufficient events, not definitive absence. Stress and hydration explanations are interpretations, and some rare events above the slab may be false.",
      "ai_role": "The time-frequency neural picker detected P/S arrivals in noisy ocean-bottom seismometer data; iterative semi-supervised retraining expanded picks, while phase association and double-difference relocation filtered unreliable events. This expanded catalog enabled the localized double-seismic-zone inference.",
      "evidence_urls": [
        "https://academic.oup.com/gji/article/238/2/1073/7689220"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-lhcb-observes-the-doubly-charmed-cc-baryon-with-an-8-7-global-significance",
    "title": "LHCb observes the doubly charmed Ωcc⁺ baryon with an 8.7σ global significance",
    "summary": "Using 6.3 fb⁻¹ of 13.6 TeV proton-proton collisions recorded in 2024, LHCb observed an Ωc⁰π⁺ mass peak with 8.7σ global significance and identified it as the weakly decaying doubly charmed Ωcc⁺ baryon. The reported mass is 3725.9 ± 1.0 (stat) ± 0.2 (syst) ± 0.4 (lifetime) ± 0.6 (external) MeV/c². A boosted decision tree suppressed random-track combinatorial background; the result is a preprint submitted to Physical Review Letters.",
    "field": "physics",
    "ai_technology": "Boosted decision tree classifier (TMVA toolkit)",
    "organization": "LHCb Collaboration; CERN",
    "source_url": "https://arxiv.org/abs/2609.21921",
    "source_type": "paper",
    "publication_date": "2026-09-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Using 6.3 fb⁻¹ of 13.6 TeV proton-proton collisions recorded in 2024, LHCb observed an Ωc⁰π⁺ mass peak with 8.7σ global significance and identified it as the weakly decaying doubly charmed Ωcc⁺ baryon.",
      "unknown": "The source is an LHCb collaboration preprint submitted to PRL, not yet a peer-reviewed journal publication as of its arXiv version. The observed peak has 8.7σ global significance; the mass uncertainty includes an explicit lifetime-dependent selection-bias term because the Ωcc⁺ lifetime is unknown. BDT use is an analysis component, not the sole basis of the observation.",
      "ai_role": "The BDT was trained on simulated Ωcc⁺ signal decays and wrong-charge Ωc⁰π⁻ combinations from data and used to suppress the dominant random-track combinatorial background in the offline candidate selection. The background rejection supported the mass-spectrum analysis yielding the first observation.",
      "evidence_urls": [
        "https://arxiv.org/abs/2609.21921"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-lhcb-observes-cp-violation-for-the-first-time-in-baryon-decays",
    "title": "LHCb observes CP violation for the first time in baryon decays",
    "summary": "Using 9 fb⁻¹ of LHCb Run 1 and Run 2 data, the collaboration measured a phase-space-averaged CP asymmetry of (2.45 ± 0.46 ± 0.10)% in Λb⁰ → pK⁻π⁺π⁻ and its conjugate, a 5.2σ observation. In the resonance-dominated R(pπ⁺π⁻)K⁻ region, the asymmetry was (5.4 ± 0.9 ± 0.1)%, 6.0σ from zero. Machine-learning BDT selection suppressed combinatorial background; neural-network PID variables mitigated particle misidentification.",
    "field": "physics",
    "ai_technology": "Boosted decision tree event classifier; neural-network particle-identification variables",
    "organization": "LHCb Collaboration; CERN",
    "source_url": "https://www.nature.com/articles/s41586-025-09119-3",
    "source_type": "paper",
    "publication_date": "2025-07-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09119-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using 9 fb⁻¹ of LHCb Run 1 and Run 2 data, the collaboration measured a phase-space-averaged CP asymmetry of (2.45 ± 0.46 ± 0.10)% in Λb⁰ → pK⁻π⁺π⁻ and its conjugate, a 5.2σ observation.",
      "unknown": "This is the first reported baryon-decay CP-violation observation, with significance 5.2σ for the phase-space-integrated result and 6.0σ in the specified resonance region. Reported asymmetries average over phase space; the paper says a full amplitude-structure investigation remains future work. Small asymmetries may cancel across angular-momentum amplitudes. The measured CP asymmetry is distinct from raw yield asymmetry; production/detection nuisance asymmetries are subtracted using a control channel.",
      "ai_role": "The BDT used decay topology, displacement and kinematics to suppress random combinations in the four-track signal selection. Neural-network PID variables helped mitigate misidentified final-state particle backgrounds. These tools supported signal purity for the baryon–antibaryon asymmetry measurement.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-025-09119-3"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-dense-array-seismicity-analysis-infers-a-buried-fault-beneath-the-three-gorges-reservoir-area",
    "title": "Dense-array seismicity analysis infers a buried fault beneath the Three Gorges Reservoir area",
    "summary": "A two-phase dense-array catalog processed with the LOC-FLOW machine-learning workflow revealed microseismicity in three zones and shallow (0–6 km) and deep (6–15 km) layers near the northern Xiannvshan Fault Zone. A steeply dipping seismic band is interpreted as a previously hidden western extension of the Tianyangping Fault; the authors also propose fluid-permeable connections between depth layers. The enhanced catalog detects over an order of magnitude more events than the regional network catalog for the same period.",
    "field": "physics",
    "ai_technology": "LOC-FLOW machine-learning seismic catalog workflow",
    "organization": "Wuhan University of Technology; Institute of Seismology, China Earthquake Administration; collaborators",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S1674984726001278",
    "source_type": "paper",
    "publication_date": "2026-10-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.geog.2026.08.002",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A two-phase dense-array catalog processed with the LOC-FLOW machine-learning workflow revealed microseismicity in three zones and shallow (0–6 km) and deep (6–15 km) layers near the northern Xiannvshan Fault Zone.",
      "unknown": "Verified from the original publisher abstract and publication metadata; full text was inaccessible in this review. The steep band is interpreted as an inferred buried fault, not independently imaged or proven fault geometry. The proposed fluid-permeable connection and dual-mechanism reservoir-modulated seismicity are hypotheses. The catalog increase reflects a dense temporary array plus machine-learning workflow, compared with the regional network catalog.",
      "ai_role": "The study used LOC-FLOW with dense-array observations to detect, locate and relocate microearthquakes for a higher-resolution catalog. The catalog made the three activity zones, depth layers and steep seismic band visible for the authors’ structural interpretation.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S1674984726001278"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-07-cms-makes-the-first-hadron-collider-observation-of-boosted-zcc-decay",
    "title": "CMS makes the first hadron-collider observation of boosted Z→cc̄ decay",
    "summary": "In 138 fb⁻¹ of 13 TeV proton-proton data, CMS observed boosted Z→cc̄ production in association with jets for the first time at a hadron collider. The measured signal strength relative to the Standard Model prediction is 1.00 with combined uncertainty +0.19/−0.17. A deep neural network (DeepDoubleX) identified large-radius jets from charm-quark pairs; the paper also sets a 95% confidence-level upper limit of 47 times the Standard Model prediction on boosted H→cc̄ production.",
    "field": "physics",
    "ai_technology": "DeepDoubleX deep neural network for double-charm jet tagging",
    "organization": "CMS Collaboration; CERN",
    "source_url": "https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.131.041801",
    "source_type": "paper",
    "publication_date": "2023-07-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevLett.131.041801",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In 138 fb⁻¹ of 13 TeV proton-proton data, CMS observed boosted Z→cc̄ production in association with jets for the first time at a hadron collider.",
      "unknown": "The first-observation claim is specifically for Z→cc̄ in association with jets at high transverse momentum at a hadron collider; it is not the first observation of Z decay generally. The neural tagger is an important analysis tool, while the observation derives from the full CMS event selection and likelihood fit. In the same paper, boosted H→cc̄ was not observed; only an upper limit is reported.",
      "ai_role": "The neural network processed secondary-vertex and particle-flow candidate features to classify boosted two-prong jets as charm versus light flavor and charm versus bottom. Its charm tagging separated the Z→cc̄ signal from dominant multijet backgrounds and supported the first observation in the boosted high-transverse-momentum regime.",
      "evidence_urls": [
        "https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.131.041801"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-t-sne-selected-lmc-radio-source-receives-a-spectroscopic-agn-classification-at-z-2-80",
    "title": "t-SNE-selected LMC radio source receives a spectroscopic AGN classification at z = 2.80",
    "summary": "SALT spectroscopy of the previously unclassified source LMC tSNE b8 (R.A. 06:22:41.86, Dec. −68:59:21.9) identified it as an active galactic nucleus at redshift z = 2.80. The source was selected from the blazar-like sequence in a t-SNE map combining VISTA near-infrared and ASKAP radio survey data with multi-wavelength photometry. This is one specific spectroscopic classification from the paper’s broader follow-up sample.",
    "field": "physics",
    "ai_technology": "t-distributed stochastic neighbour embedding (t-SNE) applied to multi-wavelength source photometry",
    "organization": "Keele University; Southern African Large Telescope collaboration; collaborators",
    "source_url": "https://arxiv.org/abs/2611.15880",
    "source_type": "paper",
    "publication_date": "2026-11-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "SALT spectroscopy of the previously unclassified source LMC tSNE b8 (R.A.",
      "unknown": "This is an arXiv preprint. The selected source was classified as an AGN at z=2.80 from SALT spectroscopy, but the paper’s broader 18-target blazar-sequence follow-up also contains Magellanic objects and two low-S/N unconfirmed sources; those are not counted as confirmed discoveries. The claim here is limited to LMC tSNE b8. The paper calls the selection sequence blazar-like; that label does not establish this specific source as a blazar.",
      "ai_role": "t-SNE mapped 6,652 Large Magellanic Cloud and 1,356 Small Magellanic Cloud sources using optical-to-radio features. The unsupervised map exposed a blazar-like sequence from which LMC tSNE b8 was selected for SALT follow-up; spectroscopy then established its AGN classification and redshift.",
      "evidence_urls": [
        "https://arxiv.org/abs/2611.15880"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-lhcb-makes-the-first-high-significance-observation-of-the-rare-p-decay",
    "title": "LHCb makes the first high-significance observation of the rare Σ⁺ → pμ⁺μ⁻ decay",
    "summary": "Using 5.4 fb⁻¹ of 13 TeV proton-proton collision data collected in 2016–2018, LHCb observed 237 ± 16 Σ⁺ → pμ⁺μ⁻ decays and measured a branching fraction of (1.08 ± 0.17) × 10⁻⁸. The dimuon spectrum showed no evidence of resonant structures, and the results are compatible with Standard Model expectations. This is the first high-significance observation; earlier HyperCP and LHCb Run 1 results were evidence only.",
    "field": "physics",
    "ai_technology": "LHCb boosted decision tree (BDT) classifier",
    "organization": "LHCb Collaboration; CERN",
    "source_url": "https://journals.aps.org/prl/abstract/10.1103/r3v2-kmmp",
    "source_type": "paper",
    "publication_date": "2025-07-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2504.06096",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using 5.4 fb⁻¹ of 13 TeV proton-proton collision data collected in 2016–2018, LHCb observed 237 ± 16 Σ⁺ → pμ⁺μ⁻ decays and measured a branching fraction of (1.08 ± 0.17) × 10⁻⁸.",
      "unknown": "The paper reports an observation of the nonresonant rare decay, not evidence for the hypothetical intermediate X⁰ suggested by the earlier HyperCP candidates; no resonant dimuon structure is found. The measured branching fraction is compatible with Standard Model expectations. AI supported event selection and background reduction but is not the sole basis of the result.",
      "ai_role": "The BDT combined kinematic and geometric variables, trained on simulated signal and data sidebands representing combinatorial background, and was used in the final candidate selection. Its output threshold was optimized together with particle-identification and Λ-veto criteria to maximize expected significance. This classifier contributed to background suppression; it was one component of the analysis, alongside upgraded trigger selections, detector reconstruction, and the statistical fit.",
      "evidence_urls": [
        "https://journals.aps.org/prl/abstract/10.1103/r3v2-kmmp"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-04-spectroscopy-validates-machine-selected-populations-of-obscured-active-galaxies",
    "title": "Spectroscopy validates machine-selected populations of obscured active galaxies",
    "summary": "Using a sample selected by unsupervised machine learning on optical and mid-infrared photometry, the team obtained 178 MMT spectra and found the brighter class was 85% Type I active galactic nuclei (AGN), with <3% non-AGN contamination, while a fainter class was 65% Type II AGN with 15% contamination, mainly star-forming galaxies. The follow-up showed that the machine-selected classes recover redder and more obscured AGN populations than typical optical spectroscopic samples.",
    "field": "physics",
    "ai_technology": "Unsupervised UMAP and DBSCAN photometric AGN selection pipeline",
    "organization": "University of Arizona; Princeton University; MMT Observatory",
    "source_url": "https://iopscience.iop.org/article/10.3847/1538-3881/ad28b4",
    "source_type": "paper",
    "publication_date": "2024-04-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2402.05169",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using a sample selected by unsupervised machine learning on optical and mid-infrared photometry, the team obtained 178 MMT spectra and found the brighter class was 85% Type I active galactic nuclei (AGN), with <3% non-AGN contamination, while a fainter class was 65% Type II AGN with 15% contamination, mainly star-forming galaxies.",
      "unknown": "This is a follow-up validation of two machine-selected candidate classes, not a claim that every cataloged object is confirmed. Spectroscopy recovered redshifts for 139 of 178 spectra; the fainter Type II class was limited to the brightest 75% of candidates and had 15% contamination, primarily from star-forming galaxies. The unsupervised pipeline is described in related work listed as in preparation in this paper; findings here are bounded to the spectroscopically analyzed sample.",
      "ai_role": "The selection pipeline applied UMAP dimensionality reduction and DBSCAN clustering to optical and mid-infrared colors and morphology, with matched spectroscopy and X-ray information used to identify AGN candidate classes. Those classes defined the targets for Hectospec follow-up; the paper then used optical spectra, emission-line diagnostics, and spectral stacking to test their AGN content. The AI selected the candidate populations, while spectroscopic diagnostics established the reported classifications.",
      "evidence_urls": [
        "https://iopscience.iop.org/article/10.3847/1538-3881/ad28b4"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-phasenet-catalog-reveals-a-2023-injection-linked-earthquake-cluster-in-the-raton-basin",
    "title": "PhaseNet catalog reveals a 2023 injection-linked earthquake cluster in the Raton Basin",
    "summary": "Applying PhaseNet-based phase picking to eight years of recordings from eight broadband stations, the study built a roughly 96,000-event Raton Basin catalog (local magnitudes −1 to 4.3) and resolved a previously unreported 2023 cluster with a Mw 4.3 event. Nearby wells had elevated injection volumes for about a year before the cluster; analysis of relocated events and 90 full centroid moment tensors also supports basement-rooted normal faults and possible elevated pore pressure. The authors interpret injection as a likely influence, not a demonstrated sole cause.",
    "field": "physics",
    "ai_technology": "PhaseNet deep neural-network seismic phase picker in the LOC-FLOW workflow",
    "organization": "Southern University of Science and Technology; University of New Mexico; U.S. Geological Survey",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL114675",
    "source_type": "paper",
    "publication_date": "2025-07-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025GL114675",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Applying PhaseNet-based phase picking to eight years of recordings from eight broadband stations, the study built a roughly 96,000-event Raton Basin catalog (local magnitudes −1 to 4.3) and resolved a previously unreported 2023 cluster with a Mw 4.3 event.",
      "unknown": "The authors report an association between nearby injection and seismicity and infer elevated pore pressure from positive isotropic components; this does not establish injection as the sole cause. They note possible residual isotropic-component artifacts from velocity heterogeneity or noise and explicitly limit interpretation of CLVD components. The paper gives slightly inconsistent catalog totals: 95,993 in the abstract and most text, but 95,933 in one discussion passage; this summary therefore says roughly 96,000. The newly resolved cluster is distinct from earlier Raton Basin seismicity studies cited by the authors.",
      "ai_role": "PhaseNet picked P and S arrivals directly from continuous waveforms at eight broadband stations, including four stations not telemetered for routine cataloging. These picks enabled event association, relocation, and the expanded 2016–2024 catalog used to identify and analyze the 2023 cluster. PhaseNet enabled the denser earthquake observations; the injection comparison and full-CMT mechanism analysis were subsequent statistical and geophysical analyses.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL114675"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-01-graph-neural-network-enables-first-observation-of-c-ne-e-decay",
    "title": "Graph neural network enables first observation of Λc⁺ → ne⁺νe decay",
    "summary": "Using 4.5 fb⁻¹ of BESIII electron–positron collision data, the collaboration reports the first observation of the Cabibbo-suppressed Λc⁺ → ne⁺νe decay at greater than 10σ significance. It measures a branching fraction of (3.57 ± 0.34(stat) ± 0.14(syst)) × 10⁻³ and extracts |Vcd| = 0.208 ± 0.011(exp) ± 0.007(LQCD) ± 0.001(lifetime) for the first time from a charmed-baryon decay.",
    "field": "physics",
    "ai_technology": "Graph neural network (GNN) trained on BESIII electromagnetic-calorimeter shower patterns",
    "organization": "BESIII Collaboration",
    "source_url": "https://www.nature.com/articles/s41467-024-55042-y",
    "source_type": "paper",
    "publication_date": "2025-01-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2410.13515",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using 4.5 fb⁻¹ of BESIII electron–positron collision data, the collaboration reports the first observation of the Cabibbo-suppressed Λc⁺ → ne⁺νe decay at greater than 10σ significance.",
      "unknown": "The observation is an experimental result from the BESIII data sample; the branching-fraction and |Vcd| uncertainties include statistical, systematic, lattice-QCD and lifetime components as reported. The paper studies data-versus-simulation domain shift and network uncertainty; residual modeling differences contribute to systematic uncertainty. The |Vcd| extraction depends on lattice-QCD input.",
      "ai_role": "The GNN separated neutron signal showers from the dominant Λc⁺ → Λe⁺νe background, whose neutron-plus-π⁰ calorimeter patterns are difficult to distinguish with conventional selection or boosted trees. The authors trained and calibrated it using large J/ψ control samples, then fitted its output distributions to extract the signal yield.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-55042-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-machine-learning-earthquake-catalog-reveals-scale-dependent-fault-controls-at-gonghe-geothermal-site",
    "title": "Machine-learning earthquake catalog reveals scale-dependent fault controls at Gonghe geothermal site",
    "summary": "A machine-learning workflow catalogued 7,346 microearthquakes (ML −2.0 to 2.18) recorded during 2021 stimulation and circulation tests at China’s Gonghe enhanced-geothermal-system site. The catalog resolves a ring-shaped distribution with four clusters and varying faulting styles. The authors infer that site-scale seismicity reactivated mapped, pre-existing natural faults; at borehole scale, the principal seismogenic zone coincides with strong SHmax rotation. Stress modeling estimates a low fault-friction coefficient near 0.23 (95% CI 0.19–0.48) and the low-to-moderate fluid overpressure needed for slip.",
    "field": "physics",
    "ai_technology": "Machine-learning earthquake-detection and cataloging workflow (model details are not stated in the publisher abstract)",
    "organization": "Southern University of Science and Technology; Institute of Geomechanics, Chinese Academy of Geological Sciences; ETH Zürich",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JB033158",
    "source_type": "paper",
    "publication_date": "2026-05-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025JB033158",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning workflow catalogued 7,346 microearthquakes (ML −2.0 to 2.18) recorded during 2021 stimulation and circulation tests at China’s Gonghe enhanced-geothermal-system site.",
      "unknown": "The source checked here is the original publisher abstract and plain-language summary. The catalog is observed, while fault reactivation, chlorite-rich gouge consistency, stress controls and required fluid overpressure are the authors’ interpretations/model estimates, not direct causal proof. No claim is made that the AI model alone establishes these mechanisms.",
      "ai_role": "The authors state that machine learning was used to detect earthquakes and compile the high-precision 7,346-event catalog; the resolved spatial pattern supplies the seismicity observations used in their scale-dependent fault-structure interpretation and stress analysis.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JB033158"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-03-particlenet-enables-first-hadron-collider-w-boson-mass-measurement-from-hadronic-decays",
    "title": "ParticleNet enables first hadron-collider W-boson mass measurement from hadronic decays",
    "summary": "Using 138 fb⁻¹ of 13 TeV proton–proton data, CMS reports the first measurement of the W-boson mass in an all-jets final state at a hadron collider: 80.83 ± 0.55 GeV. The result comes from unfolded groomed jet-mass distributions for boosted hadronic W+jets events. The mass-decorrelated ParticleNet graph neural network tagger provides the main result by selecting the W-jet sample and suppressing quark/gluon multijet background; a non-neural N₂ tagger is used as a cross-check.",
    "field": "physics",
    "ai_technology": "ParticleNet mass-decorrelated graph neural network jet tagger",
    "organization": "CMS Collaboration",
    "source_url": "https://cms-results.web.cern.ch/cms-results/public-results/publications/SMP-24-012/index.html",
    "source_type": "paper",
    "publication_date": "2026-03-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2603.19963",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using 138 fb⁻¹ of 13 TeV proton–proton data, CMS reports the first measurement of the W-boson mass in an all-jets final state at a hadron collider: 80.83 ± 0.55 GeV.",
      "unknown": "This is an all-jets boosted-W extraction with 0.55 GeV total uncertainty, much less precise than the CMS leptonic-channel W-mass result; the paper frames it as a first step toward future all-jets precision measurements and as a test of jet-substructure modeling. The earlier preliminary CMS result of 80.77 ± 0.57 GeV was superseded by this same paper’s published 80.83 ± 0.55 GeV result and is not counted separately. Jet tagging, QCD background estimation, unfolding, hadronization and jet-energy modeling contribute analysis dependence and uncertainties.",
      "ai_role": "The analysis uses ParticleNet as its main W-jet tagger. Its graph neural network treats jet constituents as an unordered particle cloud and classifies W→qq′ jets against QCD multijet backgrounds; its mass-decorrelated output defines the signal/control regions used for the primary measurement. The N₂ substructure observable is a separate cross-check.",
      "evidence_urls": [
        "https://cms-results.web.cern.ch/cms-results/public-results/publications/SMP-24-012/index.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-06-atlas-graph-neural-network-enables-first-observation-of-four-top-quark-production",
    "title": "ATLAS graph neural network enables first observation of four-top-quark production",
    "summary": "Using 140 fb⁻¹ of 13 TeV proton–proton collisions, ATLAS observed simultaneous production of four top quarks in same-sign dilepton and multilepton events, with 6.1σ observed (4.3σ expected) significance. The measured cross-section is 22.5⁺⁶·⁶₋₅·₅ fb, consistent with the Standard Model prediction within 1.8σ. A graph neural network separated the four-top signal from dominant backgrounds in the signal extraction.",
    "field": "physics",
    "ai_technology": "Event-level graph neural network built with graph_nets/TensorFlow",
    "organization": "ATLAS Collaboration; CERN",
    "source_url": "https://link.springer.com/article/10.1140/epjc/s10052-023-11573-0",
    "source_type": "paper",
    "publication_date": "2023-06-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2303.15061",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using 140 fb⁻¹ of 13 TeV proton–proton collisions, ATLAS observed simultaneous production of four top quarks in same-sign dilepton and multilepton events, with 6.1σ observed (4.3σ expected) significance.",
      "unknown": "This is an ATLAS observation from its own 140 fb⁻¹ analysis after earlier ATLAS/CMS evidence; it is one paper-level observation, not a separate count for CMS or later combinations. The measured cross-section uncertainty is large and its central value is 1.8σ above the SM prediction, so this is not evidence of a significant deviation from the Standard Model. Background control regions and systematic uncertainties are part of the analysis.",
      "ai_role": "The GNN discriminant classified events using a fully connected graph of reconstructed jets, leptons and missing transverse momentum, with node, edge and global features. It separated four-top signal from background and supplied the discriminant used in the signal extraction and cross-section fit.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1140/epjc/s10052-023-11573-0"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-08-atlas-boosted-decision-trees-enable-first-observation-of-triple-w-boson-production",
    "title": "ATLAS boosted decision trees enable first observation of triple-W boson production",
    "summary": "In 139 fb⁻¹ of 13 TeV LHC data, ATLAS observed simultaneous WWW production at 8.0σ (5.4σ expected) and measured an inclusive cross-section of 820 ± 100(stat) ± 80(syst) fb. The analysis selected same-sign dilepton-plus-jets and trilepton final states. Two boosted decision trees—one per channel—improved signal/background separation and supplied distributions for the fit. The measured rate is about 2.6σ above the Standard Model prediction, not a significant discrepancy.",
    "field": "physics",
    "ai_technology": "Two boosted decision tree classifiers for the dilepton and trilepton channels",
    "organization": "ATLAS Collaboration; CERN",
    "source_url": "https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.129.061803",
    "source_type": "paper",
    "publication_date": "2022-08-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2201.13045",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In 139 fb⁻¹ of 13 TeV LHC data, ATLAS observed simultaneous WWW production at 8.0σ (5.4σ expected) and measured an inclusive cross-section of 820 ± 100(stat) ± 80(syst) fb.",
      "unknown": "The final peer-reviewed paper reports 8.0σ (5.4σ expected), superseding the earlier 2021 conference-note/press account that quoted 8.2σ and 850 fb. CMS had previously observed combined inclusive production of three massive gauge bosons; this ATLAS paper is the first observation specifically of WWW production. The measured cross-section is about 2.6σ above the theory prediction, below a significant deviation, with data-driven background and systematic uncertainties included.",
      "ai_role": "The BDTs were trained separately for the two- and three-lepton channels, using 12 and 11 well-modelled event variables respectively, to discriminate WWW signal from background. Their output distributions entered the likelihood fit; the paper says the improved separation helped enable the observation.",
      "evidence_urls": [
        "https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.129.061803"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-deep-learning-lens-search-leads-to-first-confirmed-galaxy-scale-lensed-superluminous-supernova",
    "title": "Deep-learning lens search leads to first confirmed galaxy-scale lensed superluminous supernova",
    "summary": "The HOLISMOKES search identified the ZTF transient SN 2025wny at the position of the previously selected lens candidate PS1 J0716+3821. Follow-up imaging resolved multiple supernova images and spectroscopy classified it as a Type I superluminous supernova at z=2.008 ± 0.001; the second deflector galaxy was measured spectroscopically at z=0.375 ± 0.001. This is the first confirmed galaxy-scale strongly lensed supernova with expected time delays of days to weeks, offering a new system for time-delay cosmography.",
    "field": "physics",
    "ai_technology": "Deep-learning Pan-STARRS 3π lens-candidate classification and transient-alert cross-matching pipeline",
    "organization": "Technical University of Munich; Max Planck Institute for Astrophysics; Technical University of Denmark; HOLISMOKES Collaboration",
    "source_url": "https://www.aanda.org/articles/aa/full_html/2026/06/aa57847-25/aa57847-25.html",
    "source_type": "paper",
    "publication_date": "2026-06-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2510.21694",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The HOLISMOKES search identified the ZTF transient SN 2025wny at the position of the previously selected lens candidate PS1 J0716+3821.",
      "unknown": "The redshift, SN type and lensing are based on follow-up observations; time delays are predicted from image separations and were still awaiting high-cadence measurement in this paper. The authors note that the SN’s brightness and resolvability introduce selection effects. An independent paper analyzed the same transient; the discovery is counted once here, not as separate detections by each follow-up team.",
      "ai_role": "The discovery followed a workflow that first classified all 3 billion Pan-STARRS 3π sources with deep learning to select high-quality strong-lens candidates, then continuously cross-matched those positions to ZTF alerts. The team matched the 2025 transient to the preselected lens candidate and triggered imaging and spectroscopic follow-up, which established the lensed SLSN result.",
      "evidence_urls": [
        "https://www.aanda.org/articles/aa/full_html/2026/06/aa57847-25/aa57847-25.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-atlas-graph-neural-network-enables-first-lhc-observation-of-single-top-s-channel-production",
    "title": "ATLAS graph neural network enables first LHC observation of single-top s-channel production",
    "summary": "Using 140 fb⁻¹ of 13 TeV proton–proton collisions, ATLAS reports the first observation of single-top-quark s-channel production at the LHC, with 5.1σ observed (5.2σ expected) significance. The measured cross-section is 10.8 ± 1.5(stat) ± 1.6(syst) pb, consistent with the Standard Model. A graph neural network discriminant separates the signal from top-pair, W+jets and t-channel single-top backgrounds; the analysis also uses broadened event acceptance, improved b tagging and refined modeling.",
    "field": "physics",
    "ai_technology": "Event-level graph neural network; ATLAS GN2 graph-transformer b-jet tagger",
    "organization": "ATLAS Collaboration; CERN",
    "source_url": "https://atlas.cern/Updates/Briefing/Run2-Top-s-Channel",
    "source_type": "press release",
    "publication_date": "2026-10-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2610.11465",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Using 140 fb⁻¹ of 13 TeV proton–proton collisions, ATLAS reports the first observation of single-top-quark s-channel production at the LHC, with 5.1σ observed (5.2σ expected) significance.",
      "unknown": "This is an ATLAS preprint submitted to PRL as of the checked date; prior ATLAS analyses had reported evidence, and the Tevatron had already observed the process in a combined CDF+DØ result. The new claim is first observation at the LHC and by one experiment. The cross-section has about 20% total uncertainty and agrees with the Standard Model; the network is one of several improvements contributing to sensitivity, not the sole cause.",
      "ai_role": "The GNN uses event kinematics to distinguish s-channel signal from major backgrounds, with its score distribution used in the likelihood fit. GN2 improves b-jet identification; the paper attributes increased sensitivity jointly to the GNN, broader acceptance, improved b tagging, calibrations and theory modeling.",
      "evidence_urls": [
        "https://atlas.cern/Updates/Briefing/Run2-Top-s-Channel"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-boosted-decision-tree-enables-first-observation-of-two-simultaneously-longitudinally-polarized-z-bosons",
    "title": "Boosted decision tree enables first observation of two simultaneously longitudinally polarized Z bosons",
    "summary": "Using 164 fb⁻¹ of 13.6 TeV proton–proton data and a statistical combination with the Run 2 measurement, ATLAS reports the first observation of ZZ production with both Z bosons longitudinally polarized, at 6.5σ observed (5.6σ expected). The combined signal strength is 1.22 ± 0.21; the Run 3 measurement also reports polarization fractions across the fiducial phase space and two four-lepton invariant-mass ranges, consistent with Standard Model predictions.",
    "field": "physics",
    "ai_technology": "Boosted decision tree (BDT) classifier",
    "organization": "ATLAS Collaboration; CERN",
    "source_url": "https://atlas.cern/Updates/Briefing/Polarised-Weak-Bosons",
    "source_type": "press release",
    "publication_date": "2026-08-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2607.12879",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Using 164 fb⁻¹ of 13.6 TeV proton–proton data and a statistical combination with the Run 2 measurement, ATLAS reports the first observation of ZZ production with both Z bosons longitudinally polarized, at 6.5σ observed (5.6σ expected).",
      "unknown": "The ATLAS paper was submitted to PRL as a preprint on 2026-07-14; the cited official ATLAS briefing is dated 2026-08-04. The 6.5σ first observation combines the Run 3 analysis with the prior Run 2 measurement, whose standalone significance was 4.3σ. The Run 3 analysis alone reports 5.1σ. The measured fractions agree with the Standard Model; the statistical uncertainty dominates and polarization-template modeling is the leading systematic uncertainty. This is an experimental observation from ATLAS data, not evidence of physics beyond the Standard Model.",
      "ai_role": "The BDT is trained on polarization-sensitive lepton angular observables to separate longitudinal-longitudinal ZZ events from transverse or mixed-polarization events. Its score distribution is the fitted observable used to extract the signal strength and significance.",
      "evidence_urls": [
        "https://atlas.cern/Updates/Briefing/Polarised-Weak-Bosons"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-02-ai-built-baihetan-reservoir-catalog-reveals-fault-controlled-and-depth-dependent-seismicity",
    "title": "AI-built Baihetan reservoir catalog reveals fault-controlled and depth-dependent seismicity",
    "summary": "LOC-FLOW analysis of 55 stations and continuous waveforms from January 2021 to July 2022 identified 53,838 events around China’s Baihetan Reservoir, about four times the manual catalog, with reported relative location errors of tens of metres and recall of about 99% of manual events. The authors infer that post-impoundment seismicity is concentrated along regional structures and fractures, differs across the Jinsha River, and includes migration away from the inundation zone and with depth. They report depth-dependent b-value and water-depth/event-rate patterns consistent with different shallow and deep controls.",
    "field": "physics",
    "ai_technology": "LOC-FLOW automated earthquake-detection and location workflow using PhaseNet deep-learning phase picking",
    "organization": "China Earthquake Administration; Institute of Geophysics; Institute of Earthquake Forecasting",
    "source_url": "https://www.sciencedirect.com/science/article/abs/pii/S0040195124004037",
    "source_type": "paper",
    "publication_date": "2025-02-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.tecto.2024.230601",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "LOC-FLOW analysis of 55 stations and continuous waveforms from January 2021 to July 2022 identified 53,838 events around China’s Baihetan Reservoir, about four times the manual catalog, with reported relative location errors of tens of metres and recall of about 99% of manual events.",
      "unknown": "The catalog covers one reservoir and January 2021–July 2022. The reported location errors are relative errors, not absolute location accuracy. Spatial distribution and migration are observations from the catalog; proposed tectonic, lithologic, fluid-diffusion, poroelastic, water-load, and pore-pressure explanations are author interpretations and are not uniquely established by this study. The paper discusses the need to assess whether the larger automated catalog contains only genuine events. Summary claims are limited to the published abstract and article metadata checked here.",
      "ai_role": "The AI phase picker enabled automatic processing of continuous records from 55 stations into the high-resolution event catalog used for relocations and the spatiotemporal, depth, and b-value analyses. The authors report 53,838 detections, roughly four times the manual catalog, while recalling about 99% of the manual events.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/abs/pii/S0040195124004037"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-09-machine-learning-guided-discovery-of-enhanced-anomalous-hall-resistivity-in-feirpt-films",
    "title": "Machine learning guided discovery of enhanced anomalous Hall resistivity in Fe–Ir–Pt films",
    "summary": "A machine-learning screen trained on experimental Fe–heavy-element binary alloys pointed to a ternary Fe–Ir–Pt composition range with enhanced anomalous Hall effect. High-throughput composition-spread measurements experimentally found a peak anomalous Hall resistivity of 6.5 μΩ·cm, above the prior 5.25 μΩ·cm maximum reported for Fe–X binary systems. Scaling analysis attributed the enhancement to an extrinsic contribution.",
    "field": "materials",
    "ai_technology": "Gaussian process regression using Magpie composition descriptors",
    "organization": "National Institute for Materials Science (NIMS)",
    "source_url": "https://www.nature.com/articles/s41524-025-01757-5",
    "source_type": "paper",
    "publication_date": "2025-09-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning screen trained on experimental Fe–heavy-element binary alloys pointed to a ternary Fe–Ir–Pt composition range with enhanced anomalous Hall effect.",
      "unknown": "This is a thin-film anomalous Hall resistivity result; the paper does not demonstrate a finished magnetic sensor or device. The machine-learning model overpredicted absolute values for the measured Fe–Ir–Pt compositions, although it qualitatively identified the enhancement and composition trend. The reported extrinsic origin comes from the paper’s scaling analysis. The result is limited to the measured composition-spread films and does not establish performance across bulk material forms or devices.",
      "ai_role": "The model was trained on measured Fe–X binary anomalous Hall data and screened 13,260 possible Fe-based ternary compositions for a larger effect. Its predicted Fe–Ir–Pt region guided experimental testing in composition-spread films; the machine-learning prediction and high-throughput measurement together identified the enhanced-effect composition range.",
      "evidence_urls": [
        "https://www.nims.go.jp/eng/press/2025/09/202509040.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-11-bayesian-optimization-experimentally-identifies-a-high-ahe-feconitair-amorphous-film",
    "title": "Bayesian optimization experimentally identifies a high-AHE Fe–Co–Ni–Ta–Ir amorphous film",
    "summary": "An autonomous closed-loop search combining composition-spread experiments with Bayesian optimization reached a maximum anomalous Hall resistivity of 10.9 μΩ·cm in an amorphous Fe44.9Co27.9Ni12.1Ta3.3Ir11.7 thin film deposited at room temperature. This measured value exceeded the study’s target of 10 μΩ·cm and the values found in the initial random-exploration cycles.",
    "field": "materials",
    "ai_technology": "Bayesian optimization for composition-spread experiments",
    "organization": "National Institute for Materials Science (NIMS); University of Tokyo",
    "source_url": "https://www.nature.com/articles/s41524-025-01828-7",
    "source_type": "paper",
    "publication_date": "2025-11-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-025-01828-7",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An autonomous closed-loop search combining composition-spread experiments with Bayesian optimization reached a maximum anomalous Hall resistivity of 10.9 μΩ·cm in an amorphous Fe44.9Co27.9Ni12.1Ta3.3Ir11.7 thin film deposited at room temperature.",
      "unknown": "The 10.9 μΩ·cm maximum is the best among 234 compositions explored over 18 cycles in this search, not a global proof of the maximum over all compositions. The demonstrated film is amorphous and was measured at room temperature on thermally oxidized silicon; the study does not report a completed sensor device or independent reproduction. Although the experimental decision loop was automated, human sample transfers between instruments remained.",
      "ai_role": "The method selects which five-element alloy composition-spread film to make next and which pair of elements to grade across it, using prior AHE measurements. The closed loop then executes deposition, multi-device fabrication and AHE measurement, feeding results into the next composition proposal; the Bayesian optimization was central to reaching the high-AHE Fe–Co–Ni–Ta–Ir composition.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01828-7"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-ai-assisted-work-proves-a-latin-square-imbalance-bound-with-optimal-examples-through-order-52",
    "title": "AI-assisted work proves a Latin-square imbalance bound, with optimal examples through order 52",
    "summary": "A collaborative LLM-plus-symbolic-computation study proved that every n×n Latin square with n ≡ 1 (mod 3) has imbalance at least 4n(n−1)/9. It introduced near-perfect permutations that attain the bound and computationally found independently verified matching examples for every such order from 4 through 52. The universal lower-bound theorem was formally checked in Lean 4.",
    "field": "mathematics",
    "ai_technology": "Claude Opus 4.5 LLM agent with SageMath, Rust exhaustive solver, simulated annealing and Lean 4 verification tools",
    "organization": "Cornell University; TU Wien",
    "source_url": "https://arxiv.org/html/2603.08322v2",
    "source_type": "paper",
    "publication_date": "2026-03-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2603.08322",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A collaborative LLM-plus-symbolic-computation study proved that every n×n Latin square with n ≡ 1 (mod 3) has imbalance at least 4n(n−1)/9.",
      "unknown": "This is an arXiv preprint (v2 dated 2026-08-13), not yet peer-reviewed in the sources checked. The universal proven theorem is a lower bound for n ≡ 1 mod 3. Matching constructions, and hence demonstrated tightness, are computationally established only for n=4,7,…,52; the paper does not prove that the bound is attained for every larger order. Lean 4 formalizes the lower-bound inequality, not the finite construction search or universal tightness.",
      "ai_role": "The agent identified an all-even row-distance parity pattern in numerical search data, formulated and helped prove the lower bound, and used model review to catch proof errors. SageMath and a Rust solver supported symbolic analysis and enumeration; simulated annealing found matching constructions through n=52. Human strategic direction supplied the decisive pivot to the minimum-positive-imbalance question. Lean 4 formally verifies the lower-bound theorem itself.",
      "evidence_urls": [
        "https://arxiv.org/html/2603.08322v2"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2026-10-recurrent-neural-network-reveals-near-critical-seeding-of-topological-defects-in-kibblezurek-simulations",
    "title": "Recurrent neural network reveals near-critical seeding of topological defects in Kibble–Zurek simulations",
    "summary": "In numerical simulations of a one-dimensional Landau–Ginzburg phase transition, a recurrent neural network predicted the final locations of topological defects from short sequences of order-parameter fluctuations during the Kibble–Zurek impulse regime, before freeze-out and before the defects were macroscopically visible. Prediction error across quench rates followed power-law scaling consistent with the Kibble–Zurek freeze-out time, indicating that information about the eventual defect pattern is encoded in near-critical fluctuations in this model.",
    "field": "physics",
    "ai_technology": "Recurrent neural network (RNN); LSTM and Transformer architectures were also checked",
    "organization": "The University of Tokyo; Los Alamos National Laboratory",
    "source_url": "https://journals.aps.org/prl/abstract/10.1103/fstv-lsvh",
    "source_type": "paper",
    "publication_date": "2026-10-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2508.20347",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In numerical simulations of a one-dimensional Landau–Ginzburg phase transition, a recurrent neural network predicted the final locations of topological defects from short sequences of order-parameter fluctuations during the Kibble–Zurek impulse regime, before freeze-out and before the defects were macroscopically visible.",
      "unknown": "This is a computational result from a noisy one-dimensional Landau–Ginzburg model, not an experimental observation of defects in a material or cosmological system. The paper uses overdamped dynamics (η=1), low noise temperature θ=10⁻⁸, a finite periodic grid and simulated training/test samples; similar behavior at another noise setting is reported in supplementary material. The finding is consistent with the Kibble–Zurek mechanism and an earlier deconstruction of its dynamics; it does not establish that this prediction method applies unchanged to all phase transitions. The PRL article was published 2026-10-05; the original arXiv preprint was submitted 2025-08-28 and revised as v2 on 2026-09-10.",
      "ai_role": "The RNN analyzed short time sequences from numerical order-parameter dynamics and inferred the final kink count and positions from small fluctuations during the impulse regime. This allowed the authors to identify when information about future defect positions becomes predictable, a quantity not directly provided by conventional Kibble–Zurek defect-density scaling. They report similar prediction behavior with LSTM and Transformer models, suggesting the result is not specific to one architecture.",
      "evidence_urls": [
        "https://journals.aps.org/prl/abstract/10.1103/fstv-lsvh"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-08-machine-learning-earthquake-catalog-reveals-transform-fault-in-the-zabargad-fracture-zone",
    "title": "Machine-learning earthquake catalog reveals transform fault in the Zabargad Fracture Zone",
    "summary": "A 12-month network of ocean-bottom and land seismometers yielded a high-resolution catalog of more than 3,900 earthquakes in the northern Red Sea Zabargad Fracture Zone and Mabahiss Deep. The southern cluster’s diffuse seismicity and predominantly left-lateral strike-slip focal mechanisms reveal a NE–SW transform fault, which the authors describe as possibly in an early development stage. The northern cluster instead shows NW–SE normal-faulting lineaments related to extension.",
    "field": "physics",
    "ai_technology": "OBSTransformer deep-learning phase picker, transfer-learned from EQTransformer and used through SeisBench",
    "organization": "King Abdullah University of Science and Technology (KAUST); GEOMAR Helmholtz Centre for Ocean Research",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GC012253",
    "source_type": "paper",
    "publication_date": "2025-08-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2025GC012253",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A 12-month network of ocean-bottom and land seismometers yielded a high-resolution catalog of more than 3,900 earthquakes in the northern Red Sea Zabargad Fracture Zone and Mabahiss Deep.",
      "unknown": "The transform fault is inferred from the southern seismicity pattern and selected-event focal mechanisms, not directly imaged along its full geometry; the authors describe it as possibly in an early development stage. The catalog covers a 12-month deployment and is a snapshot that may not capture episodic ridge seismicity. Exact spreading-axis configuration remains uncertain. The earthquake-processing workflow was not fully automatic: investigators manually reviewed and corrected picks, removed events recorded by fewer than three stations, and used manual focal-mechanism analysis.",
      "ai_role": "The researchers compared EQTransformer, OBSTransformer and PickBlue for automatic earthquake detection and P/S phase picking on continuous ocean-bottom and land-station data. OBSTransformer was selected because it detected more events than EQTransformer while yielding fewer false detections than PickBlue and closer agreement with manual picks. The resulting detections and picks were manually checked and corrected, then used in the catalog from which the clusters and focal-mechanism analysis revealed the faulting pattern.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GC012253"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-symbolic-regression-active-learning-identifies-12-acid-stable-oxide-candidates-by-hse06-calculations",
    "title": "Symbolic-regression active learning identifies 12 acid-stable oxide candidates by HSE06 calculations",
    "summary": "A SISSO symbolic-regression active-learning workflow identified 12 acid-stable oxide candidates from a pool of 1,470 within 30 active-learning iterations, using high-quality DFT-HSE06 Pourbaix calculations to evaluate oxygen-evolution stability. Many candidates had not been previously identified by widely used GGA calculations. The model’s property maps and uncertainty estimates helped direct calculations toward promising, underexplored regions of oxide space.",
    "field": "materials",
    "ai_technology": "SISSO symbolic regression with bootstrap and Monte Carlo feature-dropout ensembles in an active-learning loop",
    "organization": "NOMAD Laboratory, Fritz Haber Institute of the Max Planck Society",
    "source_url": "https://www.nature.com/articles/s41524-025-01596-4",
    "source_type": "paper",
    "publication_date": "2025-05-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-025-01596-4",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A SISSO symbolic-regression active-learning workflow identified 12 acid-stable oxide candidates from a pool of 1,470 within 30 active-learning iterations, using high-quality DFT-HSE06 Pourbaix calculations to evaluate oxygen-evolution stability.",
      "unknown": "This is a computational materials-property result, not experimental synthesis or electrochemical validation. The oxide structures were drawn from a known candidate space; the authors explicitly define “discovery” here as evaluating a previously unevaluated property, not discovering entirely new compounds. Acid stability is labeled at ΔG_Pourbaix^OER ≤ 0.10 eV/atom, a threshold that allows small positive values on the basis of potential self-passivation. The result depends on HSE06 DFT and the selected Pourbaix phases/conditions; practical catalytic stability must be tested experimentally.",
      "ai_role": "SISSO models learned compact descriptors of acid stability from existing calculations; model ensembles estimated prediction uncertainty, and a probability-of-feasibility acquisition strategy selected the next oxide for expensive DFT-HSE06 evaluation. The resulting feedback loop searched 1,470 candidates and identified 12 materials meeting the authors’ acid-stability criterion in 30 iterations, more efficiently than random selection.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01596-4"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-09-deep-learning-seismic-picks-reveal-upper-mantle-upwelling-conduits-beneath-southwestern-us-volcanic-fields",
    "title": "Deep-learning seismic picks reveal upper-mantle upwelling conduits beneath Southwestern US volcanic fields",
    "summary": "Using an expanded Pn-wave travel-time database, the authors construct a high-resolution uppermost-mantle velocity and anisotropy model for the southwestern United States. The tomographic interpretation identifies separate low-velocity, anisotropic upwelling conduits beneath the Black Rock Desert, Uinkaret, and San Francisco volcanic fields along the Colorado Plateau margins, and supports remnants of the ancient Farallon slab beneath the central Great Basin.",
    "field": "physics",
    "ai_technology": "Seis-PickNet deep-learning seismic phase picker",
    "organization": "Institute of Geology and Geophysics, Chinese Academy of Sciences; University of Chinese Academy of Sciences; Shandong Earthquake Agency",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2026JB034108",
    "source_type": "paper",
    "publication_date": "2026-09-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2026JB034108",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using an expanded Pn-wave travel-time database, the authors construct a high-resolution uppermost-mantle velocity and anisotropy model for the southwestern United States.",
      "unknown": "These are tomographic interpretations of seismic velocity and anisotropy, not direct images or direct measurements of magma conduits. The article abstract reports inferred independent thermal upwellings and supports (rather than proves uniquely) Farallon slab remnants. Resolution and tectonic interpretation depend on the travel-time picks and inversion assumptions; the paper's abstract does not establish direct volcanic plumbing connectivity.",
      "ai_role": "The authors used Seis-PickNet to automatically pick Pn phases and expand the regional travel-time database; those new picks underpin the velocity and anisotropy tomography from which the mantle structures are interpreted. The AI supplies the expanded seismic measurements used in the scientific inversion, rather than only classifying existing results.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2026JB034108"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-07-llm-guided-search-yields-three-synthesized-lithium-ion-cathodes-with-higher-initial-capacity-than-nmc811",
    "title": "LLM-guided search yields three synthesized lithium-ion cathodes with higher initial capacity than NMC811",
    "summary": "An expert-guided LLM workflow proposed three NMC811-derived cathode compositions—NMC-SiMg, NMC-SiCa and NMC-MgB—that the authors report as absent from the queried materials databases. After DFT-based computational screening, the team synthesized and characterized all three. At the third cycle, measured reversible capacities were 174, 169 and 160 mAh/g, compared with 135 mAh/g for NMC811 under the reported conditions (improvements of 28.8%, 25.2% and 18.5%).",
    "field": "materials",
    "ai_technology": "ChatBattery expert-guided LLM agent framework (GPT-3.5 for composition generation; GPT-o4 for qualitative voltage ranking), with MACE-MP machine-learning potential for computational energy evaluation",
    "organization": "Université de Montréal; Mila – Québec AI Institute; University of Oxford; University College London; University of Ottawa; National Research Council Canada",
    "source_url": "https://arxiv.org/abs/2507.16110",
    "source_type": "paper",
    "publication_date": "2025-07-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2507.16110",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "An expert-guided LLM workflow proposed three NMC811-derived cathode compositions—NMC-SiMg, NMC-SiCa and NMC-MgB—that the authors report as absent from the queried materials databases.",
      "unknown": "This is an arXiv preprint (v1, posted 2025-07-21); no peer-reviewed journal publication was identified in the checked source. The formulas are modifications within the established layered NMC cathode family, not fundamentally new chemistry classes. The capacity comparison is reported at the third cycle; long-term cycle life and practical cell performance are not established here. The paper notes minor unidentified impurity peaks and says the LLM's qualitative voltage ranking was not exhaustive. Human experts designed the prompts and performed synthesis/characterization, so this was an expert-guided AI discovery workflow, not autonomous discovery.",
      "ai_role": "ChatBattery generated and iteratively refined cathode hypotheses using domain-expert prompts, materials-database checks, rule-based capacity/charge screening and an LLM voltage-ranking step. The top candidates were selected for computational evaluation and wet-lab synthesis; MACE-MP screened sampled disordered supercells when direct DFT was too costly. Human experts designed prompts and conducted the experimental verification. This workflow led to the three experimentally synthesized compositions and their measured initial-cycle capacity improvements.",
      "evidence_urls": [
        "https://arxiv.org/abs/2507.16110"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-neural-network-satellite-soil-moisture-data-expose-global-hotspots-and-pathways-of-rainfall-coupling",
    "title": "Neural-network satellite soil-moisture data expose global hotspots and pathways of rainfall coupling",
    "summary": "An information-flow analysis of daily observations found statistically significant local surface-soil-moisture effects on precipitation over about 15.7% of global land (60°S–60°N) in the boreal warm season, with eight regional hotspots. Sensible-heat mediation dominated most hotspots, while evapotranspiration mediation dominated two African hotspots; remote moisture availability and boundary-layer-height variability helped explain regional differences. Only four of 16 assessed CMIP6 models reproduced the ERA5-derived relationship between soil-moisture variability and causal coupling.",
    "field": "climate",
    "ai_technology": "Neural-network-inversion satellite soil-moisture product (NNSm), analyzed with Liang–Kleeman information-flow causal inference",
    "organization": "Tsinghua University; Chinese Academy of Sciences; National University of Singapore; University of Connecticut; Peking University",
    "source_url": "https://www.nature.com/articles/s41467-025-63999-7",
    "source_type": "paper",
    "publication_date": "2025-10-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-63999-7",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An information-flow analysis of daily observations found statistically significant local surface-soil-moisture effects on precipitation over about 15.7% of global land (60°S–60°N) in the boreal warm season, with eight regional hotspots.",
      "unknown": "The mapped effect is local surface-soil-moisture-to-precipitation information flow, statistically significant at the authors' p<0.1 threshold over about 15.7% of land; the analysis does not imply local soil moisture causes rainfall everywhere or exclude remote/upwind effects. The AI contribution is the neural-network-derived satellite soil-moisture product; the causal analysis is not itself an AI model. The spatial pattern was broadly checked with ERA5, while pathway attribution depends on ERA5-reanalysis variables and the assumptions/limitations of information-flow causal inference. Results concern boreal warm seasons and the analyzed 2003–2020 satellite period (ERA5 cross-check spans 1979–2021).",
      "ai_role": "The study's long-term daily satellite soil-moisture series was generated with a neural-network inversion algorithm, extending the observed record used to map global coupling. Researchers applied information-flow analysis to this AI-derived product and precipitation observations, then cross-checked the main spatial pattern using independent ERA5 reanalysis. AI provided the satellite soil-moisture input for a concrete global climate result; the causal-inference method itself is an information-theoretic statistical technique, not a learned model.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-63999-7"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-12-bayesian-optimization-finds-high-performance-hole-transport-molecules-for-perovskite-solar-cells",
    "title": "Bayesian optimization finds high-performance hole-transport molecules for perovskite solar cells",
    "summary": "A closed-loop workflow combined high-throughput synthesis of organic semiconductors, measured photovoltaic-device performance, molecular descriptors and Bayesian optimization to search a virtual library of about one million candidate hole-transport materials. The machine-learning-guided iterations identified high-performing molecules; after device optimization, a perovskite solar cell using a selected material reached 26.23% power-conversion efficiency (25.88% certified). The work links candidate molecular structures to measured device performance in this specific cell system.",
    "field": "materials",
    "ai_technology": "Molecular-descriptor machine-learning models with Bayesian optimization in a high-throughput synthesis and device-testing loop",
    "organization": "Forschungszentrum Jülich / Helmholtz Institute Erlangen-Nürnberg; Friedrich-Alexander-Universität Erlangen-Nürnberg; Karlsruhe Institute of Technology; Ulsan National Institute of Science and Technology; Xiamen University; University of Electronic Science and Technology of China; University of Groningen",
    "source_url": "https://www.science.org/doi/10.1126/science.ads0901",
    "source_type": "paper",
    "publication_date": "2024-12-12",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.ads0901",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A closed-loop workflow combined high-throughput synthesis of organic semiconductors, measured photovoltaic-device performance, molecular descriptors and Bayesian optimization to search a virtual library of about one million candidate hole-transport materials.",
      "unknown": "The 26.23% champion and certified 25.88% values followed device-specific optimization of a selected molecule; this does not establish that the molecule alone accounts for the full efficiency gain or that it is globally optimal among all possible molecules. The paper reports a materials-selection workflow for a particular p-i-n perovskite device setup, with performance also affected by other cell layers and processing. The million-member search was virtual and only a small selected subset was synthesized and measured; authors do not claim to have found the single best molecule in that space.",
      "ai_role": "Models trained on measured data from 149 synthesized molecules linked calculated molecular descriptors to perovskite-device performance. Bayesian optimization selected new structures from a virtual library of roughly one million; high-throughput synthesis, characterization and device tests supplied feedback for two closed-loop iterations. Three promising molecules then received individual device optimization. This workflow identified hole-transport materials yielding the reported high-efficiency perovskite cells; the AI contribution was candidate selection and structure–performance modeling, with synthesis and measurements performed experimentally.",
      "evidence_urls": [
        "https://www.science.org/doi/10.1126/science.ads0901"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-04-ml-guided-discovery-of-5ani-passivator-yields-high-efficiency-stable-perovskite-solar-cells",
    "title": "ML-guided discovery of 5ANI passivator yields high-efficiency, stable perovskite solar cells",
    "summary": "An interpretable machine-learning and active-learning workflow screened a virtual library of 18,264 candidate passivation molecules and guided synthesis/testing of selected candidates for perovskite solar cells. The newly identified molecule 5-(aminomethyl)nicotinonitrile hydroiodide (5ANI) yielded 0.05 cm² devices at 27.22% power-conversion efficiency (27.18% certified) and 21.4 cm² mini-modules at 23.49%. Under the reported ISOS-L-1I operating test, 5ANI devices retained 98.7% of initial efficiency after 1,200 hours.",
    "field": "materials",
    "ai_technology": "SISSO interpretable symbolic regression and NGBoost active-learning models, coupled to DFT calculations and a Bayesian-optimization/symbolic-regression automated fabrication platform",
    "organization": "City University of Hong Kong; University of Cambridge; Imperial College London; Friedrich-Alexander University Erlangen-Nürnberg; Central South University; Chinese University of Hong Kong; Hong Kong University of Science and Technology; Huazhong University of Science and Technology",
    "source_url": "https://www.nature.com/articles/s41586-026-10482-y",
    "source_type": "paper",
    "publication_date": "2026-04-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-10482-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An interpretable machine-learning and active-learning workflow screened a virtual library of 18,264 candidate passivation molecules and guided synthesis/testing of selected candidates for perovskite solar cells.",
      "unknown": "This is a result for a specific perovskite composition, cell architecture and passivation/fabrication protocol. The ML regression was trained on only 20 initial known passivator/device datasets; four selected candidates were experimentally synthesized/validated. The reported champion efficiency is a small-area cell result; the mini-module efficiency is lower (23.49%) and laboratory-scale. The 1,200-hour stability result follows the stated ISOS-L-1I protocol and does not establish commercial lifetime. The paper reports high reproducibility from its automated setup, not factory-scale manufacturing. This is a different study and molecule from the 2024 Bayesian-optimized hole-transport materials record (Science DOI 10.1126/science.ads0901): here AI identifies an interfacial passivator 5ANI.",
      "ai_role": "The team used DFT descriptors and SISSO regression trained on measurements from 20 known passivator/device sets to learn structure–performance relationships. NGBoost active learning reduced the computational cost of estimating interfacial properties across an 18,264-molecule virtual library; model-guided screening selected four candidates for synthesis and experimental validation. 5ANI emerged as a passivator for which the automated device-fabrication loop optimized deposition and processing settings. The ML models guided material choice, while the closed-loop platform refined device fabrication and generated measured feedback.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-026-10482-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-reinforcement-learning-guided-experiments-produce-a-zr-based-bulk-metallic-glass-with-1-7-gpa-yield-strength-and-10-2-elongation",
    "title": "Reinforcement-learning-guided experiments produce a Zr-based bulk metallic glass with 1.7 GPa yield strength and 10.2% elongation",
    "summary": "A knowledge-augmented reinforcement-learning search over composition space, guided by learned property models and then validated by synthesis, identified a Zr-based bulk metallic glass, Zr63Cu15Al10Ni10Fe2. Single-step suction-cast specimens reached 1.7 GPa yield strength and 10.2% compressive plastic strain at room temperature. Across tested Fe-alloyed compositions, measured strength trends tracked model predictions; tungsten substitutions did not show the same ductility response.",
    "field": "materials",
    "ai_technology": "AIMATDESIGN knowledge-augmented reinforcement-learning framework, using machine-learning property predictors and large-language-model-assisted model refinement and domain-knowledge rewards",
    "organization": "Nanjing University of Science and Technology; Shanghai University; University of Science and Technology of China; University of Chinese Academy of Sciences",
    "source_url": "https://www.nature.com/articles/s41524-025-01894-x",
    "source_type": "paper",
    "publication_date": "2026-02-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-025-01894-x",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A knowledge-augmented reinforcement-learning search over composition space, guided by learned property models and then validated by synthesis, identified a Zr-based bulk metallic glass, Zr63Cu15Al10Ni10Fe2.",
      "unknown": "The reported measurements are room-temperature compression tests on single-step suction-cast samples; the paper notes that plastic strain was systematically lower than predicted for most tested specimens, partly because surface defects and residual stresses were not represented in the model. The 10.2% value is for the specific Fe2 composition and test protocol, not a general alloy-family property. This is distinct from a separate study of an ML-designed bulk metallic glass optimized for hardness (DOI 10.1038/s41467-026-73008-0): the present result concerns Zr63Cu15Al10Ni10Fe2 and its strength/ductility balance.",
      "ai_role": "An RL agent explored high-dimensional alloy compositions using ML classification/regression predictions as rewards and feasibility guidance. LLM-assisted feedback corrected unreliable model predictions and expert materials rules informed late-stage rewards. The authors used this workflow to choose Zr-based bulk-glass compositions for synthesis and mechanical testing; the experimentally tested Zr63Cu15Al10Ni10Fe2 composition achieved the reported strength and ductility result. This AI-guided composition search is the discovery contribution; the LLM was one component of a larger RL/ML workflow.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01894-x"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-machine-learning-selected-moo3-and-res2-deliver-high-measured-responsivity-in-extreme-ultraviolet-detectors",
    "title": "Machine-learning-selected α-MoO3 and ReS2 deliver high measured responsivity in extreme-ultraviolet detectors",
    "summary": "A cross-spectral machine-learning model trained on 1,927 visible/UV photodetector records screened materials for extreme-ultraviolet response and selected α-MoO3 and ReS2 for device validation. At the Australian Synchrotron, the α-MoO3 detector showed measured responsivity of about 50–160 A/W depending on photon energy and bias (about 60–70 A/W near 13.5 nm); ReS2 reached about 20 A/W at 91.8 eV and 1 mV, close to the model estimate. The α-MoO3 device retained repeatable current response through 100 bias cycles under high-flux exposure.",
    "field": "materials",
    "ai_technology": "Extremely Randomized Trees cross-spectral response prediction model",
    "organization": "RMIT University; University of Illinois Chicago; University of Illinois Urbana-Champaign; Shahid Beheshti University; Monash University; Pakistan Institute of Engineering and Applied Sciences; Concordia University; University of Shanghai for Science and Technology; Australian Synchrotron",
    "source_url": "https://www.nature.com/articles/s41467-025-60499-6",
    "source_type": "paper",
    "publication_date": "2025-07-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-60499-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A cross-spectral machine-learning model trained on 1,927 visible/UV photodetector records screened materials for extreme-ultraviolet response and selected α-MoO3 and ReS2 for device validation.",
      "unknown": "The reported measurements are for the fabricated devices and stated synchrotron energies, biases and exposure conditions; they do not establish equivalent responsivity across all EUV wavelengths or device architectures. The paper reports α-MoO3 responsivity ranging about 50–160 A/W across test conditions, while near 13.5 nm it reports about 60–70 A/W; ReS2 was tested at a specified energy/bias. The authors' silicon comparison varies with the comparison conditions and metric (the abstract gives about 225-fold, while the device discussion reports up to about 800-fold); this record therefore states the direct responsivity measurements without treating either ratio as universal. The material screening result is a small experimental validation set, not an exhaustive search.",
      "ai_role": "The model learned relations between material/device descriptors and detector response across wavelength ranges from a 1,927-record dataset of visible and ultraviolet photoresponse measurements. It screened candidate materials for EUV responsivity and selected α-MoO3 and ReS2 for fabrication and synchrotron testing. The AI contribution was transferring response patterns across spectral domains to prioritize materials; experiments then measured the EUV detector performance.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-60499-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-04-bayesian-optimization-and-graph-learning-identify-two-synthesized-ultra-incompressible-hard-materials",
    "title": "Bayesian optimization and graph learning identify two synthesized ultra-incompressible hard materials",
    "summary": "A Bayesian optimization workflow using symmetry-constrained crystal relaxation and the MEGNet graph neural network screened 399,960 transition-metal boride and carbide compositions. The study identified MoWC₂ (P6₃/mmc) and ReWB (Pca2₁), then synthesized both by in-situ reactive spark-plasma sintering; the measured hardness was reported to agree with model and DFT predictions. The result is limited to these two synthesized compounds and the authors’ tested synthesis conditions.",
    "field": "materials",
    "ai_technology": "BOWSR (Bayesian Optimization With Symmetry Relaxation) with the MEGNet graph neural network energy model",
    "organization": "University of California San Diego",
    "source_url": "https://arxiv.org/abs/2104.10242",
    "source_type": "paper",
    "publication_date": "2021-04-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.mattod.2021.08.012",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Bayesian optimization workflow using symmetry-constrained crystal relaxation and the MEGNet graph neural network screened 399,960 transition-metal boride and carbide compositions.",
      "unknown": "The arXiv abstract states that the two materials were synthesized and describes them as ultra-incompressible hard materials; the institutional project page reports experiments on eight selected candidates and confirmation of two novel hard materials. The result is from a preprint/journal study and the provided abstract does not give numeric hardness values. Do not imply commercial performance or broader synthesis robustness.",
      "ai_role": "MEGNet supplied graph-based formation-energy predictions; BOWSR used Bayesian optimization under crystal-symmetry constraints to relax candidate structures without DFT geometry relaxation. This enabled screening of 399,960 boride/carbide compositions, narrowing candidates for experimental synthesis.",
      "evidence_urls": [
        "https://arxiv.org/abs/2104.10242"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-06-machine-learning-identifies-five-synthesized-chevrel-phase-tellurides-with-unusual-channel-site-occupancy",
    "title": "Machine learning identifies five synthesized Chevrel-phase tellurides with unusual channel-site occupancy",
    "summary": "An interpretable machine-learned descriptor for decomposition enthalpy screened more than 200,000 Chevrel-phase compositions, predicting 48,501 below the study’s statistical synthesizability cutoff. Researchers selected five previously unreported tellurides for synthesis and successfully made all five; experiments also confirmed the predicted preference for intercalant cations to occupy Chevrel-phase channels, a rare motif in this family. The predicted candidate list is much larger than the experimentally attempted and confirmed subset. The date is the journal paper’s online publication; UC Davis announced it on 24 June 2021.",
    "field": "materials",
    "ai_technology": "SIFT descriptor-generation and interpretable machine-learning model Hδ",
    "organization": "University of California, Davis",
    "source_url": "https://doi.org/10.1021/jacs.1c02971",
    "source_type": "paper",
    "publication_date": "2021-06-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An interpretable machine-learned descriptor for decomposition enthalpy screened more than 200,000 Chevrel-phase compositions, predicting 48,501 below the study’s statistical synthesizability cutoff.",
      "unknown": "The original paper abstract reports five successful syntheses among five tellurides attempted and confirms a predicted channel-site occupancy trend. The other thousands of predicted candidates were not demonstrated as synthesized in this study. The result concerns a specific molybdenum-chalcogenide family.",
      "ai_role": "The SIFT method generated more than 560,000 candidate descriptors; an interpretable ML model trained against DFT-computed decomposition enthalpies assessed synthesizability and selected Chevrel-phase compositions for experimental follow-up.",
      "evidence_urls": [
        "https://chemistry.ucdavis.edu/news/machine-learning-assisted-discovery-chevrel-phase-tellurides"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-12-autonomous-machine-learning-experiments-identify-a-dual-anion-aqueous-sodium-electrolyte-with-a-wider-stability-window",
    "title": "Autonomous machine-learning experiments identify a dual-anion aqueous sodium electrolyte with a wider stability window",
    "summary": "A robotic electrochemistry stand coupled to Bayesian optimization tested 140 aqueous electrolyte formulas over 40 hours and selected a non-intuitive sodium mixture of NaClO4 and NaNO3. On platinum, the blend had a wider electrochemical stability window than the NaClO4 baseline despite lower total salt concentration; follow-up tests reported a 24 mV increase in the OER potential at the stated leakage-current criterion and 58% lower current density at 2 V. The result is a measured electrolyte formulation under the paper’s electrochemical protocols, not evidence of a commercial battery or field-ready cell.",
    "field": "materials",
    "ai_technology": "Dragonfly Bayesian optimization software coupled to the Otto robotic electrochemistry test stand",
    "organization": "Carnegie Mellon University",
    "source_url": "https://doi.org/10.1016/j.xcrp.2020.100264",
    "source_type": "paper",
    "publication_date": "2020-12-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A robotic electrochemistry stand coupled to Bayesian optimization tested 140 aqueous electrolyte formulas over 40 hours and selected a non-intuitive sodium mixture of NaClO4 and NaNO3.",
      "unknown": "The study measured electrochemical behavior primarily on platinum electrodes and reports a wider stability window for the selected aqueous blend under its stated protocol. This does not establish long-term performance, commercial readiness, or general improvement across battery chemistries.",
      "ai_role": "Dragonfly selected electrolyte formulations sequentially from real-time measurements and guided autonomous experiment requests to Otto; the paper reports the mixed-anion sodium formulation as the optimization result.",
      "evidence_urls": [
        "https://doi.org/10.1016/j.xcrp.2020.100264"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-03-machine-learning-guided-calculations-predict-four-rare-earth-free-magnetic-fecosi-compounds",
    "title": "Machine-learning-guided calculations predict four rare-earth-free magnetic Fe–Co–Si compounds",
    "summary": "A crystal-graph neural network screened 357,480 hypothetical Fe–Co–Si structures; adaptive genetic search and density-functional calculations then identified four predicted ternary structures with calculated saturation polarization ≥1.0 T, easy-axis anisotropy ≥1.0 MJ/m³, and Curie temperature >840 K. Their calculated energies lie within 70 meV/atom of the ternary convex hull, supporting but not proving experimental accessibility. The paper proposes the structures for synthesis; it does not experimentally demonstrate them.",
    "field": "materials",
    "ai_technology": "Crystal graph convolutional neural network (CGCNN) with an adaptive genetic algorithm",
    "organization": "University of Texas at Austin; and Ames National Laboratory",
    "source_url": "https://journals.aps.org/prmaterials/abstract/10.1103/PhysRevMaterials.7.034410",
    "source_type": "paper",
    "publication_date": "2023-03-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevMaterials.7.034410",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A crystal-graph neural network screened 357,480 hypothetical Fe–Co–Si structures; adaptive genetic search and density-functional calculations then identified four predicted ternary structures with calculated saturation polarization ≥1.0 T, easy-axis anisotropy ≥1.0 MJ/m³, and Curie temperature >840 K.",
      "unknown": "This is a theoretical, ML-guided first-principles prediction; the paper proposes the structures for synthesis and does not report experimental synthesis. The abstract describes four compounds; the results section mentions five structures meeting initial thresholds, of which four are dynamically stable. Report the four final candidates and their calculated properties only. The work establishes neither actual permanent-magnet performance nor synthesizability.",
      "ai_role": "CGCNN models rapidly screened hypothetical structures and predicted formation energy and magnetic polarization; first-principles results were fed back to train improved models across successive generations, while an adaptive genetic algorithm searched lower-energy structures. ML materially narrowed the space for the final density-functional magnetic-property calculations.",
      "evidence_urls": [
        "https://journals.aps.org/prmaterials/abstract/10.1103/PhysRevMaterials.7.034410"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-04-machine-learning-guided-synthesis-yields-a-new-blue-emitting-perovskite-like-single-crystal",
    "title": "Machine-learning-guided synthesis yields a new blue-emitting perovskite-like single crystal",
    "summary": "A high-throughput robotic synthesis and characterization workflow used a convolutional neural network to recognize crystal growth and a machine-learning regressor to guide synthesis conditions. The team reported the first synthesis of (3-PLA)₂PbCl₄, a new perovskite-like single crystal with a direct band gap near 4 eV and strong blue emission. The reported growth conditions were restrictive and required a carefully tuned search.",
    "field": "materials",
    "ai_technology": "Convolutional neural network image classifier and machine-learning synthesis-condition regressor",
    "organization": "University of Toronto",
    "source_url": "https://light.utoronto.ca/wp-content/uploads/2020/09/matter_perovskite_crystallization.pdf",
    "source_type": "paper",
    "publication_date": "2020-04-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.matt.2020.02.012",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A high-throughput robotic synthesis and characterization workflow used a convolutional neural network to recognize crystal growth and a machine-learning regressor to guide synthesis conditions.",
      "unknown": "The authors characterize this as a new perovskite-like material and report a laboratory single-crystal synthesis and blue emission. Synthesis conditions were restrictive; the paper does not establish device performance or scale-up.",
      "ai_role": "The CNN classified crystal-growth outcomes from images, and the model trained on high-throughput experimental results selected promising conditions for successive synthesis trials. That feedback-guided sequence enabled the first synthesis of (3-PLA)₂PbCl₄.",
      "evidence_urls": [
        "https://light.utoronto.ca/wp-content/uploads/2020/09/matter_perovskite_crystallization.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-08-machine-learning-validation-and-independent-cross-checks-classify-50-kepler-candidates-as-planets",
    "title": "Machine-learning validation and independent cross-checks classify 50 Kepler candidates as planets",
    "summary": "A Gaussian-process classifier, supported by random-forest, extra-trees and multilayer-perceptron models, was applied to Kepler transit candidates. From 83 candidates meeting the study’s stringent ML and data-quality criteria, the authors newly validated 50 as planets after updated vespa false-positive checks also gave each less than a 1% false-positive probability. The result is statistical validation using Kepler pipeline and stellar data, not new telescope observations of the planets.",
    "field": "physics",
    "ai_technology": "Gaussian process classifier, random forest, extra trees and multilayer perceptron classifiers",
    "organization": "University of Warwick",
    "source_url": "https://arxiv.org/abs/2008.10516",
    "source_type": "paper",
    "publication_date": "2020-08-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/staa2498",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Gaussian-process classifier, supported by random-forest, extra-trees and multilayer-perceptron models, was applied to Kepler transit candidates.",
      "unknown": "The 50 are statistically validated candidates from archival Kepler observations, not newly observed planets. The authors caution that their method is trained and demonstrated on Kepler, that atypical/outlier candidates may be unreliable, and that ML and vespa disagree for 33 other high-ML candidates; those were not counted as validated. This entry counts one paper-level validation result, not 50 separate records.",
      "ai_role": "Four machine-learning classifiers jointly ranked candidate transit signals against planet and false-positive scenarios. The paper required all four to cross 0.99 planet probability plus positional, stellar, binary and outlier checks; an updated vespa calculation then independently cross-checked the retained candidates.",
      "evidence_urls": [
        "https://arxiv.org/abs/2008.10516"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-07-ai-guided-polyverse-discovery-yields-a-high-temperature-dielectric-polymer-with-8-3-j-cm3-energy-density",
    "title": "AI-guided polyVERSE discovery yields a high-temperature dielectric polymer with 8.3 J cm⁻³ energy density",
    "summary": "The polyVERSE workflow used reaction-aware polymer generation and machine-learning property screens to identify PONB-2Me5Cl, which the researchers synthesized and cast into films. Experimental characterization measured a discharged energy density of 8.3 J cm⁻³ at 200 °C and a breakdown field of 750 MV m⁻¹; the paper compares this energy density with commercial polymer dielectrics at the same temperature. This is a laboratory polymer-film result, not a packaged capacitor or aerospace-ready device.",
    "field": "materials",
    "ai_technology": "polyVERSE with four polymer graph neural-network property predictors",
    "organization": "Georgia Institute of Technology; and University of Connecticut",
    "source_url": "https://www.nature.com/articles/s41467-024-50413-x.pdf",
    "source_type": "paper",
    "publication_date": "2024-07-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-50413-x",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The polyVERSE workflow used reaction-aware polymer generation and machine-learning property screens to identify PONB-2Me5Cl, which the researchers synthesized and cast into films.",
      "unknown": "The reported 8.3 J cm⁻³ is measured on synthesized polymer films under the article’s laboratory breakdown and dielectric characterization protocol. The comparison with commercial polymers is the authors’ reported comparison; the work does not demonstrate a packaged capacitor, operational lifetime, aerospace use, or scale-up. Other suggested polyimides in the paper remain predicted and are not counted here. The reported polymerization used non-green solvents.",
      "ai_role": "polyVERSE generated polymers from commercially available monomers using encoded reaction templates, then machine-learning models predicted glass-transition temperature, band gap and dielectric properties for screening. AI down-selected polymer candidates for synthesis; PONB-2Me5Cl was subsequently synthesized and its dielectric properties were measured.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-50413-x.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-08-ml-guided-experiments-synthesize-single-phase-basr-cerate-perovskite-solid-solutions",
    "title": "ML-guided experiments synthesize single-phase Ba–Sr cerate perovskite solid solutions",
    "summary": "Machine-learning screening and expert down-selection identified Ba–Sr cerate oxide compositions for synthesis. The researchers used an automated rapid-sintering and dielectric-characterization platform and reported X-ray diffraction evidence of single-phase solid-solution specimens with refined orthorhombic Pnma structures. The finding is an experimental validation of selected ML-predicted compositions and a process-structure result; it does not establish device performance or broad predictive success.",
    "field": "materials",
    "ai_technology": "Gradient-boosting perovskite classifier; VAE and composition-embedding screening",
    "organization": "Queen Mary University of London; QinetiQ",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11297172/",
    "source_type": "paper",
    "publication_date": "2024-08-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-50884-y",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Machine-learning screening and expert down-selection identified Ba–Sr cerate oxide compositions for synthesis.",
      "unknown": "The paper experimentally reports selected ML-predicted Ba–Sr cerate compositions as single-phase by XRD and refines their structures, while additional tested compositions showed multiple phases; ML prediction alone did not guarantee synthesis. The text is inconsistent about the Sr fraction for one reported specimen (abstract/table/figure versus one results sentence), so this entry avoids assigning that composition. The paper validates selected members of a prior predicted set rather than claiming a universal synthesis predictor. Dielectric analysis is reported, but no application/device performance is established.",
      "ai_role": "A gradient-boosting classifier trained on experimental perovskite/non-perovskite structures screened roughly 0.6 million possible compositions, assigning high perovskite likelihood to a candidate set. Variational-autoencoder and element-composition embeddings, plus human literature and synthesis-feasibility review, down-selected compositions that were then synthesized and characterized.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11297172/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-09-data-science-guided-experiments-discover-a-cotasn-rutile-alloy-oxide-with-tunable-transparency-and-acid-stable-electrocatalysis",
    "title": "Data-science-guided experiments discover a Co–Ta–Sn rutile alloy oxide with tunable transparency and acid-stable electrocatalysis",
    "summary": "A high-throughput optical screen and data-science models identified a Co–Ta–Sn composition region whose optical trends indicated phase behavior beyond simple mixtures. Follow-up X-ray diffraction verified a rutile substitutional-alloy oxide, (Sn,Co,Ta)O₂. Composition tuning produced optical transparency and oxygen-evolution electrocatalytic activity with stability in strong-acid electrolyte, as reported by the authors.",
    "field": "materials",
    "ai_technology": "Physics-aware data-science phase-diagram and emergent-property models",
    "organization": "Google Research; California Institute of Technology",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8449358/",
    "source_type": "paper",
    "publication_date": "2021-09-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2106042118",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A high-throughput optical screen and data-science models identified a Co–Ta–Sn composition region whose optical trends indicated phase behavior beyond simple mixtures.",
      "unknown": "The phase inference initially came from optical data and the authors discuss potential false positives from extrinsic film morphology; the selected Co–Ta–Sn rutile phase was followed up and verified by XRD. The optical screen identifies promising regions, not all possible phases, and may miss phases without emergent optical signatures. Reported oxygen-evolution performance and acid stability are laboratory electrocatalysis results; no commercial device or scale-up is established.",
      "ai_role": "Models analyzed high-throughput optical absorption measurements across 376,752 compositions in 108 three-cation oxide systems. They inferred candidate phase diagrams and flagged optical trends inconsistent with lower-order phase mixtures, guiding the Co–Ta–Sn follow-up composition and XRD experiments that verified a rutile alloy phase; this screening also guided catalytic characterization.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8449358/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-05-machine-learning-guided-synthesis-realizes-four-single-phase-refractory-carbide-compositions",
    "title": "Machine-learning-guided synthesis realizes four single-phase refractory carbide compositions",
    "summary": "A random-forest model ranked new chromium-containing multicomponent carbide compositions by predicted entropy-forming ability. After density-functional calculations and experimental synthesis, X-ray diffraction identified four previously unreported single-phase rock-salt compositions: CrMoNbVWC₅, CrMoNbTaWC₅, CrMoTaVWC₅, and CrMoTiVWC₅. The first three were high-EFA selections; CrMoTiVWC₅ was a medium-EFA composition and required extended annealing to reach chemical homogeneity.",
    "field": "materials",
    "ai_technology": "Random-forest models using chemical attributes, with and without CALPHAD thermodynamic features",
    "organization": "University of California San Diego; Duke University",
    "source_url": "https://www.nature.com/articles/s41524-020-0317-6",
    "source_type": "paper",
    "publication_date": "2020-05-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-020-0317-6",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A random-forest model ranked new chromium-containing multicomponent carbide compositions by predicted entropy-forming ability.",
      "unknown": "Only seven of the 70 model-screened candidates were selected for DFT and synthesis, and several tested compositions formed multiple phases. Single-phase identification used XRD and was supported by microstructure, elemental mapping and EBSD; the medium-EFA CrMoTiVWC₅ sample required extended annealing. The paper establishes synthesis and phase formation, not measured application performance or engineering readiness.",
      "ai_role": "The random-forest models learned entropy-forming ability from 56 previously reported carbide compositions and rapidly screened 70 new chromium-containing candidates. Model rankings guided selection of seven compositions for DFT calculations and synthesis; experimental phase outcomes agreed with predicted rankings, including four single-phase materials.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-020-0317-6"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-generative-machine-learning-identifies-and-validates-new-dual-and-tri-phase-eutectic-compositionally-complex-alloys",
    "title": "Generative machine learning identifies and validates new dual- and tri-phase eutectic compositionally complex alloys",
    "summary": "A conditional variational autoencoder generated candidate alloy compositions and an artificial neural network ranked eutectic-forming likelihood. The team experimentally validated previously unreported eutectic compositionally complex alloys across quaternary to senary systems. Reported examples include Al₁₆Cr₁₈Fe₂₄Ni₄₂ with a dual-phase FCC/BCC eutectic lamellar structure and Al₁₁.₃Nb₁₃.₇Co₃₆.₆Cr₈.₂Fe₇.₉Ni₂₂.₃ with a tri-phase FCC/BCC/Laves eutectic structure, supported by XRD, SEM and DSC.",
    "field": "materials",
    "ai_technology": "Conditional variational autoencoder (CVAE) and artificial neural network (ANN)",
    "organization": "City University of Hong Kong; Shenzhen University",
    "source_url": "https://www.nature.com/articles/s41524-024-01385-5",
    "source_type": "paper",
    "publication_date": "2024-09-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-024-01385-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A conditional variational autoencoder generated candidate alloy compositions and an artificial neural network ranked eutectic-forming likelihood.",
      "unknown": "The authors report five newly identified ECCA compositions across the studied systems; this is one paper-level result, not five records. The entry names two representative experimentally examined compositions. Validation is limited to the reported arc-melted samples and XRD/SEM/DSC characterization; no long-term mechanical performance or application testing is established. The training data were sparse and imbalanced, and dataset filtering was part of the framework.",
      "ai_role": "A CVAE generated candidate compositions conditioned on eutectic/non-eutectic microstructure labels; an ANN trained on composition and thermodynamics-derived descriptors estimated eutectic-forming likelihood and selected generated candidates for experimental validation. Dataset preprocessing with PCA and K-means addressed class imbalance. The ML pipeline directly proposed the compositions whose eutectic structures were examined experimentally.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-024-01385-5"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-06-x-ray-machine-learning-analysis-reveals-opposing-cd-re-displacements-and-re-5d-involvement-in-cd2re2o7-structural-order",
    "title": "X-ray machine-learning analysis reveals opposing Cd/Re displacements and Re 5d involvement in Cd₂Re₂O₇ structural order",
    "summary": "Unsupervised X-ray diffraction temperature clustering (X-TEC) analyzed high-volume temperature-dependent diffraction data from Cd₂Re₂O₇. The extracted selection rules indicate that Cd and Re displacements are approximately equal in amplitude and opposite in phase, revealing a previously unrecognized involvement of Re 5d electrons in the structural order. The analysis also resolves X-ray diffraction evidence for fluctuations associated with the structural transitions. This is a data-driven interpretation of measured diffraction, not a new direct imaging of atom trajectories.",
    "field": "materials",
    "ai_technology": "X-ray diffraction temperature clustering (X-TEC), an interpretable unsupervised machine-learning method using Gaussian-mixture clustering and reciprocal-space label smoothing",
    "organization": "Cornell University; Argonne National Laboratory; University of Tennessee, Knoxville; University of Maryland; National Institute of Standards and Technology",
    "source_url": "https://eunahkim.lassp.cornell.edu/sites/kim/files/publications/Harnessing%20interpretable%20and%20unsupervised%20machine%20learning%20to%20address%20big%20data%20from%20modern%20X-ray%20diffraction.pdf",
    "source_type": "paper",
    "publication_date": "2022-06-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2109665119",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Unsupervised X-ray diffraction temperature clustering (X-TEC) analyzed high-volume temperature-dependent diffraction data from Cd₂Re₂O₇.",
      "unknown": "The displacement relation and Re 5d involvement are physical interpretations inferred from XRD selection rules and X-TEC clustering, not direct atom-by-atom imaging. The work reports XRD evidence for a structural Goldstone mode; it does not claim that all Goldstone-mode physics in Cd₂Re₂O₇ was previously unknown. The findings concern the analyzed crystal and temperature-dependent synchrotron diffraction data.",
      "ai_role": "X-TEC clustered temperature trajectories of diffraction intensities across a high-volume reciprocal-space dataset to identify structural order parameters and weaker diffuse-scattering fluctuations. Its data-driven selection rules exposed the reciprocal-space pattern from which researchers inferred approximately equal, out-of-phase Cd and Re displacements and the Re 5d contribution; the manual crystallographic analysis would have required selecting only a small number of peaks.",
      "evidence_urls": [
        "https://eunahkim.lassp.cornell.edu/sites/kim/files/publications/Harnessing%20interpretable%20and%20unsupervised%20machine%20learning%20to%20address%20big%20data%20from%20modern%20X-ray%20diffraction.pdf"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-07-ai-classified-xpcs-links-local-relaxation-episodes-with-bulk-stress-changes-in-a-colloidal-glass",
    "title": "AI-classified XPCS links local relaxation episodes with bulk stress changes in a colloidal glass",
    "summary": "In rheo-XPCS measurements of a sheared silica-nanoparticle glass, an unsupervised convolutional autoencoder and K-means clustering grouped two-time correlation patterns by relative relaxation rate. Changes in the frequencies of slow-dynamics classes tracked non-monotonic shear-stress relaxation, while some later local slow-dynamics episodes appeared without a corresponding change in the bulk stress signal. This is a measured microscopic-to-macroscopic relationship in the study’s model colloidal system.",
    "field": "materials",
    "ai_technology": "AI-NERD: convolutional autoencoder for XPCS correlation images with K-means clustering in the learned latent space",
    "organization": "Argonne National Laboratory; University of Chicago; University of Illinois Chicago",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11251071/",
    "source_type": "paper",
    "publication_date": "2024-07-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-49381-z",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "In rheo-XPCS measurements of a sheared silica-nanoparticle glass, an unsupervised convolutional autoencoder and K-means clustering grouped two-time correlation patterns by relative relaxation rate.",
      "unknown": "The findings come from rheo-XPCS of a model concentrated silica-nanoparticle suspension under the reported shear/relaxation protocol. Cluster labels encode relative relaxation rates and latent-space distances are not quantitative physical measures. The authors describe the stress/dynamics relationship qualitatively and say further work is needed to extract quantitative physical information. A publisher correction dated 2024-09-09 corrected corresponding-author information; it did not report a change to the results.",
      "ai_role": "The unsupervised neural autoencoder compressed experimental two-time X-ray photon correlation functions into a latent representation, then K-means grouped frames by dynamical similarity. Class-transition frequencies over experimental time let researchers compare local relaxation patterns with simultaneous rheometer stress measurements, revealing where local slow dynamics coincided with stress minima and where they did not.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11251071/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-08-deep-learning-detects-an-emerging-anthropogenic-warming-fingerprint-in-daily-precipitation",
    "title": "Deep learning detects an emerging anthropogenic warming fingerprint in daily precipitation",
    "summary": "A convolutional neural network trained on 80 CESM2 large-ensemble members detected an emerging climate-change signal in observed daily precipitation fields. Across four satellite, merged-observation and reanalysis products, the model’s annual estimated global-mean temperature exceeded the historical internal-variability range from the mid-2010s onward. The signal was driven mainly by precipitation variability on timescales shorter than 10 days, with hotspots including the tropical eastern Pacific and mid-latitude storm tracks; rain-gauge checks over the eastern United States were consistent with the increase in high-frequency variability.",
    "field": "climate",
    "ai_technology": "Deep detection (DD) convolutional neural network",
    "organization": "Chonnam National University; Pohang University of Science and Technology; University of Washington; University of Hawaiʻi at Mānoa; Institute for Basic Science; Pusan National University",
    "source_url": "https://www.nature.com/articles/s41586-023-06474-x",
    "source_type": "paper",
    "publication_date": "2023-08-30",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06474-x",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A convolutional neural network trained on 80 CESM2 large-ensemble members detected an emerging climate-change signal in observed daily precipitation fields.",
      "unknown": "This is an observational detection and attribution result using a CNN trained on climate-model simulations; it is not a direct causal experiment or a forecast of individual storms. The long-term mean precipitation change remained indistinguishable from natural variability in the analysis. Precipitation products disagree in some estimates, especially where direct ocean measurements are sparse, and the reported hotspots depend on the selected datasets and attribution methods. The paper reports an emerging signal beginning in the mid-2010s, not a uniform increase in all regions or at every timescale.",
      "ai_role": "The CNN was trained on paired daily precipitation anomaly maps and annual global-mean surface-air-temperature values from 80 CESM2 Large Ensemble members. Applied to observed precipitation products, it learned spatial and temporal patterns that predicted the warming metric and exposed an emergent signal that conventional linear detection did not resolve. Explainability analyses (occlusion sensitivity, with SHAP and integrated-gradients checks) localized the strongest contributions to high-frequency precipitation variability.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-023-06474-x"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-11-autonomous-microscopy-discovers-an-edge-plasmon-in-layered-mnps3",
    "title": "Autonomous microscopy discovers an edge plasmon in layered MnPS₃",
    "summary": "Deep-kernel-learning-guided STEM-EELS measurements identified a low-energy edge-localized plasmon response in the layered antiferromagnet MnPS₃, distinct from the higher-energy bulk plasmon response. The system used local structure to predict spectra and uncertainty, then autonomously selected where to measure based on a physics-defined objective. After 10–25 spectra, about 1% of the image-space points, it localized the edge mode to the flake boundary.",
    "field": "physics",
    "ai_technology": "Deep kernel learning (feed-forward neural-network feature embedding with a Gaussian-process kernel and physics-based acquisition/scalarizer)",
    "organization": "Oak Ridge National Laboratory; University of Tennessee, Knoxville",
    "source_url": "https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.202203422",
    "source_type": "paper",
    "publication_date": "2022-11-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2108.03290",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Deep-kernel-learning-guided STEM-EELS measurements identified a low-energy edge-localized plasmon response in the layered antiferromagnet MnPS₃, distinct from the higher-energy bulk plasmon response.",
      "unknown": "The paper demonstrates the mode in the reported exfoliated MnPS₃ flake and STEM-EELS setup. The physical search objective was supplied by researchers (a spectral feature below the bulk-plasmon peak), so the AI did not independently formulate the scientific question. The authors report that the edge response varied with local boundary shape and present this as a proof-of-concept discovery workflow, not a complete theory of MnPS₃ plasmon behavior.",
      "ai_role": "The model learned a structure-to-electron-energy-loss-spectrum relationship from sequential measurements and used its predicted response and uncertainty to choose subsequent STEM probe locations. A user-specified spectral scalarizer prioritized a low-energy feature relative to the bulk plasmon; the closed-loop instrument then discovered and mapped an edge-localized response without pretraining on MnPS₃ sample data.",
      "evidence_urls": [
        "https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.202203422"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-11-machine-learning-assisted-neutron-scattering-shows-hydrostatic-pressure-tunes-interactions-in-spin-ice",
    "title": "Machine-learning-assisted neutron scattering shows hydrostatic pressure tunes interactions in spin ice",
    "summary": "Machine-learning-assisted analysis of diffuse neutron scattering from Dy₂Ti₂O₇ at ambient pressure and about 1.3 GPa extracted pressure-dependent Hamiltonian parameters and found that hydrostatic pressure modifies further-neighbor magnetic interactions and the material’s nanoscale magnetic-glass ordering. The analysis supports using pressure to tune the spin-ice glass state. A transition at substantially higher pressure is a model-based prediction, not an observed transition in this experiment.",
    "field": "materials",
    "ai_technology": "Nonlinear autoencoder, radial-basis generative surrogate and hierarchical clustering for neutron-scattering analysis",
    "organization": "Oak Ridge National Laboratory; Argonne National Laboratory; Instituto de Física de Líquidos y Sistemas Biológicos, UNLP-CONICET",
    "source_url": "https://www.nature.com/articles/s43246-022-00306-7",
    "source_type": "paper",
    "publication_date": "2022-11-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2110.15817",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Machine-learning-assisted analysis of diffuse neutron scattering from Dy₂Ti₂O₇ at ambient pressure and about 1.3 GPa extracted pressure-dependent Hamiltonian parameters and found that hydrostatic pressure modifies further-neighbor magnetic interactions and the material’s nanoscale magnetic-glass ordering.",
      "unknown": "The reported pressure comparison is between ambient pressure and an estimated hydrostatic pressure of about 1.3 GPa, using separate measurement setups. The change in magnetic couplings and glass morphology is inferred by fitting a spin model to scattering with ML-assisted analysis, not measured as a direct microscopic coupling. The authors predict that substantially higher pressure may induce a phase transition; no such transition was observed here. Some analysis and experimental steering remained manual.",
      "ai_role": "The nonlinear autoencoder compressed measured and simulated diffuse neutron-scattering volumes; a trained generative surrogate enabled fast forward prediction of scattering from spin-Hamiltonian parameters, while clustering and inverse mapping constrained parameters against the experiment. Those ML components made it practical to analyze pressure-dependent measurements and infer the change in further-neighbor couplings during the neutron experiment workflow.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43246-022-00306-7"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-04-random-forest-alert-screening-leads-to-spectroscopic-confirmation-of-four-changing-look-agns",
    "title": "Random-forest alert screening leads to spectroscopic confirmation of four changing-look AGNs",
    "summary": "A hierarchical random-forest classifier in the ALeRCE broker applied to Zwicky Transient Facility alert-stream light curves helped select candidate changing-look active galactic nuclei from objects previously classified as type 2. New spectra confirmed significant changes in broad Balmer emission in four of six observed candidates, establishing new type 2-to-type 1 transitions. The other two observed objects had only tentative evidence and are not counted as confirmed transitions.",
    "field": "physics",
    "ai_technology": "ALeRCE light-curve classifier (hierarchical imbalanced random forest)",
    "organization": "Universidad de Valparaíso; Universidad de Chile; European Southern Observatory; Millennium Institute of Astrophysics; California Institute of Technology",
    "source_url": "https://academic.oup.com/mnrasl/article/513/1/L57/6563628",
    "source_type": "paper",
    "publication_date": "2022-04-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2203.15040",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A hierarchical random-forest classifier in the ALeRCE broker applied to Zwicky Transient Facility alert-stream light curves helped select candidate changing-look active galactic nuclei from objects previously classified as type 2.",
      "unknown": "Only six of the 30 visually retained candidates received new spectra in this paper, selected partly for observability; four showed clear >3σ broad-line changes and two were tentative. Thus the reported ≥66% success rate is a small, selected follow-up sample, not an unbiased prevalence estimate. The authors note possible 30–50% misclassification in the parent type-2 sample and that the ZTF alert threshold makes the method incomplete.",
      "ai_role": "The ALeRCE random-forest classifier used variability features from ZTF light curves and ZTF/AllWISE photometric colors to classify alert sources as AGN-like. Its classifications were cross-matched against a pre-existing spectroscopic type-2 AGN sample to produce candidates with current type-1-like variability; this AI-assisted shortlist enabled follow-up spectroscopy that confirmed the transitions.",
      "evidence_urls": [
        "https://academic.oup.com/mnrasl/article/513/1/L57/6563628"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-07-bayesian-search-and-synthesis-reveal-a-high-anisotropy-fe-co-fe-ni-13-multilayer",
    "title": "Bayesian search and synthesis reveal a high-anisotropy [Fe/Co/Fe/Ni]₁₃ multilayer",
    "summary": "A first-principles and Bayesian-optimization workflow selected [Fe/Co/Fe/Ni]₁₃ for experimental fabrication from four-element, four-layer-period magnetic multilayer candidates. The researchers grew the repeated monoatomic-layer structure and measured it with X-ray diffraction and SQUID magnetometry. They report a magnetic anisotropy constant of 374 kJ m⁻³ and saturation magnetization of 1313 emu cm⁻³, and identify the material as a new ferromagnetic multilayer with reported magnetocrystalline anisotropy exceeding L1₀ FeNi and FeCo benchmarks.",
    "field": "materials",
    "ai_technology": "Gaussian-process Bayesian optimization with expected-improvement acquisition and a Matérn 5/2 kernel, coupled to first-principles magnetic-anisotropy calculations",
    "organization": "Tokyo University of Science; National Institute for Materials Science (NIMS)",
    "source_url": "https://www.tandfonline.com/doi/pdf/10.1080/27660400.2022.2094698",
    "source_type": "paper",
    "publication_date": "2022-07-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1080/27660400.2022.2094698",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A first-principles and Bayesian-optimization workflow selected [Fe/Co/Fe/Ni]₁₃ for experimental fabrication from four-element, four-layer-period magnetic multilayer candidates.",
      "unknown": "This is a proof-of-concept search over a restricted four-element, four-layer-period design space; the best candidate required 52 monoatomic layers in its repeated stack. The sample’s measured spontaneous easy axis remained in-plane and its shape-corrected effective anisotropy was negative, even though the authors report a large intrinsic magnetocrystalline-anisotropy value. Their proposed local perpendicular domains were a hypothesis requiring microscopy follow-up. The study does not establish a practical permanent magnet or device.",
      "ai_role": "The model sequentially used first-principles magnetocrystalline-anisotropy calculations to rank discrete layer sequences and search for high-anisotropy multilayers. The top three candidates were fabricated and characterized; the resulting [Fe/Co/Fe/Ni]₁₃ experimental sample gave the paper’s highest measured anisotropy and saturation magnetization among the three. This ML-guided candidate selection and iterative calculation materially directed the new synthesis.",
      "evidence_urls": [
        "https://www.tandfonline.com/doi/pdf/10.1080/27660400.2022.2094698"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-03-autonomous-ml-search-and-experiments-find-fecoir-pt-alloys-above-the-slaterpauling-magnetization-limit",
    "title": "Autonomous ML search and experiments find Fe–Co–Ir/Pt alloys above the Slater–Pauling magnetization limit",
    "summary": "A game-tree search using Gaussian-process predictions and first-principles calculations proposed dilute Ir and Pt additions as a counterintuitive way to raise Fe–Co magnetization. Combinatorial sputtering and magnetic measurements experimentally found Fe–Co–Ir and Fe–Co–Pt compositions with magnetization exceeding the conventional Fe₃Co₁ Slater–Pauling limit; the highest-moment points reported in the ternary maps were near Fe₈₁Co₁₅Ir₄ and Fe₈₅Co₁₁Pt₄. The study’s calculations attribute the effect to Ir/Pt-induced increases in local Fe and Co moments.",
    "field": "materials",
    "ai_technology": "Autonomous materials search using a Gaussian-process regression model with game-tree search and an upper-confidence-bound acquisition rule, coupled to KKR-CPA first-principles calculations",
    "organization": "NEC Corporation; National Institute of Advanced Industrial Science and Technology (AIST); Tohoku University; University of Tokyo; Japan Atomic Energy Agency",
    "source_url": "https://www.nature.com/articles/s43246-021-00135-0",
    "source_type": "paper",
    "publication_date": "2021-03-19",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43246-021-00135-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A game-tree search using Gaussian-process predictions and first-principles calculations proposed dilute Ir and Pt additions as a counterintuitive way to raise Fe–Co magnetization.",
      "unknown": "The highest-moment composition suggested by the autonomous search (Fe₈₂Co₁₃Ir₄Pt₁) was not itself the maximum in the measured ternary composition spreads; the Fe–Co–Ir and Fe–Co–Pt maps identified different peak compositions. The finding is based on sputtered, annealed thin-film composition spreads and room-temperature MOKE/SQUID measurements, not bulk ingots or device testing. The eight-element search space was narrowed by preliminary calculations. KKR-CPA results assume ideal disordered bulk states at 0 K, unlike the film measurements.",
      "ai_role": "The autonomous system iteratively selected compositions in an eight-component disordered-alloy search using ab-initio magnetic-moment estimates and a game-tree Gaussian-process model balancing exploration and exploitation. After about 43 days it proposed Fe₈₂Co₁₃Ir₄Pt₁; that hypothesis directed combinatorial synthesis and testing of Fe–Co–Ir and Fe–Co–Pt spreads. The follow-up measurements found compositions surpassing the prior Fe₃Co₁ benchmark and motivated theoretical analysis of impurity-enhanced Fe/Co local moments.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43246-021-00135-0"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-10-real-time-discovery-and-characterization-of-at2020xnd-an-at2018cow-like-fast-transient",
    "title": "Real-time discovery and characterization of AT2020xnd, an AT2018cow-like fast transient",
    "summary": "ZTF observations identified AT2020xnd (ZTF20acigmel) at redshift 0.243 as the first unambiguous AT2018cow analogue found and confirmed in real time. Follow-up showed a rapid optical rise, persistently hot blue emission, luminous radio, millimetre and X-ray output, no classical radioactive-supernova peak, and a dwarf-galaxy host. The authors argue that these properties support a distinct fast-blue-optical-transient phenomenon, while its progenitor and central engine remain uncertain.",
    "field": "physics",
    "ai_technology": "Zwicky Transient Facility real-time alert stream with a deep-learning real-bogus classifier and a custom contextual transient filter",
    "organization": "Zwicky Transient Facility collaboration; California Institute of Technology; Liverpool John Moores University; University of California, Berkeley; collaborating observatories",
    "source_url": "https://academic.oup.com/mnras/article/508/4/5138/6385772",
    "source_type": "paper",
    "publication_date": "2021-10-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stab2785",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "ZTF observations identified AT2020xnd (ZTF20acigmel) at redshift 0.243 as the first unambiguous AT2018cow analogue found and confirmed in real time.",
      "unknown": "This is one transient and a small class comparison, not a general population measurement. The AI contribution was alert filtering, not autonomous physical interpretation or confirmation; telescope follow-up was essential. The suggested distinct phenomenon is supported by shared observed properties, while the progenitor/engine interpretation remains unresolved and model-dependent.",
      "ai_role": "The paper’s custom search filter required the ZTF alert’s deep-learning real-bogus score to exceed 0.65, along with repeated detections, a high Galactic latitude and no coincident stellar counterpart. This AI-based alert filtering contributed to selecting AT2020xnd in real time; rapid multiband photometry and spectroscopic and multiwavelength human follow-up established its astrophysical nature and unusual properties.",
      "evidence_urls": [
        "https://academic.oup.com/mnras/article/508/4/5138/6385772"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-06-machine-learning-guided-experiments-identify-a-high-power-factor-pbse-snsb-thermoelectric-family",
    "title": "Machine-learning-guided experiments identify a high-power-factor PbSe:SnSb thermoelectric family",
    "summary": "A closed-loop error-correction learning workflow guided synthesis of PbSe alloyed with SnSb and documented a previously unexplored low-temperature thermoelectric family. At 300 K, the best tested composition, 2 wt% SnSb-doped PbSe, had a measured power factor above 850 µW mK⁻² and more than twice that of PbSe. Density-functional calculations attribute the modeled transport improvement to Sn/Sb interactions that reduce hole effective mass.",
    "field": "materials",
    "ai_technology": "Closed-loop error-correction learning with a neural-network update to a prior materials-property model; composition representations compared included Magpie, Roost and CrabNet",
    "organization": "University of Toronto; King Abdullah University of Science and Technology (KAUST)",
    "source_url": "https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202302575",
    "source_type": "paper",
    "publication_date": "2023-06-28",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/adma.202302575",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A closed-loop error-correction learning workflow guided synthesis of PbSe alloyed with SnSb and documented a previously unexplored low-temperature thermoelectric family.",
      "unknown": "The result is specific to the tested powder-processing, composition and measurement protocol, including synthesis below 300 °C; it is not evidence of device-level or commercial performance. The reported mechanism for the increased power factor is supported by DFT calculations, not a direct measurement of carrier effective mass. The paper compares the 2 wt% composition with PbSe and reports more than 2× power factor at room temperature.",
      "ai_role": "A model trained on published experimental thermoelectric data ranked candidate compositions, and a dense neural-network error-correction model incorporated local experimental comparisons to adapt predictions to the authors’ synthesis and measurement conditions. Iterative model-guided experiments selected the PbSe:SnSb family and 2 wt% SnSb-doped PbSe for characterization. AI supplied composition down-selection and adaptive experiment guidance; the authors synthesized and measured the samples.",
      "evidence_urls": [
        "https://advanced.onlinelibrary.wiley.com/doi/10.1002/adma.202302575"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-06-machine-learning-guided-fe2p-type-alloys-reach-cryogenic-magnetocaloric-temperatures-without-rare-earths",
    "title": "Machine-learning-guided Fe₂P-type alloys reach cryogenic magnetocaloric temperatures without rare earths",
    "summary": "An artificial neural network trained on reported experimental compositions guided optimization of Fe₂P-type Mn–Fe–P–Si alloys for sub-77 K magnetocaloric use. Experiments measured a 97 K transition for Mn₁.₇₀Fe₀.₃₀P₀.₆₃Si₀.₃₇ at 1 T; replacing a small fraction of Fe with Co lowered the transition to 73 K. The Co-containing composition showed a 7.5 J kg⁻¹ K⁻¹ magnetic entropy change and a 1.4 K reversible adiabatic temperature change under a 0–5 T field.",
    "field": "materials",
    "ai_technology": "Artificial neural network trained on a curated, literature-mined dataset of experimental Fe₂P-type magnetocaloric compounds",
    "organization": "National Institute for Materials Science (NIMS); South China Normal University",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S1359645422003238",
    "source_type": "paper",
    "publication_date": "2022-06-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.actamat.2022.117942",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An artificial neural network trained on reported experimental compositions guided optimization of Fe₂P-type Mn–Fe–P–Si alloys for sub-77 K magnetocaloric use.",
      "unknown": "The reported compositions are rare-earth-free but still contain transition metals and phosphorus/silicon; results characterize laboratory alloys and do not establish a working hydrogen-liquefaction refrigerator. The 73 K transition requires Co substitution and is reported with a modest reduction in magnetization. The 1.4 K reversible adiabatic change is measured over 0–5 T; stated performance and operating window should not be generalized beyond the investigated alloys.",
      "ai_role": "The team mined published experimental composition/property results to train an artificial neural network for composition optimization toward lower transition temperatures. Predictions guided selection of Mn₁.₇₀Fe₀.₃₀P₀.₆₃Si₀.₃₇, then a minor Co-for-Fe substitution was tested experimentally to reduce its transition temperature below 77 K. AI guided composition selection; the reported magnetic and caloric properties came from measurements on synthesized samples.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S1359645422003238"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-06-explainable-neural-network-analysis-isolates-aerosol-sensitive-regional-climate-signals-in-cesm1",
    "title": "Explainable neural-network analysis isolates aerosol-sensitive regional climate signals in CESM1",
    "summary": "Using explainable AI on single-forcing CESM1 large ensembles, the study identified the North Atlantic, Southern Ocean and Southeast Asia as key regional temperature patterns for dating the climate simulations, particularly before the early 21st century. When applied to reanalysis observations, the neural network trained on an ensemble with industrial aerosols fixed at 1920 levels predicted the actual year more closely than networks trained on the all-forcing or greenhouse-gas-fixed ensembles. The authors interpret this as evidence that simulated regional temperature signals are sensitive to aerosol forcing.",
    "field": "climate",
    "ai_technology": "Artificial neural network trained on CESM1 single-forcing and all-forcing large ensembles, interpreted using layer-wise relevance propagation and an uncertainty mask",
    "organization": "Colorado State University; National Center for Atmospheric Research; collaborators",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2021MS002464",
    "source_type": "paper",
    "publication_date": "2021-06-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2021MS002464",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using explainable AI on single-forcing CESM1 large ensembles, the study identified the North Atlantic, Southern Ocean and Southeast Asia as key regional temperature patterns for dating the climate simulations, particularly before the early 21st century.",
      "unknown": "This is a model-based attribution analysis from one climate model (CESM1) and historical/RCP8.5 simulations, with reanalysis used for evaluation; it is not a direct experiment or causal attribution of every observed regional trend. The relevance maps identify patterns useful to the trained network and do not by themselves prove a physical mechanism. The authors discuss other forcings, feedbacks and internal variability, and the net influence of aerosols remains uncertain.",
      "ai_role": "The ANN learned to predict year from global near-surface-temperature maps in simulations with different combinations of greenhouse-gas and industrial-aerosol forcing. Layer-wise relevance propagation identified regional patterns driving those predictions, and the trained networks were then tested on 20th Century Reanalysis observations. This analysis exposed the model’s aerosol-sensitive regional signal and the higher year-prediction correlation of the network trained with aerosols held at 1920 levels.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2021MS002464"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-12-ai-derived-symbolic-formula-improves-dark-matter-halo-overdensity-estimates-in-a-cosmological-simulation",
    "title": "AI-derived symbolic formula improves dark-matter halo overdensity estimates in a cosmological simulation",
    "summary": "A graph neural network trained on a detailed N-body cosmological dark-matter simulation was distilled by symbolic regression into an analytic estimator for a halo’s overdensity from its mass and the masses, positions and velocities of nearby halos. On the analyzed simulation snapshot, the new expression reduced mean absolute error to 0.0882, compared with 0.121 for the authors’ existing hand-designed estimator. The formula’s mass-weighted neighborhood term decreases with separation and includes dependence on the central halo’s mass.",
    "field": "physics",
    "ai_technology": "Sparse-representation Graph Neural Network followed by Eureqa genetic-programming symbolic regression",
    "organization": "Princeton University; DeepMind; New York University; Flatiron Institute; Carnegie Mellon University",
    "source_url": "https://proceedings.neurips.cc/paper/2020/hash/c9f2f917078bd2db12f23c3b413d9cba-Abstract.html",
    "source_type": "paper",
    "publication_date": "2020-12-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://proceedings.neurips.cc/paper_files/paper/2020/file/c9f2f917078bd2db12f23c3b413d9cba-Paper.pdf",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A graph neural network trained on a detailed N-body cosmological dark-matter simulation was distilled by symbolic regression into an analytic estimator for a halo’s overdensity from its mass and the masses, positions and velocities of nearby halos.",
      "unknown": "The formula was derived and evaluated on a single final-time-step dataset from an N-body simulation (215,854 halos), not validated against observations or independent cosmological simulations. The authors note that some terms’ physical interpretation was unclear and reserved detailed interpretation for future astrophysics work. It is a new useful estimator for the studied simulated halo population, not a confirmed universal law of dark-matter structure.",
      "ai_role": "The team trained a graph neural network to predict dark-matter halo overdensity using a cosmological N-body simulation. L1 regularization encouraged a compact learned message representation; Eureqa symbolic regression then fitted explicit equations to the learned graph-network components and replaced them with a closed-form model. This process yielded a previously unreported analytic halo-environment estimator and enabled comparison against the existing hand-designed formula.",
      "evidence_urls": [
        "https://proceedings.neurips.cc/paper/2020/hash/c9f2f917078bd2db12f23c3b413d9cba-Abstract.html"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-04-active-learning-experiments-reveal-distinct-on-field-and-off-field-polarization-switching-behavior-in-ferroelectrics",
    "title": "Active-learning experiments reveal distinct on-field and off-field polarization-switching behavior in ferroelectrics",
    "summary": "An active-learning workflow paired with automated piezoresponse-force microscopy discovered relationships between local ferroelectric domain structure and polarization-switching hysteresis. The experimentally selected sampling paths and points differed substantially between on-field and off-field hysteresis loops, indicating that the two measurement modes are dominated by different mechanisms. This is a bounded result from the reported ferroelectric sample and spectroscopy workflow, not a universal mechanism established for all ferroelectrics.",
    "field": "materials",
    "ai_technology": "Machine-learning active-learning framework for automated piezoresponse force microscopy and spectroscopy",
    "organization": "Oak Ridge National Laboratory; Toyota Technological Institute at Chicago; Nagoya Institute of Technology and collaborators",
    "source_url": "https://arxiv.org/abs/2108.06037",
    "source_type": "paper",
    "publication_date": "2022-04-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42256-022-00460-0",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An active-learning workflow paired with automated piezoresponse-force microscopy discovered relationships between local ferroelectric domain structure and polarization-switching hysteresis.",
      "unknown": "The authors report experimental correlation and different on-field/off-field discovery paths, interpreting this as evidence that the measurement modes are dominated by different mechanisms. The study does not establish a universal causal account across materials; it reports a specific automated PFM/spectroscopy investigation.",
      "ai_role": "The model learned correlations between high-dimensional local domain structure and hysteresis-loop switching characteristics, and used loop descriptors and uncertainty to guide the automated microscopy and spectroscopy measurements. Human physics insight was encoded in the acquisition function.",
      "evidence_urls": [
        "https://arxiv.org/abs/2108.06037"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-02-machine-learning-and-ab-initio-screening-predict-60-stable-two-dimensional-ferroelectric-bimetal-phosphates",
    "title": "Machine learning and ab initio screening predict 60 stable two-dimensional ferroelectric bimetal phosphates",
    "summary": "A data-driven machine-learning search combined with high-throughput ab initio calculations screened 2,964 two-dimensional bimetal phosphates and identified 60 computationally stable structures with out-of-plane ferroelectric polarization: 16 predicted ferroelectric metals and 44 predicted ferroelectric semiconductors, including seven predicted multiferroics. The calculations analyze how bimetal displacement can support polarization alongside metallicity. These are theoretical candidates and proposed physical explanations, not experimentally synthesized or measured phases.",
    "field": "materials",
    "ai_technology": "Data-driven machine-learning screening with orbital-based descriptors, combined with high-throughput ab initio calculations",
    "organization": "University of Chinese Academy of Sciences; and collaborators",
    "source_url": "https://arxiv.org/abs/2004.08527",
    "source_type": "paper",
    "publication_date": "2021-02-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.scib.2020.09.010",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A data-driven machine-learning search combined with high-throughput ab initio calculations screened 2,964 two-dimensional bimetal phosphates and identified 60 computationally stable structures with out-of-plane ferroelectric polarization: 16 predicted ferroelectric metals and 44 predicted ferroelectric semiconductors, including seven predicted multiferroics.",
      "unknown": "The 60 phases and their stability and ferroelectric properties are computational predictions from high-throughput ab initio calculations. The paper does not report experimental synthesis or measurement of these predicted structures; practical stability and device behavior remain unverified.",
      "ai_role": "Machine learning with newly developed electronic-orbital descriptors enabled systematic screening of 2,964 candidate 2D bimetal phosphates; first-principles calculations evaluated stability, polarization, and electronic structure for the reported candidate family.",
      "evidence_urls": [
        "https://arxiv.org/abs/2004.08527"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-11-deep-learning-extends-the-mapped-extent-of-a-north-anatolian-fault-slow-slip-event",
    "title": "Deep learning extends the mapped extent of a North Anatolian Fault slow-slip event",
    "summary": "A deep convolutional autoencoder trained on synthetic deformation and atmospheric-noise time series recovered millimeter-scale ground motion from real InSAR without being given the fault location. Applied to COSMO-SkyMed observations of the 2013 North Anatolian Fault slow-slip event, it detected about 2 mm of slip beyond the previously identified segment and revised the estimated along-strike event length from about 10 km to 15–20 km. The same paper used the approach to resolve transient millimeter-scale subsidence at California’s Coso geothermal field. This is one paper-level finding enabled by the model’s extraction of deformation signals from noisy satellite data.",
    "field": "physics",
    "ai_technology": "Deep convolutional autoencoder trained on synthetic InSAR time series",
    "organization": "Los Alamos National Laboratory; École Normale Supérieure; University of California, Santa Cruz and collaborators",
    "source_url": "https://arxiv.org/abs/2012.13849",
    "source_type": "paper",
    "publication_date": "2021-11-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-021-26254-3",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A deep convolutional autoencoder trained on synthetic deformation and atmospheric-noise time series recovered millimeter-scale ground motion from real InSAR without being given the fault location.",
      "unknown": "The newly resolved extension is a reanalysis of an existing 2013 InSAR event, not a newly occurring earthquake. The paper’s reported extension is model-derived; the original expert analysis had detected the larger slow-slip segment, while the autoencoder identifies an additional lower-amplitude portion. The model assumes atmospheric noise characteristics represented by its synthetic training data and omits some noise sources, including ionospheric perturbations. Coso results are a separate application within the same paper and are not counted separately.",
      "ai_role": "The deep autoencoder separated persistent ground deformation patterns from atmospheric noise in real InSAR data without prior fault-location information, resolving the additional 2 mm slow-slip segment; the model was trained on synthetic data and was not fine-tuned on the real case.",
      "evidence_urls": [
        "https://arxiv.org/abs/2012.13849"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-01-machine-learning-selection-and-gaia-astrometry-confirm-a-62-jupiter-mass-brown-dwarf-orbiting-g-15-6",
    "title": "Machine-learning selection and Gaia astrometry confirm a 62-Jupiter-mass brown dwarf orbiting G 15-6",
    "summary": "A machine-learning search of Gaia DR3 two-body astrometric orbits flagged G 15-6 as a likely low-mass companion system. Combining Gaia’s astrometric orbit with archival radial-velocity measurements resolved the inclination ambiguity and yielded a companion mass of 62.43 (+6.16/−5.44) Jupiter masses, firmly in the brown-dwarf range. The authors identify it as the first independently validated brown-dwarf companion selected by their ML method. The radial-velocity data predate the study; the new result is the Gaia orbit and AI-guided identification that resolve the companion mass, not a newly observed companion in 2025.",
    "field": "physics",
    "ai_technology": "Semi-supervised anomaly detection, extreme gradient boosting, and random-forest classifiers applied to Gaia DR3 astrometric-orbit solutions",
    "organization": "European Space Agency (ESA), European Space Astronomy Centre",
    "source_url": "https://arxiv.org/abs/2404.09350",
    "source_type": "paper",
    "publication_date": "2025-01-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/staf018",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning search of Gaia DR3 two-body astrometric orbits flagged G 15-6 as a likely low-mass companion system.",
      "unknown": "The source study is an ML-assisted target-identification and Gaia-orbit analysis, not a 2025 direct detection. Archival radial-velocity observations already established binary motion but did not constrain the inclination enough to prove a substellar mass; Gaia astrometry supplies that missing orbital constraint. The wider candidate list includes false-positive Gaia solutions and unresolved stellar/substellar cases, which are not counted as confirmed companions here.",
      "ai_role": "The machine-learning pipeline ranked Gaia DR3 two-body systems for likely low-mass companions and selected G 15-6 as a brown-dwarf candidate. The Gaia astrometric orbit, combined with the independent archival radial-velocity orbit, resolved the inclination and established the companion’s mass; the models used only the Gaia orbital solutions for their candidate selection.",
      "evidence_urls": [
        "https://arxiv.org/abs/2404.09350"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2020-07-machine-learning-identifies-seismic-tremor-precursors-to-whakaari-eruptions",
    "title": "Machine learning identifies seismic tremor precursors to Whakaari eruptions",
    "summary": "A structured machine-learning analysis of nine years of Whakaari seismic records identified four-hour energy bursts that occurred hours to days before most of the five major eruptive episodes studied. The authors interpret the pattern as possible charging of the vent hydrothermal system by hot magmatic fluids. In cross-validation, the prototype alert model gave advance warning for four of five unseen eruption instances, including at least four hours before the 2019 eruption.",
    "field": "physics",
    "ai_technology": "Structured machine-learning pipeline using engineered seismic time-series features, a random forest, and an alert decision model",
    "organization": "University of Auckland; University of Otago",
    "source_url": "https://www.nature.com/articles/s41467-020-17375-2",
    "source_type": "paper",
    "publication_date": "2020-07-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-020-17375-2",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A structured machine-learning analysis of nine years of Whakaari seismic records identified four-hour energy bursts that occurred hours to days before most of the five major eruptive episodes studied.",
      "unknown": "This is a study of five major eruptive episodes at one volcano. The 4/5 warning result is cross-validation on this historical record, not independent operational validation; the model was described as a prototype. The ensemble alert score is not a calibrated eruption probability, and alert thresholds trade missed events against false alarms. The proposed hydrothermal charging mechanism is an interpretation of the seismic pattern.",
      "ai_role": "The machine-learning pipeline analyzed continuous seismic tremor data, selected precursor patterns, and modeled short-term eruption alerts. Its feature analysis identified the four-hour energy-burst pattern; the proposed hydrothermal charging mechanism is the authors’ geophysical interpretation, not a direct measurement of subsurface fluid movement.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-020-17375-2"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-12-unsupervised-seismic-clustering-reveals-distinct-drainage-signals-at-gornergletscher",
    "title": "Unsupervised seismic clustering reveals distinct drainage signals at Gornergletscher",
    "summary": "An unsupervised analysis of five weeks of seismic records from Gornergletscher, Switzerland, separated cryoseismic signals into clusters associated with the 2007 Gornersee lake flood and with afternoon/evening meltwater flow. One high-energy, broadband noise cluster began about two days before the reported onset of subglacial flooding; the authors suggest it captures the earlier initiation of englacial and subglacial drainage and improves the flood chronology.",
    "field": "physics",
    "ai_technology": "SpecUFEx unsupervised feature extraction: non-negative matrix factorization, hidden Markov modeling, principal component analysis, and k-means clustering",
    "organization": "Columbia University; ETH Zurich; Swiss Federal Institute for Forest, Snow and Landscape Research (WSL)",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JF006909",
    "source_type": "paper",
    "publication_date": "2022-12-14",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2022JF006909",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An unsupervised analysis of five weeks of seismic records from Gornergletscher, Switzerland, separated cryoseismic signals into clusters associated with the 2007 Gornersee lake flood and with afternoon/evening meltwater flow.",
      "unknown": "The analysis covers five weeks at one glacier using the 2007 flood record. The inferred drainage start precedes lake-level evidence by about two days, but the source cannot definitively exclude ice-bed slip, till deformation, or small fractures as explanations for some spectral features. The study interprets cluster timing and spectra as hydrologic signals; it does not directly image subglacial drainage pathways.",
      "ai_role": "The unsupervised learning pipeline reduced seismic spectrograms to fingerprints and grouped similar continuous-noise and icequake signals without labels. Comparing the learned clusters with lake level, GPS, temperature, and event timing exposed distinct clusters associated with lake-flood drainage and daily meltwater flow, which supported the revised drainage chronology.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JF006909"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-08-machine-learning-isolates-aerosol-driven-cloud-cover-increase-after-the-holuhraun-eruption",
    "title": "Machine learning isolates aerosol-driven cloud-cover increase after the Holuhraun eruption",
    "summary": "Using MODIS satellite observations and a random-forest counterfactual trained on meteorology and non-eruption years, researchers estimated cloud responses to the 2014 Holuhraun volcanic aerosol plume. They found cloud fraction increased by about 10% across the studied region; this change contributed more to the estimated aerosol cooling than the smaller radiative effect from brighter clouds with smaller droplets. The study supplies an observational constraint on aerosol–cloud climate forcing.",
    "field": "climate",
    "ai_technology": "Random-forest ML-MODIS surrogate trained on meteorological reanalysis and MODIS satellite data from 2001–2020, excluding the 2014 eruption year",
    "organization": "University of Exeter; ETH Zurich; UK Met Office; NASA Goddard Space Flight Center; University of Cambridge; University of Leeds",
    "source_url": "https://www.nature.com/articles/s41561-022-00991-6",
    "source_type": "paper",
    "publication_date": "2022-08-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41561-022-00991-6",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using MODIS satellite observations and a random-forest counterfactual trained on meteorology and non-eruption years, researchers estimated cloud responses to the 2014 Holuhraun volcanic aerosol plume.",
      "unknown": "The result concerns one volcanic natural experiment, Holuhraun in 2014, and satellite observations of a North Atlantic/Northern European region. The cloud response is estimated by comparing observations with an ML counterfactual and is not a randomized intervention; it depends on model and MODIS retrieval assumptions. The paper reports about a 10% cloud-cover increase and attributes the leading radiative contribution to cloud fraction, while the droplet-size effect is also present and the liquid-water-path contribution is weak.",
      "ai_role": "The random forest estimated the cloud properties expected under observed meteorological conditions without volcanic aerosol, creating a counterfactual baseline. Comparing these predictions with 2014 MODIS observations let the authors isolate cloud changes associated with the Holuhraun aerosol perturbation from meteorological co-variability and estimate the relative radiative contributions of cloud-fraction, liquid-water-path, and droplet-size changes.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41561-022-00991-6"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-04-machine-learning-reveals-tropical-aerosol-driven-cloud-expansion-despite-stronger-rainfall",
    "title": "Machine learning reveals tropical aerosol-driven cloud expansion despite stronger rainfall",
    "summary": "Using satellite observations of four Kīlauea volcanic outgassing episodes in 2008 and 2018, an ML counterfactual analysis found that volcanic aerosols increased cloud cover and reflected sunlight over tropical marine shallow-convective clouds, even as rainfall strengthened. Cloud-cover adjustment contributed about 65% of aerosol–cloud cooling under mixed meteorological conditions and 84–87% in humid conditions; under humid, stable conditions, a 30% increase in droplet number could correspond to up to a 50% relative increase in cloud cover. The result identifies a tropical regime in which aerosol cloud expansion is strong and its response varies with humidity and stability.",
    "field": "climate",
    "ai_technology": "Random-forest machine-learning surrogates of MODIS cloud properties, radiative fluxes, and precipitation, trained on 20 years of satellite observations and meteorology while excluding volcanic years 2008 and 2018",
    "organization": "University of Exeter; Met Office; ETH Zurich; NASA Goddard Space Flight Center; University of Cambridge; University of Leeds; University of Reading",
    "source_url": "https://www.nature.com/articles/s41561-024-01427-z",
    "source_type": "paper",
    "publication_date": "2024-04-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41561-024-01427-z",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Using satellite observations of four Kīlauea volcanic outgassing episodes in 2008 and 2018, an ML counterfactual analysis found that volcanic aerosols increased cloud cover and reflected sunlight over tropical marine shallow-convective clouds, even as rainfall strengthened.",
      "unknown": "This is an observational ML-counterfactual analysis of four Kīlauea plume months over tropical marine shallow-convective clouds, not a controlled aerosol intervention or a direct test of deliberate cloud seeding. The source reports strong cloud-cover responses and radiative estimates, especially under humid/stable conditions, but results depend on satellite retrievals and the learned unperturbed baseline. The authors’ proposed mechanism for increased cloud cover alongside stronger rainfall is explicitly a hypothesis requiring further testing; do not treat the marine-cloud-brightening scenario as experimentally demonstrated.",
      "ai_role": "The random-forest surrogates estimated unperturbed cloud properties and radiative fluxes for given meteorological conditions. Comparing those counterfactuals with observations in four volcanic plume episodes isolated aerosol-associated signals from meteorological co-variability and enabled estimates of cloud-cover, rainfall, and radiative-forcing responses across distinct tropical meteorological regimes.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41561-024-01427-z"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2024-10-ai-reconstruction-extends-the-1911-french-heatwave-signal-into-september",
    "title": "AI reconstruction extends the 1911 French heatwave signal into September",
    "summary": "A partial-convolution U-Net trained on CMIP6 climate simulations and transferred to European station-based extreme indices reconstructed the poorly observed 1911 heatwave. The model estimates that in parts of southern France, more than 40% of September days exceeded the 90th percentile of maximum temperature, extending the heatwave signal into fall. The reconstructed pattern is similar to 20th Century Reanalysis Version 3 and aligns spatially with higher French mortality in 1911; the HadEX3 gridded record does not show the same pronounced heat.",
    "field": "climate",
    "ai_technology": "Climate Reconstruction AI (CRAI), a U-Net with partial convolutions trained on CMIP6 historical simulations and transferred to the HadEX-CAM European extreme-index dataset",
    "organization": "German Climate Computing Center (DKRZ); Met Office; University of New South Wales; Freie Universität Berlin",
    "source_url": "https://www.nature.com/articles/s41467-024-53464-2",
    "source_type": "paper",
    "publication_date": "2024-10-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-53464-2",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A partial-convolution U-Net trained on CMIP6 climate simulations and transferred to European station-based extreme indices reconstructed the poorly observed 1911 heatwave.",
      "unknown": "The September 1911 values are AI reconstructions rather than newly recovered thermometer readings; the paper states the source observations are sparse or missing. CRAI is trained on CMIP6 simulations and applied to European station-based extreme-index fields, so model-prior dependence remains. The article reports a similar spatial pattern in 20CRv3 and correlation with 1911 mortality, but these support the reconstructed pattern rather than directly validate each missing grid cell. The reported >40% refers to predicted percentage of days above the local 90th percentile for maximum temperature.",
      "ai_role": "The deep-learning reconstruction inferred missing monthly temperature-extreme indices from spatial context and learned climate-model patterns. This enabled a detailed estimate for the poorly observed September 1911 period; the model output exposes hot extremes in southern France that are not clearly represented in the standard HadEX3 map and supports comparison with independent 20th Century Reanalysis and historical mortality patterns.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-53464-2"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-09-ml-guided-ptfecu-nanocatalyst-shows-enhanced-oxygen-reduction-activity-and-durability",
    "title": "ML-guided PtFeCu nanocatalyst shows enhanced oxygen-reduction activity and durability",
    "summary": "A first-principles database and neural-network-potential search explored 396,862 Pt–Fe–Cu nanoparticle configurations and mapped candidate compositions. The team synthesized PtFeCu variants and measured oxygen-reduction reaction (ORR) performance. A tested PtFeCu formulation showed about 3.1-fold higher specific activity (776 μA cm⁻² Pt) and 3.2-fold higher mass activity (0.67 A mg⁻¹ Pt) than Pt/C at 0.9 V; the authors attribute activity and improved durability to Cu-dependent surface-strain modulation and reduced Fe segregation, supported by calculations and experimental tests.",
    "field": "materials",
    "ai_technology": "DFT-integrated neural-network potentials and machine-learning configuration search",
    "organization": "Yonsei University; Korea Institute of Energy Research; Hyundai Motor Company",
    "source_url": "https://doi.org/10.1016/j.checat.2021.06.001",
    "source_type": "paper",
    "publication_date": "2021-09-16",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A first-principles database and neural-network-potential search explored 396,862 Pt–Fe–Cu nanoparticle configurations and mapped candidate compositions.",
      "unknown": "The result is a PtFeCu catalyst-composition/configuration finding, not a claim that the ternary system itself was previously unknown. The paper labels the high-/low-Cu variants inconsistently between its abstract/results and conclusion, so the quoted activity values are not tied here to a disputed variant name. Activity was measured in O2-saturated 0.1 M HClO4 using a rotating-disk setup; durability used a 30,000-cycle accelerated stress test, not a full fuel-cell operating lifetime. The reported 2 g batch capacity is a synthesis-scale result, not commercial manufacturing validation.",
      "ai_role": "The authors built neural-network potentials from first-principles data, screened 396,862 PtFeCu nanoparticle configurations, and used the predicted ternary configuration/composition map to guide experimental synthesis and selection of catalyst variants. The model-supported search was used to identify composition-dependent ORR activity and stability relationships; measurements and physical interpretation were performed by the research team.",
      "evidence_urls": [
        "https://doi.org/10.1016/j.checat.2021.06.001"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-09-robot-guided-optimization-identifies-six-non-aqueous-electrolytes-with-improved-fast-charge-performance",
    "title": "Robot-guided optimization identifies six non-aqueous electrolytes with improved fast-charge performance",
    "summary": "A closed-loop platform paired Clio, an automated electrolyte-formulation and conductivity-testing robot, with Dragonfly Bayesian optimization. In 42 experiments over two workdays, it selected six non-aqueous electrolyte formulations from a LiPF₆ and EC/EMC/DMC design space. The team tested these formulations in 220 mAh graphite‖NMC pouch cells; all showed improved fast-charging capability relative to a baseline electrolyte selected in advance from the same design space. The search reached its reported target about six times faster than random search using the same automated setup.",
    "field": "materials",
    "ai_technology": "Dragonfly Bayesian optimization coupled to the Clio robotic electrolyte experimentation platform",
    "organization": "Carnegie Mellon University",
    "source_url": "https://www.nature.com/articles/s41467-022-32938-1",
    "source_type": "paper",
    "publication_date": "2022-09-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-022-32938-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A closed-loop platform paired Clio, an automated electrolyte-formulation and conductivity-testing robot, with Dragonfly Bayesian optimization.",
      "unknown": "This is a search within a defined design space of LiPF₆ in EC/EMC/DMC solvents, not a claim that the constituent chemicals or all six electrolyte recipes were wholly unprecedented. The optimization target was bulk ionic conductivity, with fast-charge cell performance as a subsequent validation. Pouch-cell tests used nominal 220 mAh graphite‖LiNi0.5Mn0.3Co0.2O2 cells and a baseline chosen from the same design space; results do not establish commercial-scale manufacturing or long-term fleet performance. The paper compares optimization time with random search in the same automated experiment.",
      "ai_role": "Dragonfly selected electrolyte formulations for experiments and used measurement feedback from Clio to adapt subsequent selections. The closed-loop system optimized measured ionic conductivity over a non-aqueous single-salt/ternary-solvent search space; selected formulations were then assembled and evaluated by the researchers in pouch cells.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-022-32938-1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-08-active-learning-search-experimentally-realizes-a-high-zt-cu1xagxgate2-thermoelectric",
    "title": "Active-learning search experimentally realizes a high-zT Cu₁₋ₓAgₓGaTe₂ thermoelectric",
    "summary": "Compressed-sensing symbolic regression and active learning screened the Cu₁₋ₓAgₓGaTe₂ thermoelectric series and guided synthesis of selected compositions. The published study reports a p-type Cu₀.₄₅Ag₀.₅₅GaTe₂ sample with a measured thermoelectric figure of merit zT of about 1.90 at 770 K, using measured heat capacity in the analysis. This is one paper-level materials result from a targeted search, not a claim that every composition in the series is new.",
    "field": "materials",
    "ai_technology": "SISSO compressed-sensing symbolic regression within an active-learning workflow",
    "organization": "University of Science and Technology of China; Chinese Academy of Sciences",
    "source_url": "https://pubs.rsc.org/en/content/articlelanding/2023/ta/d3ta03990k",
    "source_type": "paper",
    "publication_date": "2023-08-21",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D3TA03990K",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Compressed-sensing symbolic regression and active learning screened the Cu₁₋ₓAgₓGaTe₂ thermoelectric series and guided synthesis of selected compositions.",
      "unknown": "The zT value is the final peer-reviewed paper’s reported result using measured heat capacity; an earlier preprint version of the same study reported a different higher estimate (about 2.8 at 827 K), so that preprint value is not substituted here or counted as a separate finding. The reported result is a laboratory thermoelectric sample at elevated temperature, not a device or commercial-performance demonstration. Active learning searched a defined composition series and does not establish that every searched composition was novel.",
      "ai_role": "The authors used SISSO to derive interpretable descriptors from their experimental thermoelectric dataset, then used active learning to rank prospective compositions for synthesis and measurement. The model guided candidate selection; the researchers synthesized and characterized the samples and measured the properties.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlelanding/2023/ta/d3ta03990k"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-03-adaptive-learning-finds-higher-activity-fenc-oxygen-reduction-catalysts",
    "title": "Adaptive learning finds higher-activity Fe–N–C oxygen-reduction catalysts",
    "summary": "An adaptive-learning design-of-experiments loop guided robotic high-throughput synthesis of Fe–N–C oxygen-reduction electrocatalysts across a six-dimensional process/composition space. Starting from 36 characterized samples, the team synthesized 14 additional catalyst formulations from a nominal search space of roughly 10 billion combinations. Four outperformed the best sample in the initial dataset; the best showed about 33% higher ORR activity, reported as 16.3 mA mg⁻¹ at 0.8 V in rotating-disk measurements.",
    "field": "materials",
    "ai_technology": "Adaptive learning framework with machine-learning surrogates, statistical inference, uncertainty quantification, and global optimization, paired with robotic high-throughput synthesis",
    "organization": "Los Alamos National Laboratory; Argonne National Laboratory",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S0378775322015609",
    "source_type": "paper",
    "publication_date": "2023-03-01",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.jpowsour.2022.232583",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An adaptive-learning design-of-experiments loop guided robotic high-throughput synthesis of Fe–N–C oxygen-reduction electrocatalysts across a six-dimensional process/composition space.",
      "unknown": "The finding concerns optimized Fe–N–C catalyst formulations and synthesis conditions within the study’s defined six-dimensional search; it is not discovery of a new catalyst family or proof that all approximately 10 billion combinations were tested. The 33% improvement is relative to the best sample in the initial 36-sample dataset. Activity was measured using rotating-ring disk electrode testing, not a fuel-cell device or long-duration operating demonstration. The study reports 14 new synthesized catalysts and four that beat the initial samples; these are one paper-level result, not separate entries.",
      "ai_role": "The model-based design-of-experiments tool used current measurements and uncertainty to select subsequent synthesis conditions in a six-dimensional search space. It guided the selection of 14 samples for synthesis and measurement, producing four catalysts that exceeded the initial best. Robotic high-throughput synthesis and RRDE testing generated the experimental feedback; researchers interpreted the results.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S0378775322015609"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-06-machine-learning-unmixing-links-acid-rain-and-subsurface-flowpaths-to-watershed-co2-drawdown",
    "title": "Machine-learning unmixing links acid rain and subsurface flowpaths to watershed CO₂ drawdown",
    "summary": "Non-negative matrix factorization (NMF) was applied to stream-chemistry time series from three watersheds to infer solute endmembers and their changing contributions without predefining endmember chemistry. Combined with geological interpretation and independent geochemical evidence, the model separated shallow acid-rain-affected flow from deeper pyrite-weathering contributions. The analysis found that acid rain can reduce a watershed’s long-timescale CO₂-sequestration capacity and that omitting sulfide oxidation/acid-rain reactions overestimates weathering-driven CO₂ drawdown.",
    "field": "climate",
    "ai_technology": "Non-negative matrix factorization (NMF) machine-learning blind-source-separation model applied to stream chemistry",
    "organization": "Pennsylvania State University; Syracuse University",
    "source_url": "https://hess.copernicus.org/articles/25/3397/2021/index.html",
    "source_type": "paper",
    "publication_date": "2021-06-17",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.5194/hess-25-3397-2021",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Non-negative matrix factorization (NMF) was applied to stream-chemistry time series from three watersheds to infer solute endmembers and their changing contributions without predefining endmember chemistry.",
      "unknown": "This is an inference from historical stream chemistry and watershed-specific geology, not a direct measurement of global CO₂ sequestration or a causal intervention. The NMF endmembers were interpreted as subsurface flowpaths using geological knowledge and supporting geochemical evidence; the inferred acid-rain sulfate export at Shale Hills includes a 19–31-year lag estimate. The result applies to three studied watersheds and the weathering timescales considered. Endmember counts and interpretations depend on the model setup and watershed reaction-front assumptions.",
      "ai_role": "The NMF model simultaneously inferred endmember chemical signatures and per-sample mixing proportions from solute ratios without preset endmember compositions. Researchers interpreted those patterns using watershed geology, sulfur isotopes, weathering profiles and calcite saturation, then estimated mineral-weathering rates and CO₂ sequestration. The ML decomposition enabled source separation where conventional endmember assumptions were uncertain; it did not directly measure a global carbon flux.",
      "evidence_urls": [
        "https://hess.copernicus.org/articles/25/3397/2021/index.html"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-10-cnn-assisted-search-confirms-quadruply-lensed-quasar-ps-j21071611",
    "title": "CNN-assisted search confirms quadruply lensed quasar PS J2107−1611",
    "summary": "A convolutional neural network trained on simulated lens images screened 27,760 preselected quasar candidates with multiple nearby Pan-STARRS detections. Researchers visually inspected the top 2,000 scores and selected follow-up targets; spectroscopy and multi-survey imaging established PS J2107−1611 as a quadruply imaged quasar at redshift 2.673, with a 4.3-arcsecond image separation and a lensed host-galaxy arc. Its fold-image optical fluxes are roughly 15 times fainter than a smooth lens-mass model predicts, while radio flux ratios are consistent with that model. The date is the original preprint posting; the journal version appeared in November 2023.",
    "field": "physics",
    "ai_technology": "Convolutional neural network trained on simulated quadruply lensed quasars, applied to Pan-STARRS gri image cutouts",
    "organization": "École Polytechnique Fédérale de Lausanne; University of Liège; European Southern Observatory; Universidad Andrés Bello",
    "source_url": "https://arxiv.org/abs/2310.04494",
    "source_type": "paper",
    "publication_date": "2023-10-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1051/0004-6361/202348227",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A convolutional neural network trained on simulated lens images screened 27,760 preselected quasar candidates with multiple nearby Pan-STARRS detections.",
      "unknown": "This is a single confirmed system from a prefiltered candidate set; the authors visually reviewed high-scoring candidates and performed follow-up. Spectroscopy confirms the source is a quasar at z=2.673, while the four-image lens interpretation is supported by high-resolution imaging, lens modeling, a host-galaxy arc, and wavelength-dependent flux ratios. The ∼15-fold optical fold-pair anomaly is relative to the smooth macromodel; the cause is not resolved in this letter. Detailed JWST/VLT/EVN follow-up was left for future work.",
      "ai_role": "The CNN ranked image cutouts of 27,760 quasar catalog entries that already had multiple nearby Pan-STARRS detections. The team visually examined the top 2,000, selected 131 promising systems, and followed targets with imaging/spectroscopy; the model therefore enabled prioritization but did not itself confirm lensing. Human follow-up established the quasar redshift and lens configuration.",
      "evidence_urls": [
        "https://arxiv.org/abs/2310.04494"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-symbolic-regression-derives-aircraft-aerodynamic-scaling-across-sharp-cone-and-blunt-body-geometries",
    "title": "Symbolic regression derives aircraft aerodynamic scaling across sharp-cone and blunt-body geometries",
    "summary": "A Chain of Symbolic Regression (CoSR) analysis of hypersonic simulation data derived an interpretable non-power-product scaling relation for axial-force coefficient across sharp-cone and blunt-body configurations. It combines angle of attack, Mach number and Reynolds number through a dominant term and a viscous correction; the relation has a shared functional form but geometry-specific fitted constants. The authors interpret the constants as reflecting differences in shock/base drag, friction drag and pressure drag. This is a new model-derived scaling result from simulations, not a wind-tunnel or flight measurement.",
    "field": "physics",
    "ai_technology": "Chain of Symbolic Regression (CoSR), a progressive symbolic-regression framework for dimensionless physical relations",
    "organization": "Northwestern Polytechnical University; International Joint Institute of Intelligent Fluid Mechanics; National Key Laboratory of Aircraft Configuration Design",
    "source_url": "https://journals.aps.org/prresearch/abstract/10.1103/f2kp-pp4q",
    "source_type": "paper",
    "publication_date": "2026-11-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/f2kp-pp4q",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Chain of Symbolic Regression (CoSR) analysis of hypersonic simulation data derived an interpretable non-power-product scaling relation for axial-force coefficient across sharp-cone and blunt-body configurations.",
      "unknown": "The aerodynamic relation was inferred from hypersonic simulation data for two configurations (sharp cone and blunt body), not independently measured flight or wind-tunnel data. It is a parameterized model over the analyzed configuration/data range; external validation and broader geometry coverage are not established in the source. The simpler dominant component is linked to known hypersonic invariance, while the learned viscous correction and geometry-specific scaling provide the paper’s new modeled relationship. Do not treat the authors’ drag-component interpretations as direct measurements.",
      "ai_role": "CoSR progressively searched for interpretable dimensionless groups and combined their relationships instead of fitting one end-to-end expression. Applied to hypersonic simulation data, it identified a dominant angle-of-attack/Mach term and a Reynolds-number-dependent viscous correction, then fitted geometry-specific constants for two body shapes. Earlier canonical-law and fluid examples in the same paper are framework validations and are not counted separately here.",
      "evidence_urls": [
        "https://journals.aps.org/prresearch/abstract/10.1103/f2kp-pp4q"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-03-unsupervised-learning-guides-high-zt-scnisb-based-half-heusler-thermoelectrics",
    "title": "Unsupervised learning guides high-zT ScNiSb-based half-Heusler thermoelectrics",
    "summary": "Iterative unsupervised clustering of half-Heusler compounds, using known thermoelectric materials as labeled reference points, guided selection of ScNiSb-based compositions for experimental testing. The team measured peak zT ≈0.5 at 925 K for p-type Sc₀.₇Y₀.₃NiSb₀.₉₇Sn₀.₀₃ and ≈0.3 at 778 K for n-type Sc₀.₆₅Y₀.₃Ti₀.₀₅NiSb. The paper reports these as compositionally optimized thermoelectric properties, not a new crystal structure or a device-level demonstration.",
    "field": "materials",
    "ai_technology": "Iterative unsupervised clustering using prototype-based, density-based and hierarchical machine-learning algorithms",
    "organization": "Harbin Institute of Technology; Xi’an Jiaotong University; Inner Mongolia University of Technology",
    "source_url": "https://www.nature.com/articles/s41524-022-00723-9",
    "source_type": "paper",
    "publication_date": "2022-03-04",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-022-00723-9",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Iterative unsupervised clustering of half-Heusler compounds, using known thermoelectric materials as labeled reference points, guided selection of ScNiSb-based compositions for experimental testing.",
      "unknown": "The measured materials are doped/alloyed variants of the known ScNiSb half-Heusler parent; the article does not claim discovery of a new crystal structure. Reported zT values are laboratory measurements at elevated temperatures, not performance from a fabricated waste-heat device or a commercial deployment. The two compositions are part of one paper-level screening and experimental result; do not split them into separate records.",
      "ai_role": "The authors applied unsupervised clustering to Materials Project descriptors for 456 half-Heusler compounds, iteratively using clusters containing 20 known thermoelectric materials as reference groups to shortlist candidates. The resulting ScNiSb-based choices guided synthesis and transport-property measurements; the model supported candidate selection, while the researchers performed synthesis, characterization and interpretation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-022-00723-9"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-active-learning-discovers-three-high-entropy-thermoelectrics-with-zt-above-2",
    "title": "Active learning discovers three high-entropy thermoelectrics with zT above 2",
    "summary": "A Gaussian-process model and Bayesian active-learning loop used an average thermoelectric quality factor as a physics-informed descriptor to search a nine-element chalcogenide composition space of 16,206 candidates. Across six iterations, the team synthesized and measured 80 samples; three high-entropy chalcogenides reached peak zT values of about 2.1 at 573 K, 2.3 at 473 K and 2.1 at 723 K. Atomic-scale analysis of one composition also found Cu-rich interstitial clusters and local lattice strain, proposed to contribute to low lattice thermal conductivity.",
    "field": "materials",
    "ai_technology": "Gaussian-process regression with Bayesian active learning and a physics-informed average thermoelectric quality-factor descriptor",
    "organization": "Korea Advanced Institute of Science and Technology; Seoul National University; Korea Research Institute of Standards and Science; Hanbat National University; Korea University",
    "source_url": "https://doi.org/10.1002/adma.202515054",
    "source_type": "paper",
    "publication_date": "2025-10-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A Gaussian-process model and Bayesian active-learning loop used an average thermoelectric quality factor as a physics-informed descriptor to search a nine-element chalcogenide composition space of 16,206 candidates.",
      "unknown": "The zT values are experimental material-property measurements, not efficiencies demonstrated in a fabricated commercial thermoelectric device. The model used the average quality factor as a surrogate descriptor and trained on only 80 samples from the large composition space; it did not experimentally test all 16,206 candidates. Proposed high-entropy segmented-element efficiencies assume idealized interfaces. The Cu interstitial-cluster mechanism is supported by microscopy/spectroscopy for the characterized sample but is the authors’ interpretation of why performance improves.",
      "ai_role": "The model used uncertainty-aware exploration and exploitation to select the next formulations for synthesis from a 16,206-composition design space. Experimental feedback from 80 samples across six active-learning iterations updated the Gaussian-process surrogate; model-guided selection yielded three compositions with measured zT above 2. The authors then characterized the materials and interpreted atomic arrangements and transport properties.",
      "evidence_urls": [
        "https://doi.org/10.1002/adma.202515054"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-10-noise-aware-active-learning-selects-a-high-latent-heat-tinihf-shape-memory-alloy",
    "title": "Noise-aware active learning selects a high-latent-heat Ti–Ni–Hf shape-memory alloy",
    "summary": "A noise-aware Kriging model with active-learning acquisition selected candidate compositions from a million-composition NiTi-based alloy design space. After four experiments, researchers reported Ti₀.₂₆₆Ni₀.₄₈₄Hf₀.₂₅ with a measured latent heat of −36.08 J g⁻¹, 9.2% greater in magnitude than the best value in the initial training data (−33.04 J g⁻¹). The alloy also had an austenite-finish temperature of 481.71 °C and relatively small transformation hysteresis, making it a candidate for high-temperature thermal energy storage.",
    "field": "materials",
    "ai_technology": "Noise-aware Kriging surrogate model with an active-learning acquisition strategy incorporating experimental-data noise",
    "organization": "Shanghai University; Shanghai Jiao Tong University; Xi’an Jiaotong University",
    "source_url": "https://doi.org/10.1002/advs.202406216",
    "source_type": "paper",
    "publication_date": "2024-10-03",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A noise-aware Kriging model with active-learning acquisition selected candidate compositions from a million-composition NiTi-based alloy design space.",
      "unknown": "The reported latent heat retains the source’s negative sign; the 9.2% improvement is a comparison of absolute magnitudes against the initial dataset’s best value. This is one material result from an active-learning search, not exhaustive exploration of the million-composition space. The reported temperature/hysteresis properties are laboratory characterizations; the paper does not demonstrate a deployed thermal-storage module or independent reproduction.",
      "ai_role": "The model estimated the measurement-noise level from noisy experimental data, incorporated that level into Kriging predictions, and used a noise-augmented acquisition strategy to choose the next NiTi-based compositions. Four experiments in the closed loop identified Ti0.266Ni0.484Hf0.25 as a promising high-latent-heat alloy; researchers synthesized and measured the material properties.",
      "evidence_urls": [
        "https://doi.org/10.1002/advs.202406216"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-08-machine-learning-derives-new-idealized-ocean-eddy-momentum-and-heat-flux-closures",
    "title": "Machine learning derives new idealized ocean-eddy momentum and heat-flux closures",
    "summary": "Relevance-vector-machine sparse regression applied to high-resolution idealized ocean simulations produced compact momentum and temperature-flux closures for coarse grids. In the authors’ simulated setup, the inferred momentum terms include energy transfers absent from then-current closures; the parameterizations improved selected statistics when implemented in an idealized 30-km ocean model. This is a model-derived closure result, not an observation of a previously unknown ocean process.",
    "field": "climate",
    "ai_technology": "Relevance vector machine (RVM) sparse Bayesian regression; physically constrained convolutional neural network used as a comparison",
    "organization": "University of Oxford; University of California, San Diego",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2020GL088376",
    "source_type": "paper",
    "publication_date": "2020-08-06",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2020GL088376",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Relevance-vector-machine sparse regression applied to high-resolution idealized ocean simulations produced compact momentum and temperature-flux closures for coarse grids.",
      "unknown": "Derived from idealized, eddy-resolving double-gyre simulations and tested in an idealized coarse-resolution shallow-water model, not validated against observational ocean data or a global coupled climate model. The inferred tensor structure relates to earlier Reynolds-stress and deformation-based theories; the paper’s stated novelty is the additional data-derived terms/energy pathways and resulting parameterizations, not discovery of a new observed ocean phenomenon.",
      "ai_role": "RVM selected compact closed-form terms from candidate functions of simulated velocity, vorticity and deformation fields to infer subgrid momentum and temperature forcing. The authors report the expressions captured energy pathways omitted from existing parameterizations. A physically constrained CNN was also trained and evaluated, but the interpretable closure result comes from the RVM.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2020GL088376"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-08-theseus-derives-resource-efficient-designs-for-high-dimensional-photonic-ghz-and-heralded-bell-states",
    "title": "THESEUS derives resource-efficient designs for high-dimensional photonic GHZ and heralded Bell states",
    "summary": "The interpretable THESEUS design algorithm produced a three-dimensional six-photon GHZ-state blueprint that suppresses unwanted state terms toward arbitrarily high fidelity, with lower count rates, and a heralded three-dimensional Bell-state blueprint using destructive interference to cancel vacuum contributions. The same graph-based reasoning generalized the constructions to larger photon numbers or dimensions.",
    "field": "physics",
    "ai_technology": "THESEUS",
    "organization": "University of Toronto; Vector Institute; Griffith University; Freie Universität Berlin",
    "source_url": "https://journals.aps.org/prx/abstract/10.1103/PhysRevX.11.031044",
    "source_type": "paper",
    "publication_date": "2021-08-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevX.11.031044",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The interpretable THESEUS design algorithm produced a three-dimensional six-photon GHZ-state blueprint that suppresses unwanted state terms toward arbitrarily high fidelity, with lower count rates, and a heralded three-dimensional Bell-state blueprint using destructive interference to cancel vacuum contributions.",
      "unknown": "These are theoretically calculated experiment blueprints, not experimentally demonstrated states in this paper. For the six-photon GHZ design, unwanted amplitudes can be reduced arbitrarily by adjusting graph weights, at the cost of lower count rates; feasibility is argued for contemporary photonics. The heralded Bell design’s estimated count rate and fidelity are modeled, not measured.",
      "ai_role": "The authors’ AI-enabled automated design algorithm used an interpretable weighted, colored-graph representation of photonic experiments, gradient-based optimization, and topology reduction to search open experiment-design problems and expose generalizable physical concepts. The authors identify the work as explainable AI; the contribution is the algorithmically found design/concept, not merely faster benchmarking.",
      "evidence_urls": [
        "https://journals.aps.org/prx/abstract/10.1103/PhysRevX.11.031044"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2021-06-satellite-image-clustering-guides-discovery-of-four-late-miocene-fossil-sites-in-mozambiques-urema-rift",
    "title": "Satellite-image clustering guides discovery of four late Miocene fossil sites in Mozambique’s Urema Rift",
    "summary": "Unsupervised k-means clustering of seven Landsat 8 spectral bands highlighted a small set of priority areas for paleontological field survey in Gorongosa National Park. The 2018 campaign documented four new late Miocene fossil localities; three lay within the model’s target cluster and the fourth was less than 90 m away in the same gully system. Together, the sites expose the first reported estuarine coastal-forest fossil setting in the East African Rift System, filling a regional paleoenvironmental gap.",
    "field": "biology",
    "ai_technology": "Unsupervised k-means clustering of Landsat 8 imagery",
    "organization": "University of Oxford; University of Coimbra; University of North Carolina at Greensboro; University of Algarve; Gorongosa National Park",
    "source_url": "https://peerj.com/articles/11573/",
    "source_type": "paper",
    "publication_date": "2021-06-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.7717/peerj.11573",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Unsupervised k-means clustering of seven Landsat 8 spectral bands highlighted a small set of priority areas for paleontological field survey in Gorongosa National Park.",
      "unknown": "This was a pilot study in a 36 km² area and the model had notable false positives: reported precision was 55.56%, despite 84.6% overall accuracy. Three of four new localities overlapped the selected cluster; the fourth was less than 90 m away and was found by surveying the same gully system, so model localization for that site is indirect. The findings are four new fossil localities, not four new hominin taxa; the reported estuarine coastal-forest interpretation is a paleoenvironmental inference.",
      "ai_role": "The authors clustered seven spectral bands without pre-labeled land-cover classes, compared clusters with known fossil localities, and selected the cluster containing most known sites as a predictive target for field prospecting. This decision-support workflow directed the 2018 field survey that documented the new localities; a random forest was separately used to estimate spectral-band importance.",
      "evidence_urls": [
        "https://peerj.com/articles/11573/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-11-active-learning-identifies-a-high-performing-four-metal-perovskite-oxygen-evolution-catalyst",
    "title": "Active learning identifies a high-performing four-metal perovskite oxygen-evolution catalyst",
    "summary": "A graph-based machine-learning model and closed-loop active-learning experiments identified a four-metal perovskite oxide with an intrinsic oxygen-evolution overpotential of 391 mV at 10 mA cm⁻²oxide, among the lowest reported for four-metal perovskite oxides. The study also reports recovery of several experimentally observed composition trends. This is an electrocatalyst material result; the abstract does not establish commercial electrolyzer performance.",
    "field": "materials",
    "ai_technology": "Graph-based machine-learning model with active learning and closed-loop experimentation",
    "organization": "Institute for Basic Science; Seoul National University; Allchemy, Inc.; Polish Academy of Sciences; Ulsan National Institute of Science and Technology",
    "source_url": "https://www.nature.com/articles/s41563-023-01707-w",
    "source_type": "paper",
    "publication_date": "2023-11-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41563-023-01707-w",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A graph-based machine-learning model and closed-loop active-learning experiments identified a four-metal perovskite oxide with an intrinsic oxygen-evolution overpotential of 391 mV at 10 mA cm⁻²oxide, among the lowest reported for four-metal perovskite oxides.",
      "unknown": "The 391 mV figure is the reported intrinsic overpotential at 10 mA cm⁻²oxide and is compared by the authors with four-metal perovskite oxides. This is a material-level electrocatalysis measurement, not a commercial device result. The reported finding is one selected composition from a model-guided campaign; the abstract does not provide a full uncertainty interval.",
      "ai_role": "The model incorporated structural-characterization data and guided iterative material selection in a closed-loop experimental workflow. The authors state that it identified the composition with the reported 391 mV intrinsic overpotential; laboratory electrochemical measurements established the catalyst outcome.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41563-023-01707-w"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-11-machine-learning-screening-reveals-enhanced-formate-selectivity-of-cuga-for-co2-electroreduction",
    "title": "Machine-learning screening reveals enhanced formate selectivity of Cu–Ga for CO₂ electroreduction",
    "summary": "A machine-learning high-throughput screen of 465 binary metallic catalysts predicted previously unreported Cu–Ga selectivity for formate production in CO₂ electroreduction. Electrochemical tests of electrodeposited Cu–Ga measured a maximum formate Faradaic efficiency of about 38.4% at −1.05 V_RHE, roughly four times that on bare Cu under the reported conditions. This is a catalyst selectivity result for the tested electrode and electrolyte, not evidence of industrial-scale performance.",
    "field": "materials",
    "ai_technology": "DSTAR active-motif machine-learning models with a CO₂-reduction selectivity map",
    "organization": "Sogang University; Shanghai Jiao Tong University",
    "source_url": "https://www.nature.com/articles/s41467-023-43118-0",
    "source_type": "paper",
    "publication_date": "2023-11-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-43118-0",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A machine-learning high-throughput screen of 465 binary metallic catalysts predicted previously unreported Cu–Ga selectivity for formate production in CO₂ electroreduction.",
      "unknown": "The reported ~38.4% Faradaic efficiency is for the tested electrodeposited Cu–Ga electrode at −1.05 V_RHE in the authors’ cell and electrolyte, and the comparison is to bare Cu under those conditions. The paper discusses other predicted alloy outcomes too; this entry counts the single Cu–Ga study/result once. No device-scale or long-duration performance is established.",
      "ai_role": "The authors combined ML binding-energy predictions with a reaction-selectivity map to screen 465 binary metallic combinations and prioritize Cu–Ga and Cu–Pd behavior for follow-up. Experimental Cu–Ga measurements validated the previously unreported formate-selectivity prediction; the AI-enabled screening supplied the non-obvious target and selectivity hypothesis.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-023-43118-0"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-07-single-station-cnn-analysis-adds-273-visually-confirmed-earthquakes-to-the-laquila-near-fault-catalog",
    "title": "Single-station CNN analysis adds 273 visually confirmed earthquakes to the L’Aquila near-fault catalog",
    "summary": "A two-stage 1D CNN pipeline applied to continuous three-component seismic waveforms from one L’Aquila station identified 273 previously uncataloged local earthquake events during a one-month 2009 foreshock period. The study reports that 84% were assigned magnitude below 1 and that all 273 new detections were visually confirmed; the expanded catalog contained roughly three times as many new detections as the pre-existing catalog had events for that period.",
    "field": "physics",
    "ai_technology": "Two-stage one-dimensional convolutional neural network (CNN) pipeline",
    "organization": "Université Grenoble Alpes; CNRS",
    "source_url": "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2020JB021566",
    "source_type": "paper",
    "publication_date": "2021-07-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1029/2020JB021566",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A two-stage 1D CNN pipeline applied to continuous three-component seismic waveforms from one L’Aquila station identified 273 previously uncataloged local earthquake events during a one-month 2009 foreshock period.",
      "unknown": "The analysis covers one month in February 2009 from one station. The authors report 273 new local detections and six nonlocal events among 369 windows classified as local; all 273 new detections were visually confirmed. The pipeline does not provide exact event locations, and transfer to other regions was not established. Magnitudes are coarse classes, not precise estimates.",
      "ai_role": "The detector CNN scanned continuous raw three-component waveform windows for earthquake signals; a second CNN classified distance and magnitude, enabling the researchers to find and characterize uncataloged events from a single station. The CNN pipeline generated the new event detections; authors then visually checked them.",
      "evidence_urls": [
        "https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2020JB021566"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2025-05-scdori-identifies-myt1l-as-a-glioblastoma-plasticity-repressor",
    "title": "Deep learning identifies MYT1L as a regulator that restricts glioblastoma plasticity",
    "summary": "scDORI inferred enhancer-linked regulatory networks from single-cell multiome data and prioritized regulators for perturbation. MYT1L overexpression restricted alternative cell states and reduced tumor growth and invasion in mice; loss of MYT1L restored malignant features. This is preclinical evidence, not an established treatment. The result was public in a May 13, 2025 preprint; the peer-reviewed paper appeared November 4, 2026.",
    "field": "biology",
    "ai_technology": "scDORI; deep learning for enhancer-linked gene-regulatory network inference",
    "organization": "German Cancer Research Center (DKFZ); European Molecular Biology Laboratory (EMBL); Wellcome Sanger Institute",
    "source_url": "https://www.nature.com/articles/s41586-026-11177-0",
    "source_type": "paper",
    "publication_date": "2025-05-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-11177-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "test_type": "direct test",
      "tested": "Patient-derived cell perturbations and mouse xenograft experiments tested inferred regulators.",
      "unknown": "Clinical benefit and independent follow-up replication were not established by this review.",
      "ai_role": "Deep network inference nominated regulatory programs and MYT1L for experimental testing.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-026-11177-0",
        "https://www.biorxiv.org/content/10.1101/2025.05.13.653733v1"
      ]
    }
  },
  {
    "id": "2026-11-deltacata-guides-methyltransferase-engineering-for-icaritin-synthesis",
    "title": "DeltaCata guides methyltransferase engineering with improved catalytic efficiency in vitro",
    "summary": "The accepted-paper abstract reports that DeltaCata, a geometric graph neural network predicting mutation effects on enzyme kinetics, guided engineering of methyltransferase sGmOMT2. The authors report up to a sevenfold improvement in catalytic efficiency for in vitro icaritin synthesis. This experimental application goes beyond the model benchmarks; industrial-scale performance is not demonstrated by the reviewed abstract.",
    "field": "chemistry",
    "ai_technology": "DeltaCata; Siamese geometric graph neural network",
    "organization": "Sun Yat-sen University; National Supercomputing Center in Shenzhen",
    "source_url": "https://www.nature.com/articles/s41467-026-79287-x",
    "source_type": "paper",
    "publication_date": "2026-11-05",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-79287-x",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "test_type": "direct test",
      "tested": "The abstract reports experimental enzyme engineering and an in vitro catalytic-efficiency improvement.",
      "unknown": "Full experimental methods, industrial performance and independent replication were not assessed.",
      "ai_role": "The model predicts mutation-induced kinetic changes to prioritize enzyme variants.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-79287-x"
      ]
    }
  },
  {
    "id": "2026-11-machine-learning-guides-selective-submicromolar-ck2-inhibitors",
    "title": "Machine learning guides discovery of selective submicromolar CK2 inhibitors",
    "summary": "The accepted-paper abstract reports that docking, CDK-liability filtering, machine learning and error prediction guided more selective CK2 inhibitor leads. The authors report submicromolar hits and a 90% pipeline hit rate; kinase-assay figures support biochemical inhibition. These are discovery-stage compounds, with no demonstrated clinical benefit. The hit rate describes the authors' selected pipeline, not a general drug-discovery success rate.",
    "field": "chemistry",
    "ai_technology": "Machine learning; prediction-error modeling; molecular docking",
    "organization": "University of Helsinki; Chemotargets; University of Girona; University of Turku; Åbo Akademi University",
    "source_url": "https://www.nature.com/articles/s42004-026-02268-0",
    "source_type": "paper",
    "publication_date": "2026-11-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42004-026-02268-0",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "test_type": "direct test",
      "tested": "Kinase-assay figures report CK2 inhibition by selected hit compounds.",
      "unknown": "Full study methods, independent reproduction and clinical outcomes were not assessed.",
      "ai_role": "ML and error prediction prioritized compounds after docking and off-target filtering.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42004-026-02268-0",
        "https://www.nature.com/articles/s42004-026-02268-0_reference.pdf"
      ]
    }
  },
  {
    "id": "2026-11-dnase1l3-fragmentomic-phenotype-genotype-discordance",
    "title": "Machine learning reveals incomplete concordance between DNASE1L3 genotype and cfDNA phenotype",
    "summary": "Classifiers and UMAP analyzed 129,676 prenatal-test cfDNA samples, with genotype checks in 169 selected samples. The authors report fragmentomic profiles resembling DNASE1L3 R206C homozygotes in some non-homozygotes, while some confirmed homozygotes initially lacked the profile and acquired it in later pregnancies. This extends an already-known genotype association with phenotype discordance and within-person variation; it does not establish the mechanism or clinical utility.",
    "field": "biology",
    "ai_technology": "Supervised classifiers; UMAP",
    "organization": "Amsterdam UMC",
    "source_url": "https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2026.1887185/full",
    "source_type": "paper",
    "publication_date": "2026-11-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3389/fimmu.2026.1887185",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Existing-cohort analysis with selected ddPCR genotype validation and a small longitudinal subset.",
      "unknown": "The validation cohort was enriched for discordance; general-population accuracy and causal interpretation remain unresolved.",
      "ai_role": "Classifiers and low-dimensional embeddings identified fragmentomic subgroups and discordant cases.",
      "evidence_urls": [
        "https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2026.1887185/full"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-ml-potentials-resolve-lithium-cathode-transport-tradeoff",
    "title": "Machine-learned potentials resolve a disorder trade-off in lithium cathode transport",
    "summary": "Using actively fine-tuned machine-learning interatomic potentials with experimental validation, the authors report that cation disorder improves lithium-path connectivity but broadens site energies, suppressing diffusion in disordered-rocksalt cathodes. They develop an energy-adjusted percolation descriptor and a Li–Mn–Ti–O design map. This is a composition-specific transport result; the review does not assign individual capacity figures to modeled or measured material states.",
    "field": "materials",
    "ai_technology": "Actively fine-tuned machine-learning interatomic potentials",
    "organization": "Tsinghua University; University of Chinese Academy of Sciences; Binzhou Institute of Technology",
    "source_url": "https://www.nature.com/articles/s41467-026-79420-w",
    "source_type": "paper",
    "publication_date": "2026-11-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-79420-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher abstract describes atomistic analysis combined with experimental validation within the discovery study.",
      "unknown": "Review scope was the publisher abstract and metadata; full experimental details and capacity attribution were not assessed. Article-in-press text may change.",
      "ai_role": "Learned atomistic potentials enabled analysis of lithium connectivity and site-energy distributions.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-79420-w"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-gendirect-archival-m87-ring-reconstruction",
    "title": "GenDIReCT recovers the M87* ring from archival telescope data",
    "summary": "GenDIReCT used diffusion-based reconstruction conditioned on calibration-independent closure invariants to analyze EHT observations from 2017, 2018 and 2021. The authors recover ring morphology and estimate a diameter of 42.2 ± 1.5 microarcseconds; geometric modeling gives 42.3 ± 1.1. This is a new analysis of existing observations, overlapping the target and some data of the cataloged 2023 PRIMO reconstruction. It is not a new observation or the first black-hole-shadow discovery.",
    "field": "physics",
    "ai_technology": "GenDIReCT latent diffusion",
    "organization": "CSIRO; The University of Western Australia",
    "source_url": "https://www.nature.com/articles/s41586-026-11241-9",
    "source_type": "paper",
    "publication_date": "2026-11-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-11241-9",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Archival observational reconstruction, synthetic validation and geometric modeling; lower-quality epochs give less distinct morphology.",
      "unknown": "Sparse interferometric coverage and reconstruction assumptions remain limitations; physical mass estimates require separate scaling assumptions.",
      "ai_role": "Latent diffusion reconstructed images from closure invariants.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-026-11241-9",
        "https://arxiv.org/abs/2608.19675"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-provada-protein-steering",
    "title": "ProVADA steers a protein generator to adapt proteins to non-native biological contexts",
    "summary": "ProVADA steered a pretrained protein generator to produce active, drug-inhibitable cytosolic renin and deimmunized two murine antibody fragments while retaining antigen recognition. ESCAPE-seq measured reduced HLA-I peptide binding, and redesigned FMC63 variants retained CAR-T-cell killing. This summary is limited to the abstract of the October 18, 2026 version 2 preprint; the initial preprint dates to July 17, 2025.",
    "field": "biology",
    "ai_technology": "ProVADA (Mixture-Adaptation Directed Annealing sampler steering a pretrained protein generator)",
    "organization": "Stanford University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13742580/",
    "source_type": "paper",
    "publication_date": "2026-10-18",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1101/2025.07.11.664238",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Renin activity/inhibitor response, antibody antigen recognition and HLA-I peptide binding, and FMC63 CAR-T-cell killing, as reported in the version 2 abstract.",
      "unknown": "Only the abstract, PMC version header and original-server corresponding-institution metadata were assessed; full author-affiliation list and full paper were not reviewed. Preserve ID.",
      "ai_role": "A pretrained protein generator, steered by ProVADA's annealing sampler toward externally defined properties while preserving functional constraints.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13742580/",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1101%2F2025.07.11.664238%22&resultType=core&format=json",
        "https://api.biorxiv.org/details/biorxiv/10.1101/2025.07.11.664238"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-sik1-epilepsy-machine-learning",
    "title": "Machine learning identifies SIK1 as an epilepsy driver validated in rat models",
    "summary": "Machine-learning screening of LKB1-pathway genes in GEO data prioritized SIK1. In a PTZ-induced rat epilepsy model, SIK1 knockdown delayed seizures and shortened their duration, whereas overexpression worsened severity. The paper abstract also reports that restoring downstream NR4A2 counteracted SIK1-related effects in rescue and neuronal experiments. These are preclinical findings, not evidence of treatment efficacy in people. This summary is limited to the abstract and publication metadata.",
    "field": "medicine",
    "ai_technology": "Machine learning screening of LKB1 pathway genes on GEO dataset GSE143272",
    "organization": "The First People's Hospital of Qinzhou; The First Affiliated Hospital of Guangxi Medical University",
    "source_url": "https://europepmc.org/article/MED/42991733",
    "source_type": "paper",
    "publication_date": "2026-10-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3389/fcell.2026.1887268",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Gene prioritization from GEO data followed by rat SIK1 perturbations, NR4A2 rescue and neuronal electrophysiology.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "Machine learning screened LKB1 pathway genes and ranked SIK1 as the top epilepsy risk gene for experimental follow-up.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID%3A42991733&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-takotsubo-ecg-deep-learning",
    "title": "Deep learning ECG model separates Takotsubo syndrome from anterior STEMI, with a randomized reader study",
    "summary": "A preprint reports external validation of an ECG-image ensemble for Takotsubo syndrome versus anterior STEMI in 95 participants: accuracy 0.914 and AUC 0.970. Among 70 readers, estimated diagnostic correctness rose from 0.632 unaided to 0.864 with AI. Only the fourth phase randomized assistance conditions; AI-assisted reader accuracy remained below model accuracy. This summary is limited to the abstract and version 3 preprint dated November 8, 2026; the first version was posted October 19.",
    "field": "medicine",
    "ai_technology": "Self-supervised deep learning ensemble with Grad-CAM explanations",
    "organization": "North Medical Center Hiroshima City Asa Citizens Hospital",
    "source_url": "https://doi.org/10.64898/2026.10.15.26365116",
    "source_type": "paper",
    "publication_date": "2026-11-08",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "ECG-image external validation and a separate four-phase reader experiment; the fourth phase randomized assistance conditions.",
      "unknown": "Review limited to abstract and original-server version/corresponding-institution metadata; full paper and historical v1/v2 contents not assessed. DATE/ID MONTH CONFLICT: publication becomes 2026-11-08 while stable ID retains 2026-10.",
      "ai_role": "Ensemble classified 12-lead ECG images; its predictions and Grad-CAM maps were shown to readers.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.64898%2F2026.10.15.26365116%22&resultType=core&format=json",
        "https://api.biorxiv.org/details/medrxiv/10.64898/2026.10.15.26365116"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-sars-cov-2-3cl-protease-foundation-model-fep",
    "title": "Foundation model and generative AI yield 3CL protease inhibitors confirmed at sub-micromolar IC50",
    "summary": "An AI-assisted 3CL-protease inhibitor study selected 249 molecules from about five million generated candidates and evaluated 18 using free-energy perturbation. Five molecules were experimentally tested; all five inhibited the target below 1 micromolar IC50, and one inhibited more strongly than its lead. MMELON activity predictions had Pearson R=0.72 in a blind test. This summary is limited to the abstract and November 13, 2026 publication metadata; it does not establish antiviral efficacy in people.",
    "field": "chemistry",
    "ai_technology": "MMELON biomedical foundation model with generative molecular models",
    "organization": "IBM Research; Cleveland Clinic; Cleveland Clinic Lerner College of Medicine of Case Western Reserve University",
    "source_url": "https://doi.org/10.1007/s10822-026-00973-5",
    "source_type": "paper",
    "publication_date": "2026-11-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Five synthesized molecules tested for 3CL-protease inhibition after computational candidate generation and filtering.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "Generative models produced candidate molecules; the MMELON foundation model predicted activity and supported downselection.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID%3A43024107&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-iam1363-brain-penetrant-her2-clinical-data",
    "title": "AI-designed brain-penetrant HER2 inhibitor IAM1363 shows durable responses in Phase 1/1b",
    "summary": "Iambic reported updated results from an open-label Phase 1/1b study of its AI-designed HER2 inhibitor IAM1363. At the August 31, 2026 cutoff, 122 participants had received treatment across four study parts; the company reported antitumor and intracranial responses and that more than half of participants treated at 960 mg remained on treatment. This summary is limited to the company release of October 26, describing an October 23 conference poster. These uncontrolled early results do not establish comparative efficacy.",
    "field": "medicine",
    "ai_technology": "Iambic molecular superintelligence platform",
    "organization": "Iambic Therapeutics",
    "source_url": "https://ir.iambic.ai/news-releases/news-release-details/iambic-presents-updated-iam1363-clinical-data-esmo-2026-showing",
    "source_type": "press release",
    "publication_date": "2026-10-26",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Company-reported safety and preliminary response outcomes from an ongoing open-label Phase 1/1b trial.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "The molecule was 'designed using Iambic's molecular superintelligence platform' and progressed to clinic in about two years, as stated by the company.",
      "evidence_urls": [
        "https://ir.iambic.ai/news-releases/news-release-details/iambic-presents-updated-iam1363-clinical-data-esmo-2026-showing"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-nir-ii-fluorophores-llm-virtual-screening",
    "title": "LLM-based virtual screening identifies two NIR-II fluorophores tested in vitro and in vivo",
    "summary": "Researchers trained a large-language-model framework to predict fluorophore emission wavelengths and used virtual screening to prioritize NIR-II imaging molecules. Two candidates were experimentally characterized and showed favorable imaging performance in vitro and in vivo, according to the abstract. The abstract supplies no quantitative imaging effect sizes. This summary is limited to the abstract and October 15, 2026 publication metadata; the candidates are imaging probes.",
    "field": "chemistry",
    "ai_technology": "Large language model for structure-property prediction, with virtual screening",
    "organization": "Shanghai Institute of Materia Medica, Chinese Academy of Sciences; University of Chinese Academy of Sciences; Tianjin University of Traditional Chinese Medicine; Tsinghua Shenzhen International Graduate School; Shandong Laboratory of Yantai Drug Discovery, Bohai Rim Advanced Research Institute For Drug Discovery",
    "source_url": "https://doi.org/10.1002/smll.76251",
    "source_type": "paper",
    "publication_date": "2026-10-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Two computationally prioritized fluorescent candidates characterized for imaging in vitro and in vivo.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "A large language model predicted emission wavelengths and guided virtual screening to choose candidates.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/10.1002/smll.76251?af=R"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-wenhai-e2e-global-eddy-resolving-ocean-forecast",
    "title": "WenHai-E2E: end-to-end machine learning for global eddy-resolving ocean forecasts",
    "summary": "The November 13, 2026 paper introduces WenHai-E2E, which generates global eddy-resolving ocean forecasts directly from observations and reports better forecast accuracy than a numerical system. The abstract and publication metadata reviewed here identify no separate oceanographic discovery or specific field result beyond forecast performance. Full text was not assessed.",
    "field": "climate",
    "ai_technology": "WenHai-E2E (end-to-end deep learning ocean forecast model)",
    "organization": "Laoshan Laboratory; Ocean University of China; Shandong University; Qingdao Guoshi Technology Group",
    "source_url": "https://doi.org/10.1126/sciadv.aeh9058",
    "source_type": "paper",
    "publication_date": "2026-11-13",
    "added_date": "2026-11-15",
    "verification_status": "needs verification",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Forecast accuracy against a numerical ocean forecasting system, without a separate discovery in the reviewed abstract.",
      "unknown": "Admission review pending: primary full results needed to distinguish a specific scientific/application result from general method validation. Existing supported claims may be retained with bounded scope, but do not mark source checked until admission is resolved.",
      "ai_role": "Machine learning model produces the full eddy-resolving global ocean forecast from observations, replacing numerical initial conditions.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1126%2Fsciadv.aeh9058%22&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-pact-healthy-neural-priors-bci",
    "title": "Healthy neural priors improve BCI decoding in patients with motor-cortex gliomas",
    "summary": "PACT used healthy neural priors to decode hand gestures from recordings of patients with gliomas. In the reported comparison, decoding accuracy improved by more than 10%, and alignment with reference motor representations increased from r=0.66 to r=0.94. The study collected intraoperative ECoG from eight participants, four with motor-cortex gliomas and four with other gliomas. This proof of concept addresses a specific clinical population; offline decoding does not establish benefit from a deployed BCI. Review covered the abstract, affiliations and selected results.",
    "field": "biology",
    "ai_technology": "Pathology-Aware Consensus Training (PACT), a neural decoder trained with biophysical constraints",
    "organization": "West China Hospital of Sichuan University; Sichuan University; KU Leuven; Imperial College London",
    "source_url": "https://europepmc.org/article/MED/42913013",
    "source_type": "paper",
    "publication_date": "2026-10-23",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.adx5564",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Offline gesture decoding on intraoperative recordings; selected results describe eight glioma participants.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "PACT is the decoding model; it uses healthy neural recordings as priors and learns to suppress pathological noise and reweight spatial attention.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13693691/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-llm-brain-low-rank-alignment",
    "title": "A low-rank subspace of LLM representations carries most of their brain alignment",
    "summary": "A Dartmouth preprint studied which language-model features predict human neural responses during naturalistic listening. Across three intracranial/fMRI datasets and six model conditions, reduced-rank regression recovered full linear encoding performance with about 1–3% of embedding dimensions. These directions differed from the embeddings’ principal variance axes and transferred across held-out participants, less strongly across datasets. This summary is limited to the abstract and affiliation/date metadata of the October 29, 2026 preprint. It concerns measured brain–model alignment, not proof that brains use the same representations.",
    "field": "biology",
    "ai_technology": "Autoregressive and masked language models with ridge reduced-rank regression",
    "organization": "Dartmouth College",
    "source_url": "https://arxiv.org/abs/2610.40511",
    "source_type": "paper",
    "publication_date": "2026-10-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/pdf/2610.40511",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Linear encoding of observed intracranial/fMRI responses, reduced-rank analysis and transfer across held-out participants.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "LLM representations are the predictors of neural activity; reduced-rank regression identifies the brain-predictive subspace.",
      "evidence_urls": [
        "https://arxiv.org/abs/2610.40511",
        "https://arxiv.org/html/2610.40511v1"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-ckm-wearables-health-age",
    "title": "Wearable-based health age models for cardiovascular-kidney-metabolic risk profiling and coaching",
    "summary": "Google Research reported a wearable-based health-age model associated with subsequent hypertension, cardiovascular disease and type 2 diabetes in the All of Us cohort (22,748 participants). A companion world model simulated 12-week trajectories. In 490 blinded paired reviews, coaches preferred simulation-grounded plans in 71.4% of evaluations; physicians rated 97.6% of physiological claims accurate. This summary is limited to the November 11, 2026 institutional blog. The authors state all evaluations were offline and retrospective; clinical effectiveness and durable habit change require prospective trials.",
    "field": "medicine",
    "ai_technology": "Coachable health-age model, SensorWM action-conditioned world model, feasibility-constrained optimizer with a language model agent, multimodal CKM deviation model",
    "organization": "Google Research",
    "source_url": "https://research.google/blog/using-wearables-to-enable-cardiovascular-kidney-metabolic-ckm-risk-profiling-for-personalized-coaching/",
    "source_type": "blog",
    "publication_date": "2026-11-11",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2611.08568",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Retrospective clinical-risk modeling and blinded expert comparison of simulated coaching plans.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "Built the health-age metric and multimodal CKM model from wearable and clinical markers, forecast physiological responses to habit changes with a world model, and filtered and converted simulated habit plans into coaching text with a language model agent.",
      "evidence_urls": [
        "https://research.google/blog/using-wearables-to-enable-cardiovascular-kidney-metabolic-ckm-risk-profiling-for-personalized-coaching/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-bolero-predicts-cell-state-specific-gene-regulation-across-species",
    "title": "Bolero predicts cell-state-specific gene regulation from DNA across species",
    "summary": "Bolero predicted cell-specific responses to AP-1 activity and transferred a human-trained response model to mouse and macaque neuronal enhancers, according to Arc Institute. Its announcement also describes a computational AP-1/cofactor spacing hypothesis; these virtual edits do not establish biological causality. This summary is limited to the October 20, 2026 institutional announcement, which identifies a collaboration with Harvard Medical School and describes a 10.8-million-cell atlas. The linked paper is a preprint; full-paper methods and experimental independence were not assessed.",
    "field": "biology",
    "ai_technology": "Bolero (cell-state-conditioned sequence-to-function Transformer)",
    "organization": "Arc Institute; Harvard Medical School",
    "source_url": "https://arcinstitute.org/news/bolero",
    "source_type": "press release",
    "publication_date": "2026-10-20",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.10.19.760026",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Human-trained AP-1 response predictions compared with mouse/macaque neuronal enhancers; virtual regulatory perturbations.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "Bolero, a Transformer conditioned on cell-state embeddings, predicts cell-state-specific gene regulation directly from DNA and estimates responses in unprofiled cell states.",
      "evidence_urls": [
        "https://arcinstitute.org/news/bolero"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-iris-signaling-histories-single-cells",
    "title": "IRIS model reconstructs signaling histories of single cells",
    "summary": "IRIS inferred signaling histories from gene activity after training on human stem-cell perturbations. The researchers used it to identify a signaling intervention predicted to encourage lung development and report experimental confirmation in mouse embryos. This summary is limited to MIT/Whitehead’s October 15, 2026 announcement, which explicitly dates the Nature Methods paper to September 8. The release supplies no quantitative accuracy metrics; proposed organoid and therapeutic uses are future applications.",
    "field": "biology",
    "ai_technology": "IRIS (neural network with transfer learning)",
    "organization": "MIT; Whitehead Institute for Biomedical Research",
    "source_url": "https://news.mit.edu/2026/ai-model-decodes-language-cells-use-communicate-1015",
    "source_type": "press release",
    "publication_date": "2026-10-15",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://www.nature.com/articles/s41592-026-03213-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Inferred cell signaling histories and a lung-development prediction checked in mouse embryos, according to the institutional release.",
      "unknown": "Source checking does not establish independent confirmation. Review date: 2026-11-15.",
      "ai_role": "IRIS learns consistent pathway fingerprints in gene activity across cell types, reconstructs signaling histories, and generated the lung-development prediction that was tested in mouse embryos.",
      "evidence_urls": [
        "https://news.mit.edu/2026/ai-model-decodes-language-cells-use-communicate-1015"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-human-in-loop-ai-copilot-structural-design",
    "title": "Human-in-the-loop AI copilot improves topology-optimized structure design",
    "summary": "MIT’s November 13, 2026 announcement describes a machine-learning copilot that flags regions for human edits during topology optimization. It reports improved manufacturability and buckling-load performance, but the reviewed announcement identifies no separate structural discovery or experimentally demonstrated new structure beyond testing the design workflow. Synthetic feedback training and limited thickening/thinning edits are acknowledged. The full journal paper was not assessed. The publisher abstract specifies improved manufacturability or a 39% linear-buckling-load improvement; it does not assign 39% to both metrics. The paper was published November 10; this entry represents the November 13 announcement.",
    "field": "other",
    "ai_technology": "Human-in-the-loop AI copilot with topology optimization (machine-learning region-flagging model)",
    "organization": "MIT; Georgia Institute of Technology",
    "source_url": "https://news.mit.edu/2026/new-ai-copilot-helps-engineers-design-better-structures-1113",
    "source_type": "press release",
    "publication_date": "2026-11-13",
    "added_date": "2026-11-15",
    "verification_status": "needs verification",
    "paper_url": "https://www.nature.com/articles/s44172-026-00818-x",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Computational topology-optimization workflow demonstrations; no separate structural finding established in reviewed scope.",
      "unknown": "Admission review pending: primary full results needed to distinguish a specific scientific/application result from general method validation. Existing supported claims may be retained with bounded scope, but do not mark source checked until admission is resolved.",
      "ai_role": "A machine-learning model quickly identifies design regions a human would likely want to change, within a topology optimization workflow.",
      "evidence_urls": [
        "https://news.mit.edu/2026/new-ai-copilot-helps-engineers-design-better-structures-1113",
        "https://www.nature.com/articles/s44172-026-00818-x"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-depth-aware-fib-sem-porous-electrode-reconstruction",
    "title": "Depth-aware machine learning for 3D reconstruction of fibrous porous electrodes from FIB-SEM tomography",
    "summary": "The October 25, 2026 paper applies depth-aware machine learning to copper-coated polymer-fiber mats and compares reconstructions with independent physical measurements. The abstract supports improved morphology agreement, but identifies no separate electrode-property or materials discovery beyond validating reconstruction quality. This source check is limited to the abstract, electronic publication date and affiliations; the full paper was not assessed.",
    "field": "materials",
    "ai_technology": "Depth-aware machine learning: monocular depth estimation plus unsupervised clustering",
    "organization": "Empa, Swiss Federal Laboratories for Materials Science and Technology; EPFL; ETH Zürich; European Synchrotron Radiation Facility",
    "source_url": "https://europepmc.org/article/MED/42928641",
    "source_type": "paper",
    "publication_date": "2026-10-25",
    "added_date": "2026-11-15",
    "verification_status": "needs verification",
    "paper_url": "https://doi.org/10.1002/smtd.71142",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Reconstruction agreement with porometry, pycnometry and X-ray tomography for one fiber-mat material system.",
      "unknown": "Admission review pending: primary full results needed to distinguish a specific scientific/application result from general method validation. Existing supported claims may be retained with bounded scope, but do not mark source checked until admission is resolved.",
      "ai_role": "Monocular depth estimation model, trained on macroscale images, plus unsupervised clustering to separate in-plane from subsurface material in FIB-SEM slices.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID%3A42928641&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-ai-ecg-detection-prediction-dilated-cardiomyopathy",
    "title": "AI-enhanced electrocardiography for detection and prediction of dilated cardiomyopathy",
    "summary": "In the journal abstract, an AI-ECG model trained on more than 330,000 ECG–echocardiogram pairs detected the dilated-cardiomyopathy phenotype (AUROC 0.87–0.97) and predicted later disease (C-index 0.80–0.87). In familial cohorts, five-year negative predictive values were 95.5% in Antwerp and 91.5% in Maastricht; positive predictive values were 42.4% and 39.4%. These are diagnostic and risk-prediction findings, not demonstrated clinical benefit.",
    "field": "medicine",
    "ai_technology": "Explainable deep learning model applied to electrocardiography",
    "organization": "Imperial College London; University of Antwerp and Antwerp University Hospital; Maastricht University Medical Center; Beth Israel Deaconess Medical Center; Zhongshan Hospital of Fudan University",
    "source_url": "https://europepmc.org/article/MED/43031488",
    "source_type": "paper",
    "publication_date": "2026-11-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/eurheartj/ehag917",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Validated in UK Biobank, Zhongshan Hospital and Antwerp cohorts, and in two familial cohorts (Antwerp n=166, Maastricht n=177), compared with traditional ECG features and echocardiographic traits.",
      "unknown": "Review limited to the indexed abstract and publication metadata; full-paper methods and additional limitations were not assessed.",
      "ai_role": "Deep learning model trained to detect and predict the echocardiographic DCM phenotype from ECG; explainable AI-ECG outperformed traditional ECG parameters.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1093%2Feurheartj%2Fehag917%22&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-cochleanet-cochlear-connectomics-gene-therapy",
    "title": "CochleaNet: deep learning analysis of cochlear connectomics and gene therapy",
    "summary": "The journal abstract describes CochleaNet, a deep-learning framework for cleared-cochlea light-sheet images, covering hair cells, spiral ganglion neurons, synapses, tonotopy and gene expression. Trained on mouse data, it was applied to gerbil and lower-resolution mouse and marmoset data. The authors say they analyzed neuron types and validated gene-therapy products but the abstract does not identify the biological or therapy-comparison findings. The date represents the 2026-10-23 journal version; an earlier preprint appeared on 2025-11-17.",
    "field": "biology",
    "ai_technology": "CochleaNet: deep learning framework for volumetric image segmentation and analysis",
    "organization": "University Medical Center Göttingen; University of Göttingen",
    "source_url": "https://europepmc.org/article/MED/42920300",
    "source_type": "paper",
    "publication_date": "2026-10-23",
    "added_date": "2026-11-15",
    "verification_status": "needs verification",
    "paper_url": "https://doi.org/10.1016/j.ymthe.2026.10.025",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Manual-analysis and prior-segmentation comparisons; applications to neuron types, gene-therapy products, gerbil cochleae and lower-resolution mouse and marmoset imaging. The abstract does not state the specific biological outcomes.",
      "unknown": "The journal body was not accessible through the publisher (HTTP 403), and the linked preprint full-text API returned HTTP 500. Original preprint abstract adds no specific gene-therapy result. Eligibility remains unresolved until the actual application finding is read.",
      "ai_role": "Deep learning framework that reconstructs the cochlea and segments hair cells, neurons and synapses, and quantifies tonotopy and gene expression.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1016%2Fj.ymthe.2026.10.025%22&resultType=core&format=json",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID%3APPR1120449&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-language-model-species-distribution-alternative-stable-states",
    "title": "A language-model approach to species distribution reveals alternative stable states",
    "summary": "In the journal abstract, a language-model-inspired generative model trained on about 27,000 reef-fish surveys produced alternative coral-associated and macroalgal-associated community configurations under identical environmental conditions. It also inferred core, bridging and satellite species positions and improved occurrence and biomass predictions over XGBoost and MLP baselines. The configurations are model-derived and consistent with previously described ecological regimes; independent field confirmation is not established by the abstract.",
    "field": "biology",
    "ai_technology": "Language-model-inspired generative model for ecological communities",
    "organization": "Université de Montpellier; CNRS; Ifremer; IRD; Inria; University of Tasmania",
    "source_url": "https://europepmc.org/article/MED/42913654",
    "source_type": "paper",
    "publication_date": "2026-10-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cub.2026.09.081",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Fitted on about 27,000 reef fish surveys; predictive skill compared with XGBoost and MLP baselines; multistability interpreted against described ecological regimes.",
      "unknown": "Review limited to the indexed abstract and publication metadata; full-paper methods and additional limitations were not assessed.",
      "ai_role": "Generative language-model-style model that predicts each species and its biomass conditional on previously assembled species, used to generate community configurations.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1016%2Fj.cub.2026.09.081%22&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-data-driven-mxene-discovery",
    "title": "Machine-learning mining uncovers hidden MXenes, including 11 new rare-earth MXenes",
    "summary": "The journal abstract reports that machine-learning-assisted database mining identified 33 previously synthesized but unrecognized MXene candidates. Guided by that search, the researchers synthesized seven multilayer MXenes and realized 11 previously unexplored rare-earth-based MXenes. Experiments and modeling suggested semiconducting behavior and diverse magnetic states. Review is limited to the abstract and publication/affiliation metadata.",
    "field": "materials",
    "ai_technology": "Machine-learning-assisted database mining for MXene candidates",
    "organization": "Linköping University",
    "source_url": "https://doi.org/10.1126/science.aeh5957",
    "source_type": "paper",
    "publication_date": "2026-10-29",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Seven multilayer MXenes were synthesized and 11 rare-earth MXenes realized; experiments and modeling assess electronic and magnetic behavior.",
      "unknown": "Review limited to the indexed abstract and publication metadata; full-paper methods and additional limitations were not assessed.",
      "ai_role": "Machine learning performed database mining to screen candidate MXene compositions.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1126%2Fscience.aeh5957%22&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-previa-ptsd-risk-llm",
    "title": "PREVIA: LLM ensemble predicts PTSD risk from emergency department trauma narratives",
    "summary": "In the journal abstract, PREVIA used trauma narratives from 145 patients at two New York City trauma centers to predict PTSD assessed one month later. Reported AUROC was 78.49% ± 5.75%, with precision 46.76% ± 8.66%. On 30 randomly selected patients, its performance was comparable to one PTSD-specialist emergency clinician. The authors describe these as initial findings requiring larger independent-cohort validation; prospective collection does not establish prospective evaluation of a locked model.",
    "field": "medicine",
    "ai_technology": "PREVIA clinically informed large language model ensemble",
    "organization": "New York University Grossman School of Medicine",
    "source_url": "https://doi.org/10.1038/s41467-026-78768-3",
    "source_type": "paper",
    "publication_date": "2026-11-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "One-month PTSD prediction in 145 trauma patients; comparison with a specialist emergency clinician on 30 selected patients.",
      "unknown": "Abstract and metadata review only. Authors explicitly call for larger independent-cohort validation; full methods and timing of model selection relative to outcomes were not assessed.",
      "ai_role": "LLM ensemble extracts early post-traumatic stress symptoms, risk and protective factors from narratives and produces the risk forecast.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41467-026-78768-3%22&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-virtual-biotech-agents",
    "title": "The Virtual Biotech: a multi-agent AI framework for drug development decisions",
    "summary": "The Virtual Biotech journal abstract reports a human-guided multi-agent analysis of 55,984 trials. Drugs targeting cell-type-specific genes were associated with a 48% greater likelihood of reaching market and 32% fewer adverse events. The system also proposed a lung-cancer strategy and inferred possible reasons for a terminated ulcerative-colitis trial. These are retrospective associations and hypotheses, not demonstrated treatment benefits. The date represents the 2026-10-22 journal version; the linked preprint, including the same reported association, first appeared on 2026-02-23.",
    "field": "medicine",
    "ai_technology": "Virtual Biotech multi-agent AI framework (more than 37,000 agents)",
    "organization": "Stanford University",
    "source_url": "https://doi.org/10.1126/science.aeg6779",
    "source_type": "paper",
    "publication_date": "2026-10-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Retrospective trial-outcome/target-feature analysis across 55,984 trials, plus computational therapeutic and trial-failure hypotheses.",
      "unknown": "Review limited to the indexed abstract and publication metadata; full-paper methods and additional limitations were not assessed.",
      "ai_role": "More than 37,000 AI agents annotated trial outcomes and integrated multimodal evidence to propose and analyze therapeutic decisions.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1126%2Fscience.aeg6779%22&resultType=core&format=json",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=EXT_ID%3APPR1222685&resultType=core&format=json"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-ecg-biomarker-sudden-cardiac-death-deep-learning",
    "title": "Deep learning finds an ECG biomarker for sudden cardiac death",
    "summary": "The Nature paper, published online on 2026-06-24, reports a deep-learning ECG biomarker evaluated retrospectively in Swedish data and external US and Taiwanese datasets. Its high-risk group comprised 2.2% of the sample and had a 7.0% annual sudden-cardiac-death rate, versus 4.6% for reduced LVEF; 86.1% were not flagged by LVEF. Among high-risk patients with defibrillators, sudden cardiac death was 54.4% lower than model-expected, an observational association subject to confounding and not proof of benefit.",
    "field": "medicine",
    "ai_technology": "Deep learning ECG model with a generative ECG waveform model for visualization",
    "organization": "University of California, Berkeley; University of Chicago; Massachusetts Institute of Technology; Halmstad University; Region Halland; University of Gothenburg; Sahlgrenska University Hospital",
    "source_url": "https://doi.org/10.1038/s41586-026-10674-6",
    "source_type": "paper",
    "publication_date": "2026-06-24",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Existing Swedish ECG/death-certificate data, a held-out lockbox created before analysis, external US ventricular-arrhythmia and Taiwanese arrhythmic-arrest evaluations; observational defibrillator associations.",
      "unknown": "Main paper reviewed; supplementary attachments not read. The authors explicitly identify residual confounding and the need for randomized testing of defibrillator benefit in model-high-risk patients.",
      "ai_role": "Deep learning model discovered the predictive ECG waveform pattern; generative model used to visualize it.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41586-026-10674-6%22&resultType=core&format=json",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/PMC13323061/fullTextXML"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-11-bionic-eye-deep-learning-visual-cortex-stimulation",
    "title": "Deep learning controls electrical stimulation patterns in human visual cortex",
    "summary": "The Neuron abstract reports a deep-learning framework trained on trial-resolved recordings from a bidirectional human visual-cortex implant. A learned inverse network and gradient-based optimizer shaped stimulation-evoked population activity, achieved targets at lower currents and elicited more consistent perception than conventional methods. This is an experimental neural-control result, not evidence of restored functional sight. The paper was first published online on 2026-08-07; review here is limited to its abstract and metadata.",
    "field": "medicine",
    "ai_technology": "Deep learning inverse network and gradient-based optimizer for neural stimulation",
    "organization": "University of Zurich; ETH Zurich; University of California, Santa Barbara; Miguel Hernández University",
    "source_url": "https://doi.org/10.1016/j.neuron.2026.07.006",
    "source_type": "paper",
    "publication_date": "2026-08-07",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Within-study human visual-cortex stimulation experiments comparing two learned control approaches with conventional methods; lower currents and more consistent reported percepts.",
      "unknown": "Complete abstract and metadata reviewed. Participant count, implant duration, long-term performance and functional-vision outcomes were not assessed; full-text XML retrieval returned HTTP 500.",
      "ai_role": "Deep learning model trained on trial-resolved neural recordings to synthesize and optimize stimulation patterns.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1016%2Fj.neuron.2026.07.006%22&resultType=core&format=json"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-ml-guided-activation-domain-design",
    "title": "Machine learning guided design of strong transcriptional activation domains",
    "summary": "The preprint abstract reports computational generation of 500,000 candidate transcriptional activation domains and selection of 18,750 high-scoring sequences for yeast experiments using sequence-to-function models including three neural networks. More than 90% of 16,770 well-measured sequences functioned as activation domains. Some designed sequences retained activity in fungi, plants and human cells. Model agreement motivated the MADRIGAL consensus predictor. Review is limited to the abstract and original-server metadata.",
    "field": "biology",
    "ai_technology": "Neural-network sequence-to-function models (three networks); consensus predictor MADRIGAL",
    "organization": "University of California, Berkeley",
    "source_url": "https://europepmc.org/article/PPR/PPR1355735",
    "source_type": "paper",
    "publication_date": "2026-11-02",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.10.30.763198",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "18,750 selected sequences were tested experimentally in yeast; over 90% of 16,770 well-measured sequences functioned as activation domains, and some retained function in fungi, plants and human cells.",
      "unknown": "Review limited to the indexed abstract and publication metadata; full-paper methods and additional limitations were not assessed.",
      "ai_role": "Neural networks scored sequences generated from known activation-domain-like sequences; high-scoring sequences were selected for experimental validation and the consensus model MADRIGAL was developed.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.64898%2F2026.10.30.763198%22&resultType=core&format=json",
        "https://api.biorxiv.org/details/biorxiv/10.64898/2026.10.30.763198"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-cross-state-allosteric-site-discovery",
    "title": "Cross-state representation learning enables prospective discovery of dynamic protein allosteric sites",
    "summary": "The preprint abstract reports that AllositeUltra recovered allosteric sites among its top three predictions in 65.2% of apo structures and 70.6% of substrate-bound structures. More importantly, application to MAGI2 identified a previously unrecognized interdomain pocket, tested by small-molecule screening and X-ray crystallography. This within-study experimental finding qualifies separately from the benchmark. Review is limited to the abstract and original-server metadata.",
    "field": "biology",
    "ai_technology": "AllositeUltra (structure-aware protein language model combined with physical dynamics; residue-guided candidate supplementation)",
    "organization": "Shanghai Jiao Tong University",
    "source_url": "https://europepmc.org/article/PPR/PPR1363697",
    "source_type": "paper",
    "publication_date": "2026-11-10",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.11.09.765238",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The predicted MAGI2 interdomain pocket was validated experimentally through small-molecule screening and X-ray crystallography; benchmark recovery was measured on apo and substrate-bound structures.",
      "unknown": "Review limited to the indexed abstract and publication metadata; full-paper methods and additional limitations were not assessed.",
      "ai_role": "A deep-learning model predicted allosteric sites on apo and substrate-bound structures and was applied prospectively to MAGI2 and proteome-wide.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.64898%2F2026.11.09.765238%22&resultType=core&format=json",
        "https://api.biorxiv.org/details/biorxiv/10.64898/2026.11.09.765238"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-generative-design-de-novo-aav-vectors",
    "title": "Generative design of de novo AAV vectors with unique serological profiles",
    "summary": "The preprint abstract reports a variational autoencoder trained on natural AAV sequences to generate capsids. More than 255 designs were synthesized and tested in vitro. Seven transduced human cells at over 50% efficiency; four of those avoided neutralization by every AAV-reactive human serum sample tested, at the reported NT50 threshold below 1:50. This establishes in-vitro capsid properties, not in-vivo delivery or clinical efficacy. Review is limited to the abstract and original-server metadata.",
    "field": "biology",
    "ai_technology": "Variational autoencoder (generative model of AAV capsid sequences)",
    "organization": "University of Toronto",
    "source_url": "https://europepmc.org/article/PPR/PPR1363818",
    "source_type": "paper",
    "publication_date": "2026-11-09",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.11.06.764881",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "More than 255 AI-generated capsid designs were synthesized and tested in vitro; seven transduced human cells above 50% efficiency and four avoided neutralizing antibodies in all reactive human sera.",
      "unknown": "Review limited to the indexed abstract and publication metadata; full-paper methods and additional limitations were not assessed.",
      "ai_role": "A variational autoencoder learned AAV sequence constraints from natural sequences and generated functional capsid proteins de novo, with a second design campaign exploring uncertain regions.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.64898%2F2026.11.06.764881%22&resultType=core&format=json",
        "https://api.biorxiv.org/details/biorxiv/10.64898/2026.11.06.764881"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-germinal-multi-antibody-antigenic-evolution",
    "title": "Antibody design across observed and forecasted antigenic evolution",
    "summary": "The Germinal Multi preprint abstract reports antibodies designed against several antigen variants at once. Experimental screens of tens of designs per campaign found binders to resistance-associated EGFR variants and neutralizers of historical SARS-CoV-2 variants, including variants withheld from design. Redesign of BD55-1205 restored neutralization of a forecast-guided engineered escape variant while retaining parental-variant activity. The escape result concerns an engineered variant, not a demonstrated future circulating strain. Review is limited to the abstract and original-server metadata.",
    "field": "biology",
    "ai_technology": "Germinal Multi (computational antibody design framework); evolutionary forecasting model",
    "organization": "Stanford University",
    "source_url": "https://europepmc.org/article/PPR/PPR1367523",
    "source_type": "paper",
    "publication_date": "2026-11-13",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.11.10.765734",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Tens of designs per campaign were screened experimentally; designs bound EGFR variants and neutralized SARS-CoV-2 variants, including variants withheld from the design process.",
      "unknown": "Review limited to the indexed abstract and publication metadata; full-paper methods and additional limitations were not assessed.",
      "ai_role": "Germinal Multi computationally optimized one antibody paratope against multiple antigen variants, and an evolutionary forecasting model guided construction of an escape variant and antibody redesign.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.64898%2F2026.11.10.765734%22&resultType=core&format=json",
        "https://api.biorxiv.org/details/biorxiv/10.64898/2026.11.10.765734"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-apollai-prehospital-stroke-detection",
    "title": "APOLLAI, a multimodal AI system, versus EMS personnel for prehospital detection of stroke-related deficits",
    "summary": "The medRxiv abstract reports APOLLAI evaluation on video/audio recordings from 295 suspected-stroke patients at Nîmes University Hospital, against seven-neurologist consensus labels for five NIHSS items. Global sensitivity was 78.5% versus 59.4% for EMS personnel, specificity 87.0% versus 85.9%, and AUC 0.882. These are diagnostic comparisons on recorded patient data; the authors call for prospective validation in real EMS conditions.",
    "field": "medicine",
    "ai_technology": "APOLLAI (computer vision, audio deep learning and SVM fusion classifier)",
    "organization": "AI-Stroke; Nîmes University Hospital",
    "source_url": "https://europepmc.org/article/PPR/PPR1348415",
    "source_type": "paper",
    "publication_date": "2026-10-22",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.10.16.26365784",
    "relevance": {
      "impact": 4,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Recorded cases from 295 suspected-stroke patients; AI and EMS assessments compared with consensus labels for five neurological tasks.",
      "unknown": "Complete abstract and original-server metadata reviewed; full methods, model-selection timing and external EMS generalizability were not assessed. Authors explicitly call for real-world prospective validation.",
      "ai_role": "A multimodal machine-learning system detected stroke-related neurological deficits from video and audio recordings and was compared with clinicians' assessments.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.64898%2F2026.10.16.26365784%22&resultType=core&format=json",
        "https://api.biorxiv.org/details/medrxiv/10.64898/2026.10.16.26365784"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2026-10-agent-perrhenate-deoxydehydration-water-loss",
    "title": "AI agent maps proton-gated water loss in perrhenate-catalysed deoxydehydration",
    "summary": "The paper reports a computational mechanism for perrhenate-catalysed deoxydehydration: proton-assisted water loss enables ring closure and alcohol reduction. Under human-defined hypotheses, a Claude Code agent located transition states, which the authors validated. Coupled-cluster-anchored microkinetic modeling predicted a rate increase of about eight orders of magnitude over the best proton-free pathway. At matched conversion conditions, modeled barriers were consistent with the 31.5–34.0 kcal/mol window inferred from previously reported experimental turnover; this was not a new wet-lab demonstration.",
    "field": "chemistry",
    "ai_technology": "Claude Code LLM agent for human-directed reaction-network and transition-state search, combined with electronic-structure calculations and microkinetic modeling",
    "organization": "KU Leuven",
    "source_url": "https://doi.org/10.1039/d6sc06865k",
    "source_type": "paper",
    "publication_date": "2026-10-27",
    "added_date": "2026-11-15",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Agent-located transition states verified by the authors; computed energetics, reaction-network and microkinetic consistency checks; comparisons to existing turnover-derived barrier estimates.",
      "unknown": "Main paper reviewed; supplementary files were not read. The mechanism and rate changes are predictions compared with prior experimental kinetics, not a new experimental confirmation.",
      "ai_role": "Claude Code carried out calculation/search/triage workflows for author-specified mechanistic hypotheses; authors validated the located stationary points.",
      "evidence_urls": [
        "https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1039%2Fd6sc06865k%22&resultType=core&format=json",
        "https://www.ebi.ac.uk/europepmc/webservices/rest/PMC13720772/fullTextXML"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-ori-engineers-active-and-thermotolerant-enzymes",
    "title": "ORI engineers active lysozymes, heat-tolerant chitinases and dual-activity enzymes",
    "summary": "Tencent researchers used ontology-conditioned generation and experimental feedback to engineer enzymes. In the reported assays, lysozyme TX-RL15 exceeded the earlier TX-L6 design by 66-fold and hen egg-white lysozyme by over two orders of magnitude. Chitinase TX-SC2 retained activity after incubation at 85 °C; selected proteins combined lysozyme and chitinase activity. These are bounded biochemical results. The date represents initial journal publication; the version of record followed on May 7, 2026.",
    "field": "biology",
    "ai_technology": "ORI; protein generation model (PGM); unified sequence model (USM)",
    "organization": "Tencent AI for Life Sciences Lab",
    "source_url": "https://www.nature.com/articles/s41467-026-69855-6",
    "source_type": "paper",
    "publication_date": "2026-03-19",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Expression, enzyme-activity and heat-treatment assays on selected generated proteins.",
      "unknown": "Separate supplements and independent replication not assessed. Assay-specific advantages do not establish industrial or clinical outcomes.",
      "ai_role": "Ontology-conditioned generation and reinforcement learning from experimental feedback.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-69855-6?error=cookies_not_supported"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-09-ertransformer-generated-tcr-candidates-activate-t-cells",
    "title": "ERTransformer-generated TCR candidates activate T cells in defined receptor contexts",
    "summary": "ERTransformer generated epitope-conditioned TCR β-chain CDR3 candidates. Flow-cytometry experiments found selected artificial TCRs induced activation comparable to or greater than natural controls in defined TCR/pMHC contexts. This represents the September 2026 journal report; an ERTransformer precursor was posted October 20, 2023. The earlier report does not establish that these journal wet-lab results were already public.",
    "field": "biology",
    "ai_technology": "ERTransformer; EpitopeBERT; ReceptorBERT",
    "organization": "Tencent AI for Life Sciences Lab; City University of Hong Kong; The University of Hong Kong; Xijing Hospital of Digestive Diseases; Harvard Medical School, Harvard University; Monash University",
    "source_url": "https://link.springer.com/article/10.1186/s13059-026-04266-w",
    "source_type": "paper",
    "publication_date": "2026-09-02",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected generated candidates tested for T-cell activation by flow cytometry.",
      "unknown": "Full experimental methods and supplements not assessed: publisher rendering exposed abstract and metadata. General specificity, safety and clinical outcomes are not established by this review.",
      "ai_role": "Generates TCR β-chain CDR3 candidates conditioned on epitopes.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s13059-026-04266-w",
        "https://www.biorxiv.org/content/10.1101/2023.10.18.562845v1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-hudiff-humanized-antibodies-retain-binding",
    "title": "HuDiff humanizes antibodies and nanobodies while retaining measured binding",
    "summary": "HuDiff generated humanized variants of a murine SARS-CoV-2 RBD antibody and two alpaca nanobodies. Bio-layer interferometry measured 0.15 nM affinity for the best humanized antibody versus 0.12 nM for its parent; a humanized nanobody measured 2.52 nM versus 5.47 nM. The journal abstract also reports viral neutralization. The October 2025 journal date follows an October 25, 2024 precursor describing one antibody and one nanobody.",
    "field": "biology",
    "ai_technology": "HuDiff-Ab; HuDiff-Nb",
    "organization": "School of Basic Medical Sciences, Lanzhou University; Tencent AI Lab; Fudan University; Peking University",
    "source_url": "https://www.nature.com/articles/s42256-025-01120-9",
    "source_type": "paper",
    "publication_date": "2025-10-01",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Newly humanized molecules underwent binding assays; journal abstract reports neutralization.",
      "unknown": "Main journal methods are subscription content and were not assessed. Humanness scores do not establish reduced human immunogenicity or clinical safety.",
      "ai_role": "Adaptive autoregressive diffusion generates humanized framework sequences from CDR inputs.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42256-025-01120-9?error=cookies_not_supported",
        "https://www.biorxiv.org/content/10.1101/2024.10.22.619416v1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-tfold-designs-experimentally-binding-antibodies",
    "title": "tFold designs binding antibodies against four defined antigens",
    "summary": "Tencent researchers generated antibodies against influenza hemagglutinin, PD-1, PD-L1 and SARS-CoV-2 RBD, followed by phage-display screening and SPR characterization. Anti-SC2RBD-H330 showed neutralization IC50 of 0.42 nM. Screening remained extensive, and some confident epitope predictions failed competition tests. The date represents initial journal publication, preceding the January 19, 2026 version of record; a tFold precursor was posted February 8, 2024.",
    "field": "biology",
    "ai_technology": "tFold System; tFold-Ab; tFold-Ag; AlphaFold-Multimer",
    "organization": "Tencent AI for Life Sciences Lab",
    "source_url": "https://www.nature.com/articles/s41467-025-67361-9",
    "source_type": "paper",
    "publication_date": "2025-12-10",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Binding and competition assays on generated antibodies; RBD-antibody neutralization.",
      "unknown": "Separate supplements and independent replication not assessed. First-public dates of individual journal experiments were not established by precursor version comparison.",
      "ai_role": "Generates CDR sequences and predicts complexes for structure-guided filtering.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-67361-9?error=cookies_not_supported",
        "https://www.biorxiv.org/content/10.1101/2024.02.05.578892v1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "field": "biology",
    "source_type": "paper",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "id": "2023-07-ai-assisted-macaque-cortex-atlas-reveals-spatial-cell-type-organization",
    "title": "AI-assisted macaque cortex atlas reveals regional cell-type organization",
    "summary": "AI-assisted segmentation and Spatial-ID annotation helped map 264 transcriptome-defined cell types across the macaque cortex. Regional cell-type composition tracked cortical hierarchy, and cross-species comparisons identified primate-specific types enriched in layer 4. This is the July 12, 2023 journal version; the atlas is observational, uses male monkeys, and has limited transcript capture per spatial cell.",
    "ai_technology": "AI-assisted cell segmentation; Spatial-ID (deep neural network and graph convolutional network)",
    "organization": "Chinese Academy of Sciences; BGI-Research; Tencent AI Lab",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S0092867423006797",
    "publication_date": "2023-07-12",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "New macaque tissue measurements and spatial maps were compared across sections and biological replicates; this supports a bounded anatomical result, not causal brain function.",
      "unknown": "Not assessed: independent replication and exhaustive supplementary material. The paper identifies limited transcript sensitivity and absence of female subjects.",
      "ai_role": "Image segmentation identifies cells; Spatial-ID trains a DNN on nucleus profiles and a GCN incorporating spatial neighborhoods to assign spatial cell labels.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S0092867423006797",
        "https://github.com/TencentAILabHealthcare/Cell_Segmentation_Fluorescence",
        "https://bcas.cas.cn/issue/2023_3/202311/P020231112626745733563.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "field": "biology",
    "source_type": "paper",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "id": "2024-09-ai-assisted-cross-species-cerebellar-atlases-reveal-primate-specific-organization",
    "title": "AI-assisted cerebellar atlases reveal primate-specific cellular organization",
    "summary": "Deep-learning segmentation and spatial annotation supported cerebellar atlases of macaques, marmosets and mice, revealing primate-specific cell subtypes and regional gene-expression differences. Transcriptomic gradients corresponded with awake functional-connectivity patterns. This represents the September 27, 2024 journal version; an institutional announcement gives September 26, and the observed associations do not establish causal functions.",
    "ai_technology": "Deep-learning cell segmentation; STAGATE; DestVI",
    "organization": "BGI-Research; Chinese Academy of Sciences; Erasmus MC; Tencent AI Lab",
    "source_url": "https://www.science.org/doi/10.1126/science.ado3927",
    "publication_date": "2024-09-27",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "New spatial tissue measurements, biological replicates and comparisons with awake fMRI supported the reported anatomical associations within this study.",
      "unknown": "Not assessed: exhaustive supplements, earlier-preprint chronology and independent confirmation. Functional mechanisms are not established by spatial correlations.",
      "ai_role": "Deep-learning models segment cells; STAGATE clusters spatial domains and DestVI assigns cell-type probabilities.",
      "evidence_urls": [
        "https://pure.eur.nl/ws/files/169697439/Cross-species_single-cell_spatial_transcriptomic_atlases_of_the_cerebellar_cortex.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "field": "biology",
    "source_type": "paper",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "id": "2025-04-ai-assisted-macaque-claustrum-atlas-links-cell-types-and-projection-zones",
    "title": "AI-assisted macaque claustrum atlas links cell types with projection zones",
    "summary": "Learned segmentation and Spatial-ID annotation helped connect macaque claustral cell-type distributions with four projection zones identified through tracer experiments. Different glutamatergic compositions preferentially connected to different brain regions. This is the April 3, 2025 journal version, preceded by a related 2024 conference abstract; comparisons reuse published macaque-cortex data, and the findings do not establish a causal account of consciousness.",
    "ai_technology": "Mask R-CNN; Spatial-ID; nnUNet",
    "organization": "Chinese Academy of Sciences; Stem Cell and Brain Research Institute, Inserm; BGI-Research; Tencent AI for Life Sciences Lab",
    "source_url": "https://www.cell.com/cell/fulltext/S0092-8674(25)00273-9",
    "publication_date": "2025-04-03",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Spatial transcriptomics and retrograde tracing directly supported the cell-type/projection-zone relationship within the discovery study.",
      "unknown": "Not assessed: independent replication, complete supplements and exact first posting day of the earlier conference abstract.",
      "ai_role": "Mask R-CNN segments cells; Spatial-ID trains DNN and GCN models for spatial labels; expert-corrected nnUNet segments claustral boundaries.",
      "evidence_urls": [
        "https://hal.science/hal-05210393v1/file/Lei%20et%20al%202025%20Cell.pdf",
        "https://pubmed.ncbi.nlm.nih.gov/40185102/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-spatial-dc-pancreatic-cancer-cell-interactions",
    "title": "Spatial-DC maps a candidate fibroblast–cancer signaling interaction in pancreatic tumors",
    "summary": "Spatial-DC reconstructed cell-type-resolved spatial proteomes and suggested COL8A1–ITGA2 communication between cancer-associated fibroblasts and pancreatic cancer cells. Multiplexed immunohistochemistry supported the spatial predictions; human protein co-expression was associated with survival. The pairing was already known: the result concerns its inferred spatial deployment, not a new receptor pair or established causal mechanism. This record uses the May 27, 2026 journal version; the result already appeared in the December 20, 2024 preprint.",
    "field": "biology",
    "ai_technology": "Spatial-DC; graph-enhanced transfer learning",
    "organization": "Southern University of Science and Technology; Tencent AI for Life Sciences Laboratory",
    "source_url": "https://www.sciencedirect.com/science/article/abs/pii/S2405471226000943",
    "source_type": "paper",
    "publication_date": "2026-05-27",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cels.2026.101612",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Retrospective spatial proteomic reconstruction and human survival association. Same-study multiplexed immunohistochemistry supports spatial distributions and co-expression, not causal COL8A1–ITGA2 communication.",
      "unknown": "No causal signaling perturbation, therapeutic benefit or independent replication is established in the reviewed material. Current review read indexed publisher introduction and NLM abstract/metadata; full journal methods were inaccessible. Earlier primary preprint XML was read in the project audit on November 16, 2026; revised journal quantitative details were not assessed.",
      "ai_role": "Graph-enhanced transfer learning deconvolved mixed spots and reconstructed cell-type-resolved protein profiles used to infer spatial interactions.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/abs/pii/S2405471226000943",
        "https://pubmed.ncbi.nlm.nih.gov/42202779/"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-12-ai-assisted-alzheimer-hippocampus-plaque-associated-cell-map",
    "title": "AI-assisted hippocampal atlas maps Alzheimer-associated cellular changes around amyloid plaques",
    "summary": "In a six-donor postmortem hippocampal study, neural cell segmentation and Spatial-ID annotation helped map Alzheimer-associated cell distributions and gene-expression changes. Immunofluorescence supported the inferred accumulation patterns of microglia and astrocytes around amyloid-beta plaques. These are observational associations, not causal effects or clinical benefit. This record describes the December 7, 2023 preprint; the later Neuron paper was published online March 31, 2025 with an expanded cohort, whose counts and final methods are not pooled here.",
    "field": "biology",
    "ai_technology": "StereoCell/ESPANet cell segmentation; Spatial-ID deep neural network and graph convolutional network",
    "organization": "Zhejiang University; BGI Research; Tencent AI Lab; University of Chinese Academy of Sciences",
    "source_url": "https://www.medrxiv.org/content/10.1101/2023.12.06.23299593v1",
    "source_type": "paper",
    "publication_date": "2023-12-07",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://www.medrxiv.org/content/10.1101/2023.12.06.23299593v1.full.pdf",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Spatial transcriptomics and single-nucleus sequencing in three AD and three control donor brains; immunofluorescence checked the plaque-associated glial distribution pattern within the same study.",
      "unknown": "Postmortem observational results do not establish disease causation. No independent replication was assessed. Preprint methods retain an unfinished training-epoch placeholder. The reviewed claims and AI architectures refer to the primary preprint, not authenticated final journal methods; candidate blood biomarkers and diagnostic efficacy are outside this record's represented claim.",
      "ai_role": "Trained neural segmentation delineated cells; Spatial-ID used a DNN and GCN to assign spatial cell types, enabling plaque-associated cell-distribution analysis.",
      "evidence_urls": [
        "https://www.medrxiv.org/content/10.1101/2023.12.06.23299593v1.full.pdf",
        "https://www.medrxiv.org/content/10.1101/2023.12.06.23299593v1.full",
        "https://pubmed.ncbi.nlm.nih.gov/40168986/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-astra-assisted-gauge-invariant-lattice-fermion-measure",
    "title": "GPT-6 Astra assists a gauge-invariant lattice chiral-fermion measure construction",
    "summary": "Nathaniel Craig proposes a smooth, exponentially local, gauge-invariant fermion measure on admissible four-dimensional lattices, with restrictions covering the Standard Model gauge groups. GPT-6 Astra helped develop the refinement construction and led the detailed proof. The supplementary report states that Lean kernel replay accepted the encoded measure-existence result and supporting inputs under their recorded assumptions. This is a candidate finite-lattice definition; reflection positivity, practical simulation and an interacting continuum theory remain outside the established claim. The October 6, 2026 preprint preceded the November 15 institutional announcement.",
    "field": "physics",
    "ai_technology": "GPT-6 Astra; Lean 4 formalization",
    "organization": "University of California, Santa Barbara; OpenAI",
    "source_url": "https://arxiv.org/abs/2610.09093",
    "source_type": "paper",
    "publication_date": "2026-10-06",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/html/2610.09093v1",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Source-reported Lean kernel replay checked the concrete companion measure-existence theorem, its analytic/geometric inputs and Standard Model restrictions under encoded definitions and assumptions. Same-study mathematical and two-dimensional computational checks are also reported.",
      "unknown": "Review checked the preprint verification/conclusion sections and formal-verification supplement, not an independently executed Lean build. Formal-to-source correspondence was reviewed by AI agents. A separate frozen-reference comparator failed on generated proof names; the report distinguishes this from successful kernel replay. Kernel correctness is relative to the encoded claim and axioms; it does not establish reflection positivity, a physical continuum limit, independent scientific confirmation or practical usefulness.",
      "ai_role": "Craig supplied the refinement strategy; Astra developed key construction steps, led proof development and carried out proposed verification work.",
      "evidence_urls": [
        "https://arxiv.org/html/2610.09093v1",
        "https://arxiv.org/src/2610.09093v1/anc/lean-verification-report.pdf",
        "https://arxiv.org/src/2610.09093v1/anc/mathematical-companion.pdf",
        "https://openai.com/index/step-toward-defining-the-standard-model/"
      ],
      "test_type": "prospective or independent check"
    }
  },
  {
    "id": "2024-10-robotic-learning-identifies-nine-polyoxometalate-based-frameworks",
    "title": "Robotic learning identifies nine polyoxometalate-based frameworks",
    "summary": "A robot and an XGBoost classifier guided crystallization experiments that produced nine reported polyoxometalate-scaffolding metal-organic frameworks. Researchers defined the chemistry and performed structural and electrochemical characterization. The measured electron-transfer behavior does not establish practical energy-storage deployment.",
    "field": "materials",
    "ai_technology": "XGBoost with uncertainty-feedback retraining",
    "organization": "University of Glasgow",
    "source_url": "https://eprints.gla.ac.uk/336574/",
    "source_type": "paper",
    "publication_date": "2024-10-09",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/jacs.4c09553",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Nine frameworks were synthesized with reported repeatability.",
      "unknown": "Independent replication, exhaustive structural prior art and device performance were not assessed.",
      "ai_role": "XGBoost with uncertainty-feedback retraining supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://eprints.gla.ac.uk/336574/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Institutional publication record and primary abstract; full paper and supplements not reread."
    }
  },
  {
    "id": "2024-07-bayesian-optimization-improves-printed-bisbte-thermoelectrics",
    "title": "Bayesian optimization improves printed BiSbTe thermoelectrics",
    "summary": "Learned models guided ink formulation and printing settings for known BiSbTe-based material. The resulting printed samples reached a reported room-temperature zT of 1.3. This is an experimental processing result, rather than discovery of a new chemical composition or demonstrated deployed-device efficiency. The result was public in the July 2024 preprint before the journal version.",
    "field": "materials",
    "ai_technology": "Gaussian-process regression and constrained Bayesian optimization",
    "organization": "University of Notre Dame; Marquette University",
    "source_url": "https://arxiv.org/abs/2407.01145",
    "source_type": "paper",
    "publication_date": "2024-07-01",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D4TA03062A",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Printed samples were characterized after model-guided processing.",
      "unknown": "Generality, industrial manufacturing and the contribution of each downstream processing step remain unestablished in this review.",
      "ai_role": "Gaussian-process regression and constrained Bayesian optimization supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://arxiv.org/abs/2407.01145",
        "https://arxiv.org/pdf/2407.01145"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary preprint abstract, submission history and selected PDF passages on model and affiliations."
    }
  },
  {
    "id": "2024-06-bayesian-search-identifies-organic-metallophotocatalyst-formulations",
    "title": "Bayesian search identifies organic metallophotocatalyst formulations",
    "summary": "Gaussian-process models and chemist decisions guided synthesis and testing of organic photocatalysts for a specified cross-coupling reaction. A catalyst search reached 67% yield before reaction-condition optimization. This establishes a bounded formulation result for the studied reaction, not discovery of the reaction or general superiority to iridium catalysts.",
    "field": "chemistry",
    "ai_technology": "Gaussian-process Bayesian optimization",
    "organization": "Zhejiang Normal University; University of Liverpool; University of Birmingham; IBM Research",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11321994/",
    "source_type": "paper",
    "publication_date": "2024-06-11",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41557-024-01546-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Candidate catalysts were synthesized and reaction yields measured in repeated experiments.",
      "unknown": "General substrate scope, independent replication and exhaustive molecular prior art were not assessed.",
      "ai_role": "Gaussian-process Bayesian optimization supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11321994/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article abstract, selected Results, computational Methods and publication metadata; supplements not reviewed."
    }
  },
  {
    "id": "2022-04-machine-guided-microscopy-characterizes-an-atypical-ataxia-presentation",
    "title": "Machine-guided microscopy characterizes an atypical ataxia presentation",
    "summary": "Unsupervised image analysis detected cellular patterns not apparent by inspection in an investigation of one family with an atypical ataxia presentation. Additional cell experiments found altered ataxin-7 behavior and mitochondrial bioenergetic deficits. The proposed combined role of ATXN7 and TOP1MT variants remains a hypothesis: unaffected parents carried individual variants, and fly models did not reproduce the patient phenotype.",
    "field": "medicine",
    "ai_technology": "PhenoRipper unsupervised image-pattern analysis",
    "organization": "McMaster University; University of British Columbia; University of Toronto; University of Pittsburgh",
    "source_url": "https://www.nature.com/articles/s41439-022-00188-8",
    "source_type": "paper",
    "publication_date": "2022-04-14",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41439-022-00188-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Patient-derived cells, functional measurements and fly models were examined.",
      "unknown": "Clinical causality, variant interaction and diagnostic utility are not established by this small study.",
      "ai_role": "PhenoRipper unsupervised image-pattern analysis supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41439-022-00188-8"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher abstract, selected image-analysis Methods, Results, Discussion, dates and affiliations; supplement not reviewed."
    }
  },
  {
    "id": "2021-12-learned-soundscape-features-predict-tropical-forest-species-occurrence",
    "title": "Learned soundscape features predict tropical-forest species occurrence",
    "summary": "Learned soundscape features predicted occurrence for 34 of 39 surveyed bird and herpetofaunal species across a forest-degradation gradient in Sabah. Predictions used the broader acoustic environment rather than direct identification of every species call. Site-level evaluation supports a bounded monitoring result, not proof of causal habitat quality or a replacement for species surveys.",
    "field": "biology",
    "ai_technology": "Convolutional audio embeddings and Gaussian-mixture occurrence models",
    "organization": "Norwegian Institute for Nature Research; Imperial College London",
    "source_url": "https://nsojournals.onlinelibrary.wiley.com/doi/10.1111/oik.08525",
    "source_type": "paper",
    "publication_date": "2021-12-08",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1111/oik.08525",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Occurrence predictions were compared with field point counts using cross-validation.",
      "unknown": "Transfer to other ecosystems and causal interpretation are not established by this study.",
      "ai_role": "Convolutional audio embeddings and Gaussian-mixture occurrence models supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://nsojournals.onlinelibrary.wiley.com/doi/10.1111/oik.08525"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher metadata, affiliations, abstract and selected Methods/Results; supplements not reviewed."
    }
  },
  {
    "id": "2022-04-random-forest-selection-links-a-food-compound-to-cox-1-inhibition",
    "title": "Random-forest selection links a food compound to COX-1 inhibition",
    "summary": "A random-forest model prioritized 5-methoxysalicylic acid for a COX-1 enzyme assay. It showed about 40% inhibition at 100 micromolar, whereas a higher-ranked single-fingerprint comparator did not inhibit at that concentration. This is a modest in-vitro target result, not evidence of dietary, therapeutic or clinical benefit.",
    "field": "chemistry",
    "ai_technology": "Random-forest classification of drug-target similarity",
    "organization": "European Molecular Biology Laboratory; University of Cambridge",
    "source_url": "https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1010029",
    "source_type": "paper",
    "publication_date": "2022-04-25",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pcbi.1010029",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An enzymatic assay compared the selected compound with positive and negative controls.",
      "unknown": "Cellular target engagement, efficacy and a robust compound IC50 are not established here.",
      "ai_role": "Random-forest classification of drug-target similarity supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1010029"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher abstract, selected model training and assay Results, publication date and affiliations; supplementary data not reviewed."
    }
  },
  {
    "id": "2021-08-learned-models-reveal-nonmonotonic-bacterial-patterns-in-boreal-lakes",
    "title": "Learned models reveal nonmonotonic bacterial patterns in boreal lakes",
    "summary": "Models fitted to observations from 70 Scandinavian lakes identified nonmonotonic associations between browning indicators and bacterial community diversity. XGBoost explained up to about 45% of community-composition variation in the analyzed data. The fitted ridges are not demonstrated causal tipping points; translating this spatial gradient into future temporal trajectories requires further validation.",
    "field": "biology",
    "ai_technology": "XGBoost, random forests and feed-forward neural networks",
    "organization": "University of Oslo; Université Savoie Mont Blanc; INRAE",
    "source_url": "https://www.nature.com/articles/s43705-021-00038-8",
    "source_type": "paper",
    "publication_date": "2021-08-16",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43705-021-00038-8",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Models analyzed field-sampled microbial communities with held-out evaluation.",
      "unknown": "Covariation, causal mechanisms and long-term temporal predictions remain unresolved.",
      "ai_role": "XGBoost, random forests and feed-forward neural networks supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43705-021-00038-8",
        "https://www.nature.com/articles/s43705-021-00038-8.pdf"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher PDF abstract and selected Methods/Results/limitations; indexed publisher publication metadata. Supplements not reviewed."
    }
  },
  {
    "id": "2022-01-learned-subgroup-rules-guide-higher-component-oxidation-catalysts",
    "title": "Learned subgroup rules guide higher-component oxidation catalysts",
    "summary": "Rules learned from high-throughput catalyst measurements guided tests of more complex four- and five-component materials for propylene oxidation. Selected materials reached approximately 60% oxygenate yield in laboratory tests. The learned descriptor ranges are associative design guides, not proven atomic mechanisms or universal industrial performance rules.",
    "field": "chemistry",
    "ai_technology": "Subgroup-discovery rule learning",
    "organization": "Fritz Haber Institute of the Max Planck Society; Humboldt University of Berlin; BASF; hte",
    "source_url": "https://doi.org/10.1021/acscatal.1c04793",
    "source_type": "paper",
    "publication_date": "2022-01-31",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "More complex catalyst compositions were tested against rules learned from three-component measurements.",
      "unknown": "Independent replication and industrial durability were not assessed; temperature rules did not transfer unchanged.",
      "ai_role": "Subgroup-discovery rule learning supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://doi.org/10.1021/acscatal.1c04793",
        "https://pure.mpg.de/pubman/item/item_3369647_4/component/file_3370185/acscatal.1c04793.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher text and deposited publisher PDF abstract and selected experimental-validation Results; supplements not reviewed."
    }
  },
  {
    "id": "2023-09-ai-assisted-protein-family-analysis-identifies-a-tested-toxin-antitoxin-superfamily",
    "title": "AI-assisted protein-family analysis identifies a tested toxin-antitoxin superfamily",
    "summary": "AI-predicted structural information helped researchers prioritize previously poorly characterized protein families. Experiments supported a translation-targeting toxin-antitoxin superfamily, TumE–TumA. This is one tested family-level result; the broader proposed families and predicted fold are not hundreds of separately validated discoveries. Native-organism function and general experimental validation across the predicted families remain unestablished.",
    "field": "biology",
    "ai_technology": "AlphaFold structural models and DeepFRI functional prediction",
    "organization": "University of Basel; SIB Swiss Institute of Bioinformatics",
    "source_url": "https://www.nature.com/articles/s41586-023-06622-3",
    "source_type": "paper",
    "publication_date": "2023-09-13",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-023-06622-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected proteins received laboratory functional validation.",
      "unknown": "Independent replication and function of the broader computational candidate set were not assessed.",
      "ai_role": "AlphaFold structural models and DeepFRI functional prediction supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-023-06622-3",
        "https://edoc.unibas.ch/entities/publication/5b5fe8a1-910e-4fa9-81a1-c5bceb0297be",
        "https://www.sib.swiss/news/artificial-intelligence-tools-shed-light-on-millions-of-proteins"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher Methods and publication metadata, institutional abstract and announcing-institution text; full PDF and supplements not freshly recovered."
    }
  },
  {
    "id": "2022-10-deep-learning-control-links-single-cell-expression-to-growth-responses",
    "title": "Deep-learning control links single-cell expression to growth responses",
    "summary": "A learned predictive controller enabled high-throughput measurements linking gene-expression patterns with functional outcomes in individual engineered bacterial cells. The study exposed heterogeneous growth responses under antibiotic challenge. The original result was public in the October 2022 preprint and later appeared in a 2024 journal article. This is a laboratory model-system study, not a clinical resistance result or validated intervention.",
    "field": "biology",
    "ai_technology": "LSTM encoder-decoder predictive control",
    "organization": "Boston University",
    "source_url": "https://www.biorxiv.org/content/10.1101/2022.10.28.514305v1",
    "source_type": "paper",
    "publication_date": "2022-10-31",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-46361-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Researchers measured single-cell expression and growth behavior under controlled comparisons.",
      "unknown": "Generalization beyond the engineered system and correspondence of every quantitative journal endpoint to v1 were not established.",
      "ai_role": "LSTM encoder-decoder predictive control supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.1101/2022.10.28.514305v1",
        "https://api.biorxiv.org/details/biorxiv/10.1101/2022.10.28.514305",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10923782/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary bioRxiv API v1 metadata/abstract; journal PMC abstract and selected model Methods and functional Results. No v1 quantitative endpoints transferred."
    }
  },
  {
    "id": "2022-10-fleet-guided-observing-campaign-classifies-superluminous-supernovae",
    "title": "FLEET-guided observing campaign classifies superluminous supernovae",
    "summary": "FLEET used transient light curves and host-galaxy features to prioritize telescope follow-up. Its first two-year campaign spectroscopically classified 21 superluminous supernovae. This is one campaign-level result, not 21 separately counted discoveries or proof that FLEET first detected every object. The result was public in the October 2022 preprint before the 2023 journal article.",
    "field": "physics",
    "ai_technology": "FLEET machine-learning classifier",
    "organization": "Space Telescope Science Institute; Center for Astrophysics | Harvard & Smithsonian",
    "source_url": "https://arxiv.org/abs/2210.10811",
    "source_type": "paper",
    "publication_date": "2022-10-19",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-4357/acc536",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Spectra supported classifications in the AI-prioritized follow-up sample.",
      "unknown": "First-discovery attribution for individual objects and completeness beyond the studied alerts were not established.",
      "ai_role": "FLEET machine-learning classifier supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://arxiv.org/abs/2210.10811",
        "https://pureadmin.qub.ac.uk/ws/portalfiles/portal/555515225/Fleet.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary preprint abstract and submission history; indexed institutional publisher-PDF header and abstract for affiliations and journal chronology."
    }
  },
  {
    "id": "2024-04-laiss-assisted-review-reclassifies-anomalous-ztf-transients",
    "title": "LAISS-assisted review reclassifies anomalous ZTF transients",
    "summary": "Expert review of 455 previously observed transients changed 16 normal labels to anomalous; the random-forest model flagged 11 of these. Spectra, light curves and host context supported the reclassifications. This is a retrospective study-level result rather than a count of newly observed events or newly established physical classes. Human review and similarity searches also contributed, so the entire retrospective sample is not attributed to the classifier.",
    "field": "physics",
    "ai_technology": "LAISS random-forest anomaly classifier and similarity search",
    "organization": "University of Illinois Urbana-Champaign; Pacific Northwest National Laboratory",
    "source_url": "https://arxiv.org/abs/2404.01235v1",
    "source_type": "paper",
    "publication_date": "2024-04-01",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-4357/ad6869",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Existing photometry and classification spectra were re-examined to support revised labels.",
      "unknown": "Anomaly categories depend on expert judgments; universal first-discovery priority and new physical mechanisms are not established.",
      "ai_role": "LAISS random-forest anomaly classifier and similarity search supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://arxiv.org/abs/2404.01235v1",
        "https://arxiv.org/html/2404.01235v1"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary arXiv v1 HTML abstract, selected classifier/reclassification Results and affiliations; version history checked."
    }
  },
  {
    "id": "2020-04-adaptive-cloud-laboratory-reveals-optical-chirality-in-perovskite-nanocrystals",
    "title": "Adaptive cloud laboratory reveals optical chirality in perovskite nanocrystals",
    "summary": "An automated laboratory used measured feedback and fitted local optimization models to search for optically active CsPbBr3 nanocrystals. Spectroscopy measured circular dichroism and imaging informed a proposed defect-related explanation. The composition was known; the contribution is an experimentally characterized optical property and its optimization. The proposed mechanism and future device applications are not independently established.",
    "field": "materials",
    "ai_technology": "SNOBFIT adaptive optimization with locally fitted quadratic models",
    "organization": "Chinese University of Hong Kong, Shenzhen; Shenzhen University; Hubei University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC7184584/",
    "source_type": "paper",
    "publication_date": "2020-04-27",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-020-15728-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Nanocrystal optical properties and structures were measured during adaptive experiments.",
      "unknown": "Independent replication, isolated causal mechanisms and device-scale utility were not assessed.",
      "ai_role": "SNOBFIT adaptive optimization with locally fitted quadratic models supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC7184584/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article abstract, selected Results and optimization Methods, affiliations and publication metadata; supplements not reviewed."
    }
  },
  {
    "id": "2022-10-generative-learning-selects-experimentally-active-egfr-inhibitor-analogs",
    "title": "Generative learning selects experimentally active EGFR inhibitor analogs",
    "summary": "A recurrent generative model refined by reinforcement learning selected compounds for an EGFR enzyme assay. Four of 15 predicted-active compounds were active, including two with nanomolar inhibition; five predicted-inactive controls were inactive. The experimentally tested compounds were commercially obtainable analogs of a known privileged scaffold, not a wholly new drug scaffold or demonstrated cellular therapy.",
    "field": "chemistry",
    "ai_technology": "Recurrent neural generator with reinforcement learning",
    "organization": "Carnegie Mellon University; University of North Carolina at Chapel Hill; Enamine",
    "source_url": "https://www.nature.com/articles/s42004-022-00733-0",
    "source_type": "paper",
    "publication_date": "2022-10-18",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42004-022-00733-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An independent contract assay tested selected compounds and negative controls.",
      "unknown": "Cellular efficacy, selectivity, safety and novel-scaffold discovery are not established.",
      "ai_role": "Recurrent neural generator with reinforcement learning supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42004-022-00733-0",
        "https://www.nature.com/articles/s42004-022-00733-0.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher PDF abstract, selected experimental Results and affiliations; primary Crossref online-date metadata. Supplement not reviewed."
    }
  },
  {
    "id": "2023-03-acoustic-learning-maps-spider-monkey-associations-with-forest-disturbance",
    "title": "Acoustic learning maps spider-monkey associations with forest disturbance",
    "summary": "A learned call detector, manual confirmation and occupancy modeling analyzed recordings from 341 Costa Rican sites. In this sample, Geoffroy's spider monkey was detected only above 80% forest cover and away from nearby primary paved roads. These are regional observational associations, not causal or universal habitat thresholds; imperfect detection and the focus on one call type limit inference.",
    "field": "biology",
    "ai_technology": "Attention-equipped residual convolutional call detector",
    "organization": "Imperial College London; Osa Conservation",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10015327/",
    "source_type": "paper",
    "publication_date": "2023-03-15",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1098/rspb.2022.2473",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Model positives from field recordings were manually checked and linked to landscape observations.",
      "unknown": "Causality, independent ground-truth survey agreement and transfer to other regions were not established.",
      "ai_role": "Attention-equipped residual convolutional call detector supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10015327/"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article dates, affiliations and selected detector Methods, Results and limitations; supplement not reviewed."
    }
  },
  {
    "id": "2021-11-machine-learning-search-surfaces-extreme-t-subdwarf-candidate-cwise-j0523",
    "title": "Machine-learning search surfaces extreme T-subdwarf candidate CWISE J0523",
    "summary": "A machine-learning-assisted search of WISE time-series images helped identify the high-proper-motion object CWISE J052306.42-015355.4. Its colors suggest an extreme T-subdwarf candidate, while spectral classification, parallax, distance and metallicity remain uncertain. Human and citizen-science pathways also contributed. The candidate result was public in the November 2021 preprint before the January 2022 journal article.",
    "field": "physics",
    "ai_technology": "Machine-learning proper-motion search",
    "organization": "Northern Arizona University; Caltech IPAC; Backyard Worlds: Planet 9; NASA Goddard Space Flight Center",
    "source_url": "https://arxiv.org/abs/2111.08182v1",
    "source_type": "paper",
    "publication_date": "2021-11-16",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-3881/ac3a0a",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Archival astrometry and multi-survey photometry support the object and candidate classification.",
      "unknown": "Spectroscopic classification and measured parallax are not established in the reviewed study.",
      "ai_role": "Machine-learning proper-motion search supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://arxiv.org/abs/2111.08182v1",
        "https://ntrs.nasa.gov/citations/20220005456"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary v1 preprint metadata and abstract; NASA publication record, abstract, affiliations and journal date."
    }
  },
  {
    "id": "2022-04-learned-phase-models-identify-synergistic-photoanode-mixtures",
    "title": "Learned phase models identify synergistic photoanode mixtures",
    "summary": "Learning from measured diffraction and photocurrent data identified Bi-Cu-V oxide phase mixtures whose photoactivity exceeded an additive non-interacting-phase model. BiVO4-like phases showed beneficial associations with copper-vanadate phases. This is an experimentally grounded composition-property result; the underlying electronic, catalytic and morphological mechanisms remain unresolved, and device-level solar-fuel efficiency was not established.",
    "field": "materials",
    "ai_technology": "Deep Reasoning Networks and matrix factorization",
    "organization": "California Institute of Technology; Cornell University",
    "source_url": "https://www.nature.com/articles/s41524-022-00747-1",
    "source_type": "paper",
    "publication_date": "2022-04-05",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-022-00747-1",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Measured photocurrents and phase distributions were compared with a learned additive baseline.",
      "unknown": "Independent replication, detailed mechanisms and practical device performance were not assessed.",
      "ai_role": "Deep Reasoning Networks and matrix factorization supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-022-00747-1"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher abstract, selected phase-model and measured-synergy Results, date and affiliations; supplementary material not reviewed."
    }
  },
  {
    "id": "2022-05-learned-cluster-expansion-guides-an-acidic-oxygen-evolution-catalyst",
    "title": "Learned cluster expansion guides an acidic oxygen-evolution catalyst",
    "summary": "A fitted energy model and quantum-inspired optimizer selected a mixed-oxide catalyst family for laboratory testing. The study reports a Ru-Cr-Mn-Sb oxide catalyst with higher ruthenium-normalized activity than its RuO2 comparator and sustained operation in the tested setup. This is one bounded materials result, not commercial electrolyzer validation or evidence that the optimizer itself is a learned model. The catalyst result already appears in the May 2022 preprint.",
    "field": "materials",
    "ai_technology": "DFT-trained cluster expansion with quantum-inspired search",
    "organization": "University of Toronto; Harvard University; Fujitsu Consulting (Canada)",
    "source_url": "https://arxiv.org/abs/2205.09007v1",
    "source_type": "paper",
    "publication_date": "2022-05-18",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.matt.2022.11.031",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Selected oxide catalysts received electrochemical activity and stability measurements.",
      "unknown": "Scale-up, independent replication and general design success rates were not established.",
      "ai_role": "DFT-trained cluster expansion with quantum-inspired search supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://arxiv.org/abs/2205.09007v1",
        "https://light.northwestern.edu/wp-content/uploads/2023/01/1-s2.0-S2590238522006622-main.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary v1 preprint abstract/history; indexed publisher article and institutional publisher-PDF passages on model fitting and affiliations; supplements not reviewed."
    }
  },
  {
    "id": "2023-04-exogru-guided-experiments-support-rbm24-involvement-in-small-rna-secretion",
    "title": "ExoGRU-guided experiments support RBM24 involvement in small-RNA secretion",
    "summary": "A recurrent model of RNA-sequence features helped nominate RNA-binding proteins involved in secretion. Perturbation and binding experiments supported RBM24 involvement in the tested cell system, alongside previously implicated factors. This mechanism study was public in an April 2023 preprint before the 2024 journal version. It does not establish a universal secretion mechanism, in-vivo benefit or a therapeutic result.",
    "field": "biology",
    "ai_technology": "ExoGRU gated recurrent neural network",
    "organization": "University of California, San Francisco; Sharif University of Technology; McGill University",
    "source_url": "https://www.biorxiv.org/content/10.1101/2023.04.04.535452v1",
    "source_type": "paper",
    "publication_date": "2023-04-06",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.xgen.2024.100522",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Selected factors received perturbation and extracellular RNA-binding follow-up.",
      "unknown": "Generality across cell types and effects in vivo remain unestablished.",
      "ai_role": "ExoGRU gated recurrent neural network supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.1101/2023.04.04.535452v1",
        "https://api.biorxiv.org/details/biorxiv/10.1101/2023.04.04.535452",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11019361/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary bioRxiv API v1 metadata/abstract and journal PMC selected model, validation and limitations text; supplement not reviewed."
    }
  },
  {
    "id": "2024-01-active-learning-identifies-experimentally-improved-ammonia-catalysts",
    "title": "Active learning identifies experimentally improved ammonia catalysts",
    "summary": "Literature-trained models and iterative experimental feedback guided a predetermined ammonia-catalyst search. Selected formulations exceeded many reported comparator activities in the authors' laboratory measurements. Initial literature-only predictions were insufficient, making feedback essential. This is a bounded formulation result, not demonstrated commercial green-ammonia performance or invention of promoted ruthenium catalysis.",
    "field": "chemistry",
    "ai_technology": "Active-learning regression trained on literature and experimental feedback",
    "organization": "University of South Carolina",
    "source_url": "https://pubs.rsc.org/en/content/articlelanding/2024/ta/d3ta05939a",
    "source_type": "paper",
    "publication_date": "2024-01-02",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D3TA05939A",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers synthesized and tested model-selected catalyst formulations and updated the model.",
      "unknown": "Independent formulation priority, replication and industrial performance were not assessed.",
      "ai_role": "Active-learning regression trained on literature and experimental feedback supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlelanding/2024/ta/d3ta05939a"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher abstract, author affiliations and publication metadata; prior frozen method review retained, author manuscript freshly blocked."
    }
  },
  {
    "id": "2023-01-support-vector-regression-selects-more-selective-tetrahydropyran-catalysts",
    "title": "Support-vector regression selects more selective tetrahydropyran catalysts",
    "summary": "Models of catalyst fragments and reaction settings selected catalysts for asymmetric tetrahydropyran synthesis. A tested catalyst reached an enantiomeric ratio of 91.5:8.5, above the training examples' maximum of 82:18. The result improves a known catalyst class for a focused reaction; it does not establish invention of the reaction or broad generality.",
    "field": "chemistry",
    "ai_technology": "Support-vector regression with tunable fragment descriptors",
    "organization": "Hokkaido University; Max Planck Institute for Coal Research; University of Strasbourg",
    "source_url": "https://onlinelibrary.wiley.com/doi/full/10.1002/anie.202218659",
    "source_type": "paper",
    "publication_date": "2023-01-23",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/anie.202218659",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Previously untested catalyst substituents were prepared and experimentally evaluated.",
      "unknown": "Exhaustive catalyst prior art, independent replication and broader reaction scope were not assessed.",
      "ai_role": "Support-vector regression with tunable fragment descriptors supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/full/10.1002/anie.202218659"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher article dates, affiliations and selected model/experimental-validation Results; supplementary data not reviewed."
    }
  },
  {
    "id": "2024-09-human-guided-active-learning-improves-nickel-cobalt-methanation-tests",
    "title": "Human-guided active learning improves nickel-cobalt methanation tests",
    "summary": "A human-in-the-loop model selected nickel-cobalt catalysts and reaction settings for a 48-test laboratory campaign. Further selected experiments improved methane space-time yield by nearly 50% over the previous maximum in that dataset. Extrapolation remained harder than interpolation. This is a bounded catalyst-and-condition result, not evidence of industrial deployment or lifecycle climate benefit.",
    "field": "chemistry",
    "ai_technology": "Gaussian-process active learning with batch acquisition",
    "organization": "EPFL; Empa; Ghent University",
    "source_url": "https://pubs.rsc.org/en/content/articlehtml/2024/cy/d4cy00873a",
    "source_type": "paper",
    "publication_date": "2024-09-10",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D4CY00873A",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected catalysts and conditions were experimentally tested; model-suggested activity trends received follow-up.",
      "unknown": "Generality, industrial durability and lifecycle carbon benefit were not established.",
      "ai_role": "Gaussian-process active learning with batch acquisition supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlehtml/2024/cy/d4cy00873a"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher abstract, model Methods, selected Results, affiliations and first-online date; supplementary material not reviewed."
    }
  },
  {
    "id": "2023-10-neural-active-learning-guides-stronger-bone-scaffold-architectures",
    "title": "Neural active learning guides stronger bone-scaffold architectures",
    "summary": "Learned architecture models and finite-element simulations guided fabrication of scaffold subunits. A larger scaffold assembled through human adaptation of the learned structures showed about 20% greater load-bearing capacity than the uniform comparator. This is a mechanical bench-test result, not evidence of implantation, bone healing or clinical benefit.",
    "field": "materials",
    "ai_technology": "GAD-MALL autoencoder and convolutional active learning",
    "organization": "Tsinghua University; Peking University; Max Planck Institute for Iron Research",
    "source_url": "https://www.nature.com/articles/s41467-023-42415-y",
    "source_type": "paper",
    "publication_date": "2023-10-19",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-42415-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Printed structures received mechanical tests.",
      "unknown": "Independent replication and in-vivo outcomes were not established.",
      "ai_role": "GAD-MALL autoencoder and convolutional active learning supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-023-42415-y",
        "https://pure.mpg.de/pubman/item/item_3643037_1/component/file_3643048/s41467-023-42415-y.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher affiliations, date and selected design Results; indexed institutional publisher PDF mechanical comparison. Supplements not reviewed."
    }
  },
  {
    "id": "2023-04-learned-promoter-models-identify-human-specific-transcription-motifs",
    "title": "Learned promoter models identify human-specific transcription motifs",
    "summary": "Models trained on promoter activity measurements helped characterize a human-specific downstream promoter variant and select synthetic motifs with species-specific activity. In-vitro transcription assays tested the predicted preferences with human and Drosophila extracts. These are gene-regulation and construct-assay findings, not demonstrated effects in organisms or disease.",
    "field": "biology",
    "ai_technology": "HARPE-trained support-vector regression",
    "organization": "University of California San Diego",
    "source_url": "https://escholarship.org/uc/item/1rt3r053",
    "source_type": "paper",
    "publication_date": "2023-04-25",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1101/gad.350572.123",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Predicted promoter preferences were compared with measured transcription in extracts.",
      "unknown": "Native-chromatin regulatory effects and organism-level outcomes were not established.",
      "ai_role": "HARPE-trained support-vector regression supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://escholarship.org/uc/item/1rt3r053",
        "https://escholarship.org/content/qt1rt3r053/qt1rt3r053.pdf",
        "https://pubmed.ncbi.nlm.nih.gov/37163335/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Institutional manuscript abstract and selected SVR Methods; primary PubMed early-online date, affiliations and experimental figure captions."
    }
  },
  {
    "id": "2023-10-btsbot-workflow-automatically-confirms-and-classifies-sn-2023tyk",
    "title": "BTSbot workflow automatically confirms and classifies SN 2023tyk",
    "summary": "BTSbot selected SN 2023tyk from ZTF alerts and requested robotic spectroscopy; SNIascore supplied its Type Ia classification. The integrated workflow publicly reported the result without human intervention in those steps. This is one transient result, not a new supernova class or a claim that BTSbot alone performed the observation. It preceded the expanded 2024 study.",
    "field": "physics",
    "ai_technology": "BTSbot convolutional classifier and SNIascore",
    "organization": "Northwestern University; California Institute of Technology",
    "source_url": "https://www.wis-tns.org/astronotes/astronote/2023-265",
    "source_type": "other",
    "publication_date": "2023-10-11",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-4357/ad5666",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Announcement only",
      "tested": "Robotic spectroscopy supported the Type Ia classification.",
      "unknown": "Universal discovery priority and broader workflow completeness were not assessed.",
      "ai_role": "BTSbot convolutional classifier and SNIascore supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.wis-tns.org/astronotes/astronote/2023-265",
        "https://news.northwestern.edu/stories/2023/10/first-supernova-detected-confirmed-classified-and-shared-by-ai?fj=1"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary TNS AstroNote and institutional announcement, plus indexed published-paper abstract. Direct TNS retrieval blocked."
    }
  },
  {
    "id": "2024-08-learned-call-analysis-identifies-receiver-specific-marmoset-vocal-patterns",
    "title": "Learned call analysis identifies receiver-specific marmoset vocal patterns",
    "summary": "Classifiers identified receiver-related structure in calls from ten captive marmosets. Playback experiments also found stronger responses to calls originally directed at the listener. The authors interpret these patterns as vocal labels; a later commentary disputes whether the evidence establishes name-like reference, and the authors defend that interpretation. The bounded finding is receiver-specific acoustic and behavioral differentiation, not decoded human-like names.",
    "field": "biology",
    "ai_technology": "Random-forest acoustic classifiers",
    "organization": "Hebrew University of Jerusalem",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/39208084/",
    "source_type": "paper",
    "publication_date": "2024-08-29",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/science.adp3757",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Recorded calls and playback responses were analyzed.",
      "unknown": "Name-like reference and alternative explanations remain debated; the captive sample is small.",
      "ai_role": "Random-forest acoustic classifiers supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39208084/",
        "https://zoology.ubc.ca/edg/pdfs/Oren2024.pdf",
        "https://pubmed.ncbi.nlm.nih.gov/39747802/",
        "https://pubmed.ncbi.nlm.nih.gov/40866605/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary published-paper Methods and abstract, PubMed metadata, critical commentary abstract and authors response abstract."
    }
  },
  {
    "id": "2023-05-embryonet-links-statins-to-fgf-like-zebrafish-developmental-effects",
    "title": "EmbryoNet links statins to FGF-like zebrafish developmental effects",
    "summary": "A learned embryo-image classifier flagged statin-treated zebrafish as resembling reduced FGF signaling. Follow-up morphology and pERK measurements supported the developmental-pathway association. The mechanism remains proposed, and the zebrafish assays do not establish effects in human pregnancy or clinical use.",
    "field": "biology",
    "ai_technology": "EmbryoNet deep image classifier",
    "organization": "University of Konstanz; Friedrich Miescher Laboratory of the Max Planck Society",
    "source_url": "https://www.nature.com/articles/s41592-023-01873-4",
    "source_type": "paper",
    "publication_date": "2023-05-08",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-023-01873-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Embryonic morphology and pathway-associated fluorescence were measured.",
      "unknown": "Causal mechanism and human relevance were not established.",
      "ai_role": "EmbryoNet deep image classifier supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41592-023-01873-4",
        "https://pubmed.ncbi.nlm.nih.gov/37156842/",
        "https://www.nature.com/articles/s41592-023-01873-4.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher experimental Results and PDF passages; primary PubMed date and affiliations. Direct full-text retrieval blocked."
    }
  },
  {
    "id": "2021-07-boolean-network-learning-prioritizes-an-ibd-organoid-barrier-target",
    "title": "Boolean-network learning prioritizes an IBD organoid barrier target",
    "summary": "Learning from transcriptomic networks helped prioritize PRKAB1 for experimental follow-up. An agonist improved a barrier-resistance endpoint in about 80% of diseased patient-derived organoids in the tested sample. This is preclinical target-response evidence, not treatment of patients or clinical efficacy. The network direction incorporates sample labels and does not independently establish causal disease progression.",
    "field": "medicine",
    "ai_technology": "Boolean Network Explorer learned implication networks",
    "organization": "University of California San Diego; Assiut University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8275683/",
    "source_type": "paper",
    "publication_date": "2021-07-12",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-021-24470-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Patient-derived organoid barriers and preclinical models were tested.",
      "unknown": "Clinical safety and benefit, independent replication and causal network direction remain unestablished.",
      "ai_role": "Boolean Network Explorer learned implication networks supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8275683/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article metadata, affiliations, selected network Methods and organoid Results; supplements not reviewed."
    }
  },
  {
    "id": "2024-09-costar-predicts-drug-activity-tested-in-glioblastoma-samples",
    "title": "COSTAR predicts drug activity tested in glioblastoma samples",
    "summary": "A learned drug-target connectivity signature selected 23 predicted hits and 25 negatives for testing in four glioblastoma patient samples. Ex-vivo measurements agreed with predictions at a reported mean ROC AUC of 0.94. The small validation establishes a bounded drug-activity prediction result, not clinical response. Vortioxetine was found by the initial experimental screen and is not attributed to COSTAR nomination.",
    "field": "medicine",
    "ai_technology": "COSTAR logistic LASSO drug-target model",
    "organization": "ETH Zurich; University Hospital Zurich; University of Zurich",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11564103/",
    "source_type": "paper",
    "publication_date": "2024-09-20",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41591-024-03224-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Newly selected compounds were tested ex vivo across four patient samples.",
      "unknown": "Patient benefit, generalization and clinical safety remain unestablished.",
      "ai_role": "COSTAR logistic LASSO drug-target model supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11564103/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary full-text model Methods and selected experimental Results, publication metadata and affiliations; supplements not reviewed."
    }
  },
  {
    "id": "2025-02-machine-learning-assisted-taxonomy-identifies-limnognathia-desmeti",
    "title": "Machine-learning-assisted taxonomy identifies Limnognathia desmeti",
    "summary": "Learned analyses of transcriptomic data contributed to an integrative description of Limnognathia desmeti from the Crozet Islands, the second species described in Micrognathozoa. Phylogenomic and molecular evidence also contributed; machine learning alone did not establish the species. Greenland and Pyrenean populations were not definitively split. Limited specimen and population sampling constrains the result.",
    "field": "biology",
    "ai_technology": "Variational autoencoder and random-forest species delimitation",
    "organization": "University of Copenhagen; University of Barcelona; Natural History Museum, London",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11836703/",
    "source_type": "paper",
    "publication_date": "2025-02-19",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1098/rspb.2024.2867",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Molecular analyses, microscopy and multiple delimitation approaches supported the formal description.",
      "unknown": "Broader geographic diversity and dispersal history remain uncertain.",
      "ai_role": "Variational autoencoder and random-forest species delimitation supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11836703/"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary full-text species-delimitation Methods, taxonomic Results, date and affiliations; supplements not reviewed."
    }
  },
  {
    "id": "2020-10-opex-guided-experiments-reveal-bacterial-cross-stress-responses",
    "title": "OPEX-guided experiments reveal bacterial cross-stress responses",
    "summary": "Learned models selected bacterial stress-combination experiments and incorporated their measurements. The study reported cross-stress protection and vulnerability patterns, with follow-up supporting roles for stress-response machinery. This is a laboratory microbiology result, not a clinical resistance prediction or evidence of treatment efficacy.",
    "field": "biology",
    "ai_technology": "OPEX machine-learning experimental selection",
    "organization": "University of California, Davis",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/33024104/",
    "source_type": "paper",
    "publication_date": "2020-10-06",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-020-18785-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected experiments and subsequent biological validation tested stress responses.",
      "unknown": "Transfer beyond the studied organism and conditions, clinical outcomes and independent replication were not assessed.",
      "ai_role": "OPEX machine-learning experimental selection supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/33024104/",
        "https://escholarship.org/content/qt5cc185t3/qt5cc185t3.pdf?t=qlp51p",
        "https://tagkopouloslab.ucdavis.edu/news/2020/10/new-article-in-nature-communication/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PubMed abstract/date/affiliations, indexed deposited publisher-paper abstract and author laboratory announcement."
    }
  },
  {
    "id": "2021-01-learned-resistance-models-expose-tested-enzyme-substrate-activities",
    "title": "Learned resistance models expose tested enzyme-substrate activities",
    "summary": "Interrogating models trained on genome and resistance data nominated enzyme-substrate relationships for experimental testing. The authors report 22 previously unknown activities validated in their assay system. This is one mechanism study; it does not establish the prevalence or clinical contribution of each relationship in patients. Model performance varied substantially across organism and drug datasets.",
    "field": "biology",
    "ai_technology": "Antibiotic-specific logistic-regression classifiers",
    "organization": "McMaster University; Dalhousie University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8115898/",
    "source_type": "paper",
    "publication_date": "2021-01-08",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1099/mgen.0.000500",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Predicted relationships received targeted experimental validation.",
      "unknown": "Clinical prevalence, generalization and independent replication remain unestablished.",
      "ai_role": "Antibiotic-specific logistic-regression classifiers supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8115898/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article date, affiliations, selected logistic-regression Methods and experimental Results; supplements not reviewed."
    }
  },
  {
    "id": "2020-03-learned-genomic-models-predict-baltic-microbial-niche-placement",
    "title": "Learned genomic models predict Baltic microbial niche placement",
    "summary": "Models trained on functional-gene profiles predicted microbial distribution along salinity, depth and sampling-fraction gradients in Baltic Sea observations. The study links genomic repertoire to ecological niche placement within this ecosystem. It does not demonstrate that associated genes causally determine habitat or that the models transfer elsewhere. The genome clusters are not separately counted discoveries.",
    "field": "biology",
    "ai_technology": "Genome-feature regression and tree-based learning",
    "organization": "KTH Royal Institute of Technology; Leibniz Institute for Baltic Sea Research",
    "source_url": "https://www.nature.com/articles/s42003-020-0856-x",
    "source_type": "paper",
    "publication_date": "2020-03-13",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42003-020-0856-x",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Predictions were compared with observed distributions in the sampled ecosystem.",
      "unknown": "Causal effects, field manipulation and cross-ecosystem transfer were not established.",
      "ai_role": "Genome-feature regression and tree-based learning supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42003-020-0856-x",
        "https://epic.awi.de/52845/1/Alneberg_et_al_2020.pdf"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher abstract, date and affiliations, plus institutional publisher-PDF metadata; Methods review retained from frozen research."
    }
  },
  {
    "id": "2022-02-knowledge-graph-learning-prioritizes-experimentally-examined-alzheimer-genes",
    "title": "Knowledge-graph learning prioritizes experimentally examined Alzheimer genes",
    "summary": "A learned knowledge-graph model nominated candidate Alzheimer-associated genes for neuronal, autopsy and cell-assay follow-up. The study highlights candidate associations rather than establishing causes of human disease. Small samples, noisy negative labels and an internal discrepancy in reported knockdown-hit counts constrain interpretation; those counts are not treated as reconciled.",
    "field": "medicine",
    "ai_technology": "MPxgb(AD) knowledge-graph XGBoost",
    "organization": "University of New Mexico; University of Gothenburg; University of Copenhagen",
    "source_url": "https://www.nature.com/articles/s42003-022-03068-7",
    "source_type": "paper",
    "publication_date": "2022-02-11",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42003-022-03068-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected predicted genes received biological screening.",
      "unknown": "Independent validation, human causality and therapeutic benefit remain unestablished.",
      "ai_role": "MPxgb(AD) knowledge-graph XGBoost supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42003-022-03068-7",
        "https://www.nature.com/articles/s42003-022-03068-7.pdf",
        "https://pubmed.ncbi.nlm.nih.gov/35149761/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher PDF abstract and model description, publisher date and PubMed affiliations; detailed prior Methods review retained, direct PMC retrieval blocked."
    }
  },
  {
    "id": "2023-06-organ-specific-learned-networks-prioritize-tested-seed-lipid-regulators",
    "title": "Organ-specific learned networks prioritize tested seed-lipid regulators",
    "summary": "Learned organ classification enabled regulatory-network inference that nominated seed-lipid regulators for genetic and physiological testing in Arabidopsis. The result was public in a June 2023 preprint before the April 2024 journal article. Candidate counts differ between versions, so the entry does not transfer final-version counts or mechanistic detail to the preprint. Crop-yield benefit was not established.",
    "field": "biology",
    "ai_technology": "K-nearest-neighbor organ classification and regulatory-network inference",
    "organization": "Purdue University",
    "source_url": "https://www.biorxiv.org/content/10.1101/2023.06.17.542927v1",
    "source_type": "paper",
    "publication_date": "2023-06-18",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2322751121",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Predicted regulators received genetic and physiological experimental follow-up.",
      "unknown": "Cross-species utility, crop performance and version-specific numerical correspondence remain unestablished.",
      "ai_role": "K-nearest-neighbor organ classification and regulatory-network inference supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.1101/2023.06.17.542927v1",
        "https://api.biorxiv.org/details/biorxiv/10.1101/2023.06.17.542927",
        "https://www.pnas.org/doi/10.1073/pnas.2322751121"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary bioRxiv API v1 abstract/date plus journal abstract, selected Methods and affiliations; no full preprint numerical endpoints transferred."
    }
  },
  {
    "id": "2023-05-active-learning-maps-wind-tunnel-terrain-equivalence",
    "title": "Active learning maps wind-tunnel terrain equivalence",
    "summary": "A learned surrogate selected automated wind-tunnel experiments that mapped terrain configurations with similar second-order wind statistics. Different terrain shapes could meet the study's equivalence criterion. This does not imply identical full turbulence fields or equal structural safety; higher-order statistics can differ. Estimated experiment savings are not a fully executed randomized comparison with alternative sampling.",
    "field": "physics",
    "ai_technology": "Gaussian-process sequential active learning",
    "organization": "Johns Hopkins University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10209069/",
    "source_type": "paper",
    "publication_date": "2023-05-24",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41598-023-35257-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Automated terrain changes and measured wind profiles defined experimental equivalence regions.",
      "unknown": "Generality beyond the facility, full-field equivalence and safety outcomes remain unestablished.",
      "ai_role": "Gaussian-process sequential active learning supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10209069/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article retrieved via browser request: publication metadata, Gaussian-process Methods and terrain-study Results; supplements not reviewed."
    }
  },
  {
    "id": "2024-06-simulation-trained-inference-maps-accreted-stellar-mass-in-nearby-galaxies",
    "title": "Simulation-trained inference maps accreted stellar mass in nearby galaxies",
    "summary": "A model trained on simulated galaxy observations inferred accreted stellar fractions in nearby MaNGA galaxies. Predictions suggest in-situ formation dominates most of the studied mass range, while accreted fractions rise in the most massive systems and vary with galaxy characteristics. These are simulation-dependent inferences, not direct measurements of each star's formation history. The result preceded the July 2024 journal article.",
    "field": "physics",
    "ai_technology": "MaNGIA/TNG50-trained stellar-mass inference",
    "organization": "Instituto de Astrofísica de Canarias; University of La Laguna; Max Planck Institute for Astronomy",
    "source_url": "https://arxiv.org/abs/2407.00166v1",
    "source_type": "paper",
    "publication_date": "2024-06-28",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41550-024-02327-3",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Model inference was evaluated with simulated observations and applied to real survey data.",
      "unknown": "Observational ground truth and transfer beyond the training simulation remain unresolved.",
      "ai_role": "MaNGIA/TNG50-trained stellar-mass inference supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://arxiv.org/abs/2407.00166v1",
        "https://arxiv.org/html/2407.00166v1"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary v1 abstract, history, affiliations and selected HTML Methods; supplements not reviewed."
    }
  },
  {
    "id": "2023-02-learned-morphology-guides-targeted-isolation-of-gut-bacteria",
    "title": "Learned morphology guides targeted isolation of gut bacteria",
    "summary": "A learned morphology model selected likely target genera for robotic isolation, followed by sequencing to confirm cultured organisms. This supports a bounded, experimentally tested recovery result. The platform's broader diversity-selection component uses feature distances and is not itself described as learned. Neither every isolate nor every genome is counted as a separate discovery, and clinical benefit was not established.",
    "field": "biology",
    "ai_technology": "CAMII random-forest colony taxonomy prediction",
    "organization": "Columbia University",
    "source_url": "https://www.nature.com/articles/s41587-023-01674-2",
    "source_type": "paper",
    "publication_date": "2023-02-20",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-023-01674-2",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "New colonies selected by the taxonomy model were isolated and sequenced.",
      "unknown": "Generality beyond the studied samples and culture conditions remains unestablished.",
      "ai_role": "CAMII random-forest colony taxonomy prediction supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41587-023-01674-2",
        "https://www.nature.com/articles/s41587-023-01674-2.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher PDF targeted-isolation Methods, publisher abstract/date and affiliations; prior experimental Results review retained."
    }
  },
  {
    "id": "2023-04-learned-community-models-prioritize-a-tested-bacterial-colonization-inhibitor",
    "title": "Learned community models prioritize a tested bacterial colonization inhibitor",
    "summary": "Models trained on community composition helped identify Enterococcus faecalis as an inhibitor of Enterococcus faecium colonization in human stool-derived laboratory communities. Follow-up supported the predicted interaction under the tested conditions. The result was public in an April 2023 preprint before the March 2024 journal article. It does not demonstrate prevention of infection or beneficial intervention in humans.",
    "field": "biology",
    "ai_technology": "Random-forest and neural-ODE community models",
    "organization": "Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences; Brigham and Women's Hospital; Harvard Medical School",
    "source_url": "https://www.biorxiv.org/content/10.1101/2023.04.19.537502v1",
    "source_type": "paper",
    "publication_date": "2023-04-21",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-46766-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Model-prioritized interactions received in-vitro validation.",
      "unknown": "Host effects, clinical benefit and transfer beyond the culture conditions remain unestablished.",
      "ai_role": "Random-forest and neural-ODE community models supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.1101/2023.04.19.537502v1",
        "https://api.biorxiv.org/details/biorxiv/10.1101/2023.04.19.537502",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10944475/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary bioRxiv API v1 abstract/date; journal article selected model and intervention Results, metadata and affiliations. Detailed final-version counts not transferred."
    }
  },
  {
    "id": "2022-07-active-learning-improves-a-cell-free-carbon-fixation-cycle",
    "title": "Active learning improves a cell-free carbon-fixation cycle",
    "summary": "An ensemble learned from experimental feedback to select compositions for an engineered carbon-fixation cycle. Laboratory measurements supported improved glycolate production compared with the study's reference cycle. This is an in-vitro pathway result, not carbon capture by living organisms or a demonstrated industrial process. Other applications in the same paper are not additional catalog entries.",
    "field": "biology",
    "ai_technology": "METIS XGBoost active learning",
    "organization": "Max Planck Institute for Terrestrial Microbiology",
    "source_url": "https://www.nature.com/articles/s41467-022-31245-z",
    "source_type": "paper",
    "publication_date": "2022-07-05",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-022-31245-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected pathway compositions received biochemical measurements.",
      "unknown": "Independent replication, in-vivo function and scale-up remain unestablished.",
      "ai_role": "METIS XGBoost active learning supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-022-31245-z",
        "https://www.nature.com/articles/s41467-022-31245-z.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary publisher PDF model Methods, date and affiliations; prior primary experimental Results review retained. Direct PDF access blocked."
    }
  },
  {
    "id": "2020-07-random-forest-screening-identifies-experimentally-active-anti-crispr-families",
    "title": "Random-forest screening identifies experimentally active anti-CRISPR families",
    "summary": "A learned model ranked candidate protein families for experimental follow-up. Testing supported two previously unknown anti-CRISPR proteins in the studied system. The much larger predicted family catalog remains computational and is not counted as experimentally validated discoveries. Assay activity does not establish native-host function or improved genome-editing performance.",
    "field": "biology",
    "ai_technology": "Extremely randomized forest of genomic-context features",
    "organization": "National Institutes of Health; University of California, San Francisco",
    "source_url": "https://www.nature.com/articles/s41467-020-17652-0.pdf",
    "source_type": "paper",
    "publication_date": "2020-07-29",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-020-17652-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected predicted candidates received functional laboratory tests.",
      "unknown": "Broader family validation, native-host activity and practical utility remain unestablished.",
      "ai_role": "Extremely randomized forest of genomic-context features supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-020-17652-0.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary published PDF abstract, affiliations and model-training Methods; experimental scope checked against prior primary review."
    }
  },
  {
    "id": "2021-12-bayesian-optimization-guides-complex-nanoparticle-heterostructures",
    "title": "Bayesian optimization guides complex nanoparticle heterostructures",
    "summary": "Learning from earlier synthesis and microscopy data guided a human-supervised nanoparticle campaign. Researchers reported the desired structures for 18 of 19 selected candidates, including complex multielement particles. This success rate concerns a curated experimental set, not the full predicted composition space. Structural synthesis was demonstrated; catalytic and device performance were not.",
    "field": "materials",
    "ai_technology": "Bayesian optimization of composition and morphology",
    "organization": "Northwestern University; Toyota Research Institute",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC8694626/",
    "source_type": "paper",
    "publication_date": "2021-12-22",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.abj5505",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Synthesized candidates were examined by microscopy and elemental analysis.",
      "unknown": "Functional performance, general search-space success and independent replication were not established.",
      "ai_role": "Bayesian optimization of composition and morphology supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC8694626/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article metadata, affiliations, selected experimental Results and discussion; supplements not reviewed."
    }
  },
  {
    "id": "2022-03-learned-activity-models-improve-viral-diagnostic-assay-designs",
    "title": "Learned activity models improve viral diagnostic assay designs",
    "summary": "Learned diagnostic-activity scores guided assay design across viral variation. Experiments on selected synthetic targets supported improved detection compared with the studied conservation-based designs. The larger cross-species design catalog is computational; it is not thousands of validated diagnostics. Patient-specimen sensitivity, clinical specificity and field performance were not established.",
    "field": "biology",
    "ai_technology": "ADAPT learned sequence-activity classifier and regressor",
    "organization": "Broad Institute of MIT and Harvard; Massachusetts Institute of Technology",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9287178/",
    "source_type": "paper",
    "publication_date": "2022-03-03",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41587-022-01213-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected assay designs were tested on synthetic viral targets.",
      "unknown": "Clinical and environmental sample performance require further validation.",
      "ai_role": "ADAPT learned sequence-activity classifier and regressor supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9287178/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article metadata, affiliations, selected design Methods, experimental Results and limitations; supplements not reviewed."
    }
  },
  {
    "id": "2020-06-del-trained-models-identify-experimentally-active-small-molecule-hits",
    "title": "DEL-trained models identify experimentally active small-molecule hits",
    "summary": "Models trained on DNA-encoded-library selection data prioritized roughly 2,000 compounds for biochemical testing across three targets. The study reports active and potent hits outside the training libraries. Human filtering and chemical review contributed to selection. These are biochemical hits, not established safe or effective drugs; the authors disclose commercial interests and related patent applications.",
    "field": "chemistry",
    "ai_technology": "Graph-convolutional neural networks and random forests",
    "organization": "Google Research; X-Chem",
    "source_url": "https://accio.github.io/AMIDD/assets/2020/13-14/McCloskey-2020-ML-DELT.pdf",
    "source_type": "paper",
    "publication_date": "2020-06-11",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jmedchem.0c00452",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Newly selected compounds were tested in biochemical assays.",
      "unknown": "Cellular efficacy, selectivity, safety and clinical benefit remain unestablished.",
      "ai_role": "Graph-convolutional neural networks and random forests supported the stated study; experimental and interpretive work remained with researchers.",
      "evidence_urls": [
        "https://accio.github.io/AMIDD/assets/2020/13-14/McCloskey-2020-ML-DELT.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Author-hosted published-article PDF abstract, selected experimental Results, affiliations and disclosures; prior PubMed online date retained."
    }
  },
  {
    "id": "2021-08-ml-moth-receptor-larval-attractants",
    "title": "Machine learning identifies moth-receptor agonists that attract larvae",
    "summary": "Random-forest and k-nearest-neighbors QSAR models trained on known odorant responses prioritized natural-product candidates for two Spodoptera littoralis odorant receptors. The paper reports receptor responses for selected compounds and Y-maze preference for seven strong agonists; all seven were preferred at 10⁻² dilution, while three retained preference at 10⁻³. The percentages of receptor-active candidates apply to selected test sets and are not field trap accuracy. The fly-neuron assay is heterologous, and larval choice is laboratory evidence, not crop-protection efficacy or ecological safety. This extends the same group’s 2020 receptor-ligand study; count the 2021 behavior result as one study, not each molecule or a separate model discovery.",
    "field": "biology",
    "ai_technology": "Random-forest QSAR and k-nearest-neighbors classifier",
    "organization": "INRAE; Sorbonne Université; CNRS; IRD; Université de Paris; Université Côte d’Azur; Daegu Gyeongbuk Institute of Science and Technology",
    "source_url": "https://link.springer.com/article/10.1007/s00018-021-03919-2",
    "source_type": "paper",
    "publication_date": "2021-08-27",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher article, article metadata and author-affiliation block reviewed; summary is limited to reported assay results."
    },
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected ML-prioritized receptor ligands were tested electrophysiologically and in larval Y-maze assays; the paper reports the stated dose-dependent preference outcomes.",
      "unknown": "No field efficacy, human health benefit, broad ecological selectivity or global priority is established. The study extends earlier same-group work; its compounds and behavior result are bounded to the reported assays.",
      "ai_role": "The models ranked candidate natural-product ligands for SlitOR24/SlitOR25; researchers selected the targets and candidates and performed the receptor and larval behavioral experiments.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1007/s00018-021-03919-2"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-10-ml-human-olfactory-receptor-agonists",
    "title": "Machine learning prioritizes agonists for two human olfactory receptors",
    "summary": "A random-forest classifier trained from molecular descriptors screened five databases containing about 23 million compounds for agonists of human olfactory receptors OR1A1 and OR2W1. Following docking and researcher selection, the paper reports in-vitro activation by ethyl 2-(3-bromophenyl)acetate and methyl 2-(2-methylphenyl)acetate at OR1A1 and 2-ethoxynaphthalene at OR2W1. These are cell-assay results; they do not establish human perception, tissue safety or clinical benefit. The compounds were selected after multiple filters, so the results are not a database-wide hit rate. Prior ML work on related receptors exists; this record is limited to the reported candidates and assays.",
    "field": "biology",
    "ai_technology": "Random-forest classifier; molecular descriptors; homology-model docking",
    "organization": "Macquarie University; Duke University",
    "source_url": "https://www.mdpi.com/1422-0067/22/21/11546",
    "source_type": "paper",
    "publication_date": "2021-10-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher article, article metadata and author-affiliation block reviewed; summary is limited to reported assay results."
    },
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected candidates were tested in vitro; the paper reports two active OR1A1 ligands and one active OR2W1 ligand.",
      "unknown": "No in-vivo human olfaction or clinical effect is established. A 2018 study used ML for related human receptors; candidate-level overlap and priority remain bounded follow-up.",
      "ai_role": "The random-forest classifier screened chemical databases to rank candidates; docking and researcher selection further narrowed compounds for in-vitro validation.",
      "evidence_urls": [
        "https://www.mdpi.com/1422-0067/22/21/11546"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-02-deep-learning-labmol-antimalarial-candidates",
    "title": "Deep learning screens LabMol compounds for antiplasmodial activity",
    "summary": "Deep-learning QSAR models for parasite activity and cytotoxicity prioritized commercial candidates for assay. Primary Table 3 reports LabMol-152 EC50 values of 0.049 and 0.078 µM against 3D7 and W2; LabMol-149 values were 1.450 and 0.509 µM, inconsistent with the abstract’s under-500-nM claim. The study reports in-vitro results only; the model ranked existing compounds rather than generating them.",
    "field": "medicine",
    "ai_technology": "Deep-neural-network QSAR models",
    "organization": "University Center of Anápolis–UniEVANGÉLICA; Federal University of Goiás",
    "source_url": "https://doi.org/10.1371/journal.pcbi.1007025",
    "source_type": "paper",
    "publication_date": "2020-02-18",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Table 3 reports strain-specific EC50 and COS-7 cytotoxicity results for tested compounds.",
      "unknown": "Abstract/results inconsistency for LabMol-149 preserved; no animal or clinical evidence; earlier public version/chemical prior art unresolved.",
      "ai_role": "Activity and cytotoxicity models ranked an existing library and filters/human inspection selected compounds for assay.",
      "evidence_urls": [
        "https://doi.org/10.1371/journal.pcbi.1007025"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "publisher primary full text. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2025-01-random-forest-antimalarial-candidates",
    "title": "Random forest prioritizes investigational compounds with antiplasmodial activity",
    "summary": "A random-forest classifier screened investigational compounds; after applicability, similarity, diversity, availability and cost filters, six compounds were purchased and tested. CEP-37440 and AZD-1480 had reported P. falciparum IC50 values of 1.22 and 4.00 µM. These are in-vitro findings for existing compounds, not evidence of malaria clinical safety or a confirmed parasite target.",
    "field": "medicine",
    "ai_technology": "Random-forest classifier",
    "organization": "Birla Institute of Technology and Science Pilani; Institute of Bioinformatics and Applied Biotechnology",
    "source_url": "https://doi.org/10.1186/s13065-025-01395-4",
    "source_type": "paper",
    "publication_date": "2025-01-30",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports tested CEP-37440/AZD-1480 antiplasmodial activity.",
      "unknown": "Selected six are filtered, not a random sample; proposed parasite kinase mechanisms and patient efficacy are unverified.",
      "ai_role": "RF ranks investigational compounds from a drug database; human filters determine purchase/assay set.",
      "evidence_urls": [
        "https://doi.org/10.1186/s13065-025-01395-4"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "publisher primary full text. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2022-07-random-forest-dual-stage-antimalarial-agents",
    "title": "Random forest prioritizes dual-stage antimalarial compounds",
    "summary": "A random forest trained on parasite activity data screened commercial compounds; 100 predicted actives and 20 predicted inactives were tested. The primary manuscript reports 18 dose-response-validated compounds with liver- and blood-stage activity among 26 liver-stage hits. The selected sets are not population samples, and assays used P. berghei liver-stage and P. falciparum blood-stage parasites.",
    "field": "medicine",
    "ai_technology": "Random-forest classifier",
    "organization": "Rutgers University–New Jersey Medical School; Duke University",
    "source_url": "https://doi.org/10.1021/acsinfecdis.2c00189",
    "source_type": "paper",
    "publication_date": "2022-07-27",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary indexed Methods/Results report 18 dual-stage validated compounds and screening denominators.",
      "unknown": "Chemical prior art, complete supplemental details and clinical translation remain open; do not present 18/26 as general hit rate.",
      "ai_role": "Classifier prioritizes compounds from >1.5 million virtual candidates; human structural-diversity selection and assays follow.",
      "evidence_urls": [
        "https://doi.org/10.1021/acsinfecdis.2c00189"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "PMC-indexed primary manuscript text; publisher primary metadata/index. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2023-02-virtual-screened-beta-haematin-inhibitors",
    "title": "Virtual screening identifies antiplasmodial beta-haematin inhibitor chemotypes",
    "summary": "Molecular docking and Bayesian/SVM classifiers screened a ChemDiv 300k library. Researchers tested prioritized commercial compounds and reported coumarin and iminodipyridinopyrimidine chemotypes active against beta-haematin and whole-cell P. falciparum, including chloroquine-resistant parasites. This is one bounded in-vitro campaign, not clinical evidence.",
    "field": "medicine",
    "ai_technology": "Bayesian and support-vector classifiers; molecular docking",
    "organization": "University of Cape Town",
    "source_url": "https://doi.org/10.1038/s41598-023-29273-w",
    "source_type": "paper",
    "publication_date": "2023-02-14",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary abstract reports prioritized compounds with beta-haematin and whole-cell parasite activity.",
      "unknown": "A 2023 conference abstract may disclose an early coumarin result; the article’s specific two-chemotype campaign is retained without priority claim; no clinical evidence.",
      "ai_role": "Sequential ligand/structure virtual screening prioritizes commercial compounds for biochemical and parasite assays.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41598-023-29273-w"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "PMC-hosted primary full article; PubMed primary record/abstract. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2026-09-pca-map-antiplasmodial-haemozoin-screen",
    "title": "PCA mapping and machine learning prioritize antiplasmodial chemotypes",
    "summary": "A PCA enrichment map and machine-learning virtual screening pipeline prioritized a commercial library for beta-haematin and parasite assays. The primary indexed paper reports several chemotypes evaluated for antiplasmodial activity, beta-haematin inhibition and haemozoin formation inhibition. This is an in-vitro campaign; results should not be conflated across assays or with clinical efficacy.",
    "field": "medicine",
    "ai_technology": "PCA enrichment mapping; support-vector machine classification; virtual screening",
    "organization": "University of Cape Town",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42857431/",
    "source_type": "paper",
    "publication_date": "2026-09-29",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "PMC-hosted primary text reports PCA/ML screen and experimental parasite/haemozoin assays.",
      "unknown": "Exact structure-level overlaps with earlier UCT screens and detailed full-text quantitative results should be cross-checked before final prioritization.",
      "ai_role": "PCA map and classifier prioritize compounds for biochemical and cell assays.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42857431/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "PMC-indexed primary article; PubMed primary metadata. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2024-02-selective-bias-labmol-394-antimalarial-screen",
    "title": "Selective-bias screening nominates LabMol-394 antimalarial candidates",
    "summary": "A machine-learning workflow guided virtual screening for parasite N-myristoyltransferase, followed by testing of selected compounds in parasite assays. The primary preprint reports a bounded set of measured antimalarial candidates, including LabMol-394, while target assignment remains computational. This is not clinical efficacy or proof of target engagement.",
    "field": "medicine",
    "ai_technology": "Selective-bias virtual screening; machine-learning target/activity models",
    "organization": "Federal University of Goiás; University of Gothenburg",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10925453/",
    "source_type": "paper",
    "publication_date": "2024-02-20",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Primary preprint full text reports tested selected candidates and parasite activity.",
      "unknown": "Exact earliest journal version and compound prior art remain open; target-specific binding is not experimentally established.",
      "ai_role": "Model-based selective-bias workflow scores target/activity candidates for experimental testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10925453/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "PMC-indexed primary preprint full text; PubMed primary record/abstract and affiliations. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2021-03-qsar-natural-products-antimalarial-hits",
    "title": "QSAR screening identifies antimalarial sesquiterpene lactones",
    "summary": "QSAR virtual screening of a natural-product database prioritized compounds for in-vitro testing against chloroquine-sensitive and multidrug-resistant P. falciparum. The paper reports two sesquiterpene lactones, LDT-597 and LDT-598, with potent and selective activity. ADME/PBPK predictions are not measured pharmacokinetics, and the assays do not establish clinical efficacy.",
    "field": "medicine",
    "ai_technology": "QSAR activity models; QSPR ADME and PBPK predictors",
    "organization": "University of Campinas; Federal University of Goiás",
    "source_url": "https://doi.org/10.3390/biom11030459",
    "source_type": "paper",
    "publication_date": "2021-03-19",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports experimental validation of two prioritized compounds against parasite strains and human hepatoma cells.",
      "unknown": "Compound-level prior art and animal efficacy remain open; predicted PK is computational, not measured.",
      "ai_role": "QSAR screen prioritizes commercial natural products/derivatives for parasite and cytotoxicity assays.",
      "evidence_urls": [
        "https://doi.org/10.3390/biom11030459"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "publisher primary full text. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2023-11-maip-high-throughput-antimalarial-screen",
    "title": "MAIP enriches high-throughput screening for antimalarial hits",
    "summary": "The MAIP malaria-activity predictor was combined with a compound-selection workflow and high-throughput cascade to prioritize public-library compounds. The paper reports eight selected hits with potency and ADME profiles and 12-fold enrichment versus a random set. That enrichment is study-specific; private training-set overlap and chemical novelty remain unresolved.",
    "field": "medicine",
    "ai_technology": "MAIP machine-learning antimalarial activity predictor",
    "organization": "European Molecular Biology Laboratory, European Bioinformatics Institute; Medicines for Malaria Venture",
    "source_url": "https://doi.org/10.1021/acsmedchemlett.3c00369",
    "source_type": "paper",
    "publication_date": "2023-11-20",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports selected hits, experimental screening and 12-fold enrichment versus random selection.",
      "unknown": "Underlying partner data are partly private; selection and training overlap and exact compound prior art require review; not clinical evidence.",
      "ai_role": "Predictor enriches candidate pool, with staged workflow and human selection preceding experimental screening.",
      "evidence_urls": [
        "https://doi.org/10.1021/acsmedchemlett.3c00369"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "publisher primary metadata/abstract; author-hosted primary full paper. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2024-09-ml-predicted-approved-antileishmanial-drugs",
    "title": "Machine-learning-predicted approved drugs show in-vitro Leishmania activity",
    "summary": "Two classifiers previously prioritized approved drugs from a screened set; ten candidates were purchased for assays against L. major and L. infantum. Five showed dose-dependent effects; intracellular testing and macrophage cytotoxicity were also reported. Domperidone and dibucaine had prior antiparasitic evidence, so the bounded result concerns the study’s remaining tested drugs and assays, not ten novel therapies.",
    "field": "medicine",
    "ai_technology": "Random forest and support-vector machine classifiers",
    "organization": "Institut Pasteur de Tunis; Université de Tunis El Manar",
    "source_url": "https://doi.org/10.3389/fcimb.2024.1403589",
    "source_type": "paper",
    "publication_date": "2024-09-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary full text describes ten tested predicted candidates, three Leishmania strains, intracellular testing and host-cell counter-assays.",
      "unknown": "Earlier drug-specific literature, concentrations, achievable exposure and clinical efficacy require separate assessment.",
      "ai_role": "Previously trained activity models selected FDA-approved drugs for experimental parasite and host-cell testing.",
      "evidence_urls": [
        "https://doi.org/10.3389/fcimb.2024.1403589"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "publisher primary full text. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2026-03-graph-embedding-lc61-leishmania",
    "title": "Graph embedding identifies LC-61 with anti-Leishmania activity",
    "summary": "A multiscale graph embedding workflow prioritized LC-61, which the study experimentally evaluated against L. infantum. The finding is a specific in-vitro compound result; it does not establish clinical efficacy, safety or a validated target.",
    "field": "medicine",
    "ai_technology": "Multiscale-aware graph embedding and machine learning",
    "organization": "Federal University of Goiás",
    "source_url": "https://doi.org/10.1021/acs.jcim.5c02947",
    "source_type": "paper",
    "publication_date": "2026-03-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article/PubMed and institutional manuscript identify LC-61 and its experimental parasite assay.",
      "unknown": "Exact prior structures and broader dose/replication details remain open.",
      "ai_role": "Graph embedding ranks candidate molecules for assay; human selection and experimental testing follow.",
      "evidence_urls": [
        "https://doi.org/10.1021/acs.jcim.5c02947"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "author-institution primary full-text PDF; PubMed primary metadata. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2026-01-tranylcypromine-leishmania-major-assays",
    "title": "Tranylcypromine shows in-vitro activity against Leishmania major",
    "summary": "The study experimentally tested the previously model-predicted drug tranylcypromine against L. major promastigotes and intracellular amastigotes. It reports IC50 values of 83.6 and 31.6 µg/mL, respectively, and macrophage CC50 of 282.3 µg/mL. These relatively high in-vitro concentrations do not establish achievable exposure, patient safety or clinical efficacy; human monoamine oxidase mechanism is not shown as the parasite target.",
    "field": "medicine",
    "ai_technology": "Prior machine-learning drug-repurposing pipeline (the present study is experimental validation)",
    "organization": "Institut Pasteur de Tunis; Université de Tunis El Manar",
    "source_url": "https://doi.org/10.3389/fmicb.2025.1756231",
    "source_type": "paper",
    "publication_date": "2026-01-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary full text reports promastigote and intracellular amastigote assays with dose responses and host-cell cytotoxicity.",
      "unknown": "Prior drug/species work, tolerability and mechanism remain unresolved; no clinical inference.",
      "ai_role": "Study experimentally validates a model-prioritized repurposing candidate from earlier research.",
      "evidence_urls": [
        "https://doi.org/10.3389/fmicb.2025.1756231"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "publisher primary full text; PubMed primary metadata/abstract. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2023-12-ml-screened-leishmania-dhfr-ts-inhibitors",
    "title": "Machine learning screens cinnamic-acid derivatives for Leishmania enzyme inhibition",
    "summary": "Two random-forest classifiers and docking ranked an in-house set of cinnamic-acid derivatives; selected compounds were tested against recombinant L. major DHFR-TS. The paper reports IC50 values of 7.5, 6.1 and 10.1 µM for three named natural products versus 1.4 µM for methotrexate. These are enzyme assays, not whole-parasite treatment results.",
    "field": "chemistry",
    "ai_technology": "Random-forest classifiers; structure-based docking",
    "organization": "Universidad ECCI",
    "source_url": "https://doi.org/10.3390/molecules29010179",
    "source_type": "paper",
    "publication_date": "2023-12-28",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports recombinant DHFR-TS inhibition for selected compounds and comparator.",
      "unknown": "Methods/results disagree on number of compounds; name/formula spellings and exact prior art need checking.",
      "ai_role": "RF models and docking prioritize a small compound set for biochemical enzyme assays.",
      "evidence_urls": [
        "https://doi.org/10.3390/molecules29010179"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "PMC-hosted primary full text; publisher primary article page. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2025-11-data-centric-ml-trypanosoma-cruzi-screen",
    "title": "Data-centric machine learning prioritizes compounds for Trypanosoma cruzi assays",
    "summary": "Ensemble supervised classifiers trained on curated phenotypic activity data prioritized eight repurposing compounds for clinically relevant parasite-stage assays. The article reports six selected compounds with predicted trypanocidal effects, while identifying cytotoxicity as a caveat for two amastigote candidates; mifepristone showed trypomastigote activity. These are in-vitro strain-specific outcomes, not clinical efficacy or novel molecular entities.",
    "field": "medicine",
    "ai_technology": "Ensemble molecular-descriptor linear classifiers",
    "organization": "CONICET–Universidad de Buenos Aires",
    "source_url": "https://doi.org/10.3390/pharmaceutics17121513",
    "source_type": "paper",
    "publication_date": "2025-11-23",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports eight in-silico hits and selected-stage experimental confirmation; tables qualify cytotoxicity caveats.",
      "unknown": "Prior compound activity and some full model/supplement details remain for later audit; selected repurposed compounds are not novel molecules.",
      "ai_role": "Data-centric supervised ensembles prioritize repurposing candidates for distinct parasite-stage assays.",
      "evidence_urls": [
        "https://doi.org/10.3390/pharmaceutics17121513"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "publisher primary full text and tables. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2021-01-bayesian-schistosoma-screening",
    "title": "Bayesian models prioritize compounds for Schistosoma mansoni assays",
    "summary": "Eight stage/time-specific Bayesian classifiers trained on historical phenotypic measurements guided prospective selection of predicted active and inactive commercial compounds. In-vitro assays reported 48% and 34% hit fractions for somules and adults, followed by triage of five compounds. The selected-set fractions are not unbiased head-to-head screening rates; no animal cure, clinical mechanism, safety or efficacy is established.",
    "field": "medicine",
    "ai_technology": "Bayesian classifiers",
    "organization": "Collaborations Pharmaceuticals",
    "source_url": "https://doi.org/10.1021/acsinfecdis.0c00754",
    "source_type": "paper",
    "publication_date": "2021-01-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports prospective somule/adult screens and prioritized compounds after triage.",
      "unknown": "Compound-specific prior evidence and complete supplementary access remain open; no clinical inference.",
      "ai_role": "Stage/time-specific Bayesian models guided selection of assay compounds, with human curation and triage.",
      "evidence_urls": [
        "https://doi.org/10.1021/acsinfecdis.0c00754"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "PMC-indexed primary text; PubMed primary metadata. Material checked for the specific claims stated in the record."
    }
  },
  {
    "id": "2025-11-rfd2-mi-cysteine-proteases",
    "title": "RFD2-MI designs sequence-specific cysteine proteases",
    "summary": "RFD2-MI designs cysteine-nucleophile proteases that cleave peptide backbones in a sequence-dependent manner. The preprint abstract reports rate enhancements up to 3 × 10^7 and crystal structures close to design models. The journal version reports additional experiments; this draft is bounded to the preprint abstract result and does not transfer later results across versions.",
    "field": "biology",
    "ai_technology": "RFD2-MI; ProteinMPNN; AlphaFold 3",
    "organization": "Institute for Protein Design, University of Washington",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12668006/",
    "source_type": "paper",
    "publication_date": "2025-11-22",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1101/2025.11.21.689808",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broader deployment or clinical benefit are not established by this source review.",
      "ai_role": "RFD2-MI generates catalytic protein structures; sequence design and structure prediction support experimental selection.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12668006/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source abstract and publication metadata reviewed; full text scope is specified in source_checks."
    }
  },
  {
    "id": "2026-10-designed-metalloproteases",
    "title": "Deep-learning design of zinc metalloproteases",
    "summary": "RFdiffusion2 for Molecular Interfaces designs zinc-dependent proteases for peptide-bond hydrolysis. The primary paper reports cleavage at the targeted bond in 36% of 135 designs and selected designs that cleave disease-associated proteins in vitro. These biochemical results do not establish therapeutic efficacy.",
    "field": "biology",
    "ai_technology": "RFdiffusion2 for Molecular Interfaces (RFD2-MI); AlphaFold 3",
    "organization": "Institute for Protein Design, University of Washington",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12667880/",
    "source_type": "paper",
    "publication_date": "2026-10-14",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-026-11167-2",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 5
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broader deployment or clinical benefit are not established by this source review.",
      "ai_role": "RFD2-MI generates scaffolds around zinc catalytic and peptide-binding geometries; computational filtering guided experimental testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12667880/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source abstract and publication metadata reviewed; full text scope is specified in source_checks."
    }
  },
  {
    "id": "2025-09-crest-formate-oxidation-catalyst",
    "title": "CRESt discovers a multielement formate-oxidation catalyst",
    "summary": "CRESt combines multimodal models, knowledge-assisted Bayesian optimization and robotic synthesis and testing. Across more than 900 chemistries and 3,500 electrochemical tests, it identified an eight-element catalyst with a reported 9.3-fold improvement in cost-specific formate-oxidation performance. The metric is cost-specific, not a universal activity comparison.",
    "field": "chemistry",
    "ai_technology": "CRESt multimodal model; knowledge-assisted Bayesian optimization",
    "organization": "Massachusetts Institute of Technology",
    "source_url": "https://www.nature.com/articles/s41586-025-09640-5",
    "source_type": "paper",
    "publication_date": "2025-09-23",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41586-025-09640-5",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broader deployment or clinical benefit are not established by this source review.",
      "ai_role": "The model combines composition, literature text embeddings and microstructural images; Bayesian optimization prioritizes experiments and robotic tests feed results back.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41586-025-09640-5"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source abstract and publication metadata reviewed; full text scope is specified in source_checks."
    }
  },
  {
    "id": "2026-02-pichia-clm-protein-expression",
    "title": "Pichia-CLM improves recombinant protein expression in Komagataella phaffii",
    "summary": "A GRU encoder-decoder language model learns host-specific amino-acid-to-codon patterns and designs synonymous sequences. Across six proteins, the study reports the best titer for five against four commercial optimization tools; human serum albumin reached about threefold the native-sequence titer. Results are plate-scale expression measurements, not manufacturing economics or clinical product outcomes.",
    "field": "biology",
    "ai_technology": "Pichia-CLM (GRU encoder-decoder)",
    "organization": "Massachusetts Institute of Technology",
    "source_url": "https://www.pnas.org/doi/10.1073/pnas.2522052123",
    "source_type": "paper",
    "publication_date": "2026-02-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2522052123",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broader deployment or clinical benefit are not established by this source review.",
      "ai_role": "The learned host codon language model generates optimized coding sequences; researchers experimentally compare protein titers with native sequences and four commercial tools.",
      "evidence_urls": [
        "https://www.pnas.org/doi/10.1073/pnas.2522052123"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source abstract and publication metadata reviewed; full text scope is specified in source_checks."
    }
  },
  {
    "id": "2020-05-hob2-magnetocaloric-effect",
    "title": "HoB2 exhibits a large magnetocaloric response near 15 K",
    "summary": "Machine learning guided selection of HoB2, followed by synthesis and magnetic measurements. The paper reports a magnetic entropy change of 40.1 J kg−1 K−1 for a 5 T field change near a 15 K ferromagnetic transition. The compound was known; the reported discovery is its measured magnetocaloric response, not a demonstrated refrigerator.",
    "field": "materials",
    "ai_technology": "Machine-learning-guided composition screening; XGBoost",
    "organization": "National Institute for Materials Science; University of Tsukuba",
    "source_url": "https://www.nature.com/articles/s41427-020-0214-y.pdf",
    "source_type": "paper",
    "publication_date": "2020-05-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41427-020-0214-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broader deployment or clinical benefit are not established by this source review.",
      "ai_role": "Machine learning ranks candidate compounds for measurement; HoB2 was then synthesized and its magnetocaloric properties measured experimentally.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41427-020-0214-y.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source abstract and publication metadata reviewed; full text scope is specified in source_checks."
    }
  },
  {
    "id": "2024-06-ai-designed-printable-aluminum-alloy",
    "title": "Machine-learning design yields a high-strength printable aluminum alloy",
    "summary": "Hybrid CALPHAD/ICME calculations and Bayesian optimization selected an Al-Er-Zr-Y-Yb-Ni composition that was manufactured as powder and printed by laser powder bed fusion without the cracking reported for Al 7075. The preprint reports 50% greater strength after 400 °C aging than the strongest benchmark printable aluminum alloy. This is a measured material result, not component-scale qualification.",
    "field": "materials",
    "ai_technology": "Hybrid CALPHAD/ICME; neural-network regression; Gaussian-process Bayesian optimization",
    "organization": "Massachusetts Institute of Technology; Carnegie Mellon University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12810664/",
    "source_type": "paper",
    "publication_date": "2024-06-25",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/adma.202509507",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broader deployment or clinical benefit are not established by this source review.",
      "ai_role": "Simulation-derived features train a forward model and Bayesian optimization proposes compositions; the selected alloy is manufactured and mechanically tested.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12810664/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source abstract and publication metadata reviewed; full text scope is specified in source_checks."
    }
  },
  {
    "id": "2024-08-superalloy-extrapolation-solvus",
    "title": "Unsupervised learning guides experimental superalloy design",
    "summary": "A gradient-boosted regressor and unsupervised sampling workflow guided fabrication and differential-scanning-calorimetry measurements of nine superalloys selected outside the initial training compositions. Three alloys had gamma-prime solvus temperatures about 52 °C above the relevant comparators. This is a phase-solvus result, not a blade service-temperature or engine-performance result.",
    "field": "materials",
    "ai_technology": "Gradient-boosting regression; t-SNE sampling; SHAP analysis",
    "organization": "Northwestern Polytechnical University",
    "source_url": "https://www.nature.com/articles/s41524-024-01358-8",
    "source_type": "paper",
    "publication_date": "2024-08-04",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broad deployment are not established by this review.",
      "ai_role": "Supervised prediction is combined with unsupervised extrapolation, interpretability and composition similarity to select alloys for synthesis and measurement.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-024-01358-8"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks."
    }
  },
  {
    "id": "2024-03-mole-antimicrobial-growth-inhibitors",
    "title": "MolE prioritizes existing drugs with measured Staphylococcus aureus growth effects",
    "summary": "MolE-XGBoost prioritized existing compounds. Of six tested commercially available compounds, elvitegravir, opicapone and ebastine showed measured growth-inhibitory effects on S. aureus; visomitin was a known-antimicrobial positive control. Effects varied: elvitegravir inhibited growth at 8 μg/mL, while the other two limited or delayed growth. This is an in-vitro finding, not a clinical antibiotic.",
    "field": "biology",
    "ai_technology": "MolE molecular representation learning; XGBoost",
    "organization": "University of Tübingen",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11986102/",
    "source_type": "paper",
    "publication_date": "2024-03-11",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broad deployment are not established by this review.",
      "ai_role": "A learned molecular representation and XGBoost rank candidates for experimental screening; selected compounds undergo MIC and growth-curve assays.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11986102/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks."
    }
  },
  {
    "id": "2024-11-apexgo-peptide-optimization",
    "title": "ApexGO optimizes antimicrobial peptide activity",
    "summary": "ApexGO uses Bayesian optimization with a peptide variational autoencoder and an activity-prediction oracle to optimize derivatives of ten peptide templates. The primary article reports 100 synthesized sequences tested in vitro and activity improvements; selected candidates showed activity in mouse A. baumannii infection models. These results are preclinical and do not establish human efficacy.",
    "field": "biology",
    "ai_technology": "ApexGO; peptide variational autoencoder; Bayesian optimization; APEX predictor",
    "organization": "University of Pennsylvania",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13201158/",
    "source_type": "paper",
    "publication_date": "2024-11-27",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broad deployment are not established by this review.",
      "ai_role": "A predictor scores generated peptide derivatives and Bayesian optimization uses feedback to propose variants; researchers synthesize and test them in vitro and in mouse models.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13201158/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks."
    }
  },
  {
    "id": "2025-09-diffsyn-ufi-zeolite-synthesis",
    "title": "DiffSyn guides synthesis of a high-silica UFI zeolite",
    "summary": "DiffSyn is a diffusion model trained on synthesis recipes and conditioned on zeolite structure and organic template. Researchers synthesized a UFI material using generated routes and report an ICP Si/Al ratio of 19.0. UFI is an existing framework; the result is a model-guided synthesis route and measured composition, not discovery of a new framework. Expected thermal stability is not presented as a measured outcome.",
    "field": "chemistry",
    "ai_technology": "DiffSyn generative diffusion model",
    "organization": "Massachusetts Institute of Technology",
    "source_url": "https://arxiv.org/abs/2509.17094",
    "source_type": "paper",
    "publication_date": "2025-09-21",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broad deployment are not established by this review.",
      "ai_role": "A diffusion model proposes synthesis routes from structure/template conditioning; researchers synthesize and characterize a candidate material.",
      "evidence_urls": [
        "https://arxiv.org/abs/2509.17094"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks."
    }
  },
  {
    "id": "2025-04-ml-predicts-pancreatic-cancer-drug-synergies",
    "title": "Machine learning nominates synergistic drug combinations in pancreatic cancer cells",
    "summary": "Three research groups used machine-learning models trained on 496 measured PANC-1 drug pairs to nominate additional combinations. Of 88 nominated pairs tested, 51 met the study’s synergy threshold. The article also reports 307 synergistic combinations including the 256-pair training screen; the 307 total is not a count of new AI-nominated hits. The result is limited to cell assays.",
    "field": "medicine",
    "ai_technology": "Random forests; XGBoost; deep neural networks; graph convolutional networks",
    "organization": "National Center for Advancing Translational Sciences; University of North Carolina; Massachusetts Institute of Technology",
    "source_url": "https://www.nature.com/articles/s41467-025-56818-6",
    "source_type": "paper",
    "publication_date": "2025-04-29",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broad deployment are not established by this review.",
      "ai_role": "Models trained on measured drug-combination assays rank candidates from a virtual library; nominated combinations are tested in PANC-1 cell assays.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-56818-6"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks."
    }
  },
  {
    "id": "2026-05-biogdr-aldh1b1-cell-sensitivity",
    "title": "BioGDR predicts cell-line sensitivity to an ALDH1B1 inhibitor",
    "summary": "BioGDR combines molecular structure with learned, predicted gene-expression and kinase-effect features and transcriptomic cell representations. Prospective in-vitro tests confirmed differential sensitivity to Golgicide A-2, including T84 colorectal cells (reported IC50 1.09 μM). This is a cell-line sensitivity result, not evidence of clinical benefit or discovery of the compound itself.",
    "field": "medicine",
    "ai_technology": "BioGDR multimodal interpretable deep learning; pathway graph attention",
    "organization": "Shanghai Institute of Materia Medica, Chinese Academy of Sciences",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13396495/",
    "source_type": "paper",
    "publication_date": "2026-05-13",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and clinical benefit are not established by this source review.",
      "ai_role": "The model integrates predicted compound effects and cell transcriptomes to rank drug responses; the study prospectively tests candidate sensitivity in cell lines.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13396495/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source text reviewed; source_checks specify accessible sections."
    }
  },
  {
    "id": "2025-07-ai-generative-liquid-electrolytes",
    "title": "Generative model identifies high-conductivity liquid electrolyte formulations",
    "summary": "A predictive model and permutation-invariant generative approach were used to design liquid electrolyte mixtures. The primary v3 preprint reports three experimentally identified formulations with high ionic conductivity and anion-concentrated solvation structure. The later journal abstract additionally mentions cycling stability; that claim is excluded here because its version-specific test was not reviewed.",
    "field": "chemistry",
    "ai_technology": "Predictive model; permutation-invariant mixture model; generative design",
    "organization": "ByteDance Seed",
    "source_url": "https://arxiv.org/abs/2504.18728",
    "source_type": "paper",
    "publication_date": "2025-07-29",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and commercial deployment are not established.",
      "ai_role": "Models predict mixture properties and generate candidate formulations; researchers experimentally measure selected mixtures and solvation structure.",
      "evidence_urls": [
        "https://arxiv.org/html/2504.18728v3"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article/preprint text and metadata as listed in source_checks."
    }
  },
  {
    "id": "2026-03-deep-active-learning-lithium-electrolytes",
    "title": "Deep active learning identifies lithium-metal battery electrolytes with longer symmetric-cell lifetime",
    "summary": "A two-stage framework combines deep-kernel active learning with transfer to expanded formulation spaces. The primary article reports electrolytes that increased average lifetime of lithium-metal symmetric cells threefold after three learning iterations and improved formulations in lithium-metal full cells. These are study-specific battery tests, not commercial-pack or independent replication evidence.",
    "field": "materials",
    "ai_technology": "Deep-kernel active learning; target statistic coding; knowledge transfer",
    "organization": "Dalhousie University; University of Maryland",
    "source_url": "https://www.nature.com/articles/s41467-026-70973-4",
    "source_type": "paper",
    "publication_date": "2026-03-27",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and commercial deployment are not established.",
      "ai_role": "Deep active learning selects informative experiments and models formulation performance; statistic coding transfers learned information to expanded composition spaces.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-70973-4"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary article/preprint text and metadata as listed in source_checks."
    }
  },
  {
    "id": "2023-07-machine-learning-braid-invariants",
    "title": "Machine learning guides proofs of braid and flat-braid invariants",
    "summary": "A one-hidden-layer multilayer perceptron classified examples of three-strand braids and suggested algebraic criteria. Authors converted the observed patterns into proofs. The paper proves a complete criterion for triviality of flat braids on three strands; its result does not establish a complete invariant for arbitrary braids or the neural network itself as a proof system.",
    "field": "mathematics",
    "ai_technology": "Supervised multilayer perceptron",
    "organization": "University of Liverpool; University of Essex",
    "source_url": "https://arxiv.org/html/2307.12185",
    "source_type": "paper",
    "publication_date": "2023-07-22",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The source reports the specific mathematical or algorithmic result summarized here.",
      "unknown": "The result is attributed to the paper; no independent proof review beyond the authors’ source text is claimed.",
      "ai_role": "The MLP classified examples and exposed patterns used by human authors to formulate and prove exact invariants.",
      "evidence_urls": [
        "https://arxiv.org/html/2307.12185"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary preprint full text or abstract reviewed; evidence scope is specified in source_checks."
    }
  },
  {
    "id": "2023-04-ml-ribbon-knot-certificates",
    "title": "Machine learning finds ribbon-knot certificates and rules out candidate pairs",
    "summary": "Bayesian optimization and reinforcement learning search for sequences of knot moves that certify a knot is ribbon. The primary v1 reports that both knots in 843 specified pairs were shown ribbon, eliminating those pairs as potential counterexamples in the studied strategy. It also recovers a separately independently established set of 1,705 ribbon knots; these are not claimed as novel. This does not resolve the smooth four-dimensional Poincaré conjecture.",
    "field": "mathematics",
    "ai_technology": "Bayesian optimization; reinforcement learning",
    "organization": "Dublin Institute for Advanced Studies; Northeastern University; Stanford University",
    "source_url": "https://arxiv.org/html/2304.09304v1",
    "source_type": "paper",
    "publication_date": "2023-04-18",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The source reports the specific mathematical or algorithmic result summarized here.",
      "unknown": "The result is attributed to the paper; no independent proof review beyond the authors’ source text is claimed.",
      "ai_role": "The algorithms search move sequences; a successful path is a verifiable certificate that authors use to prove ribbon status.",
      "evidence_urls": [
        "https://arxiv.org/html/2304.09304v1"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary preprint full text or abstract reviewed; evidence scope is specified in source_checks."
    }
  },
  {
    "id": "2024-04-ai-guided-six-colorings",
    "title": "Machine learning guides new six-colorings of the Euclidean plane",
    "summary": "A custom machine-learning approach suggested constructions that authors formalized as two six-colorings. The primary source reports an off-diagonal distance range of 0.354 to 0.657, extending the previously known range. This concerns a specified off-diagonal variant and does not solve the original Hadwiger–Nelson problem.",
    "field": "mathematics",
    "ai_technology": "Custom machine-learning approach; neural networks; differentiable optimization",
    "organization": "Technical University of Berlin",
    "source_url": "https://arxiv.org/abs/2404.05509",
    "source_type": "paper",
    "publication_date": "2024-04-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The source reports the specific mathematical or algorithmic result summarized here.",
      "unknown": "The result is attributed to the paper; no independent proof review beyond the authors’ source text is claimed.",
      "ai_role": "Neural optimization suggests geometric configurations; authors formalize valid colorings and prove the construction.",
      "evidence_urls": [
        "https://arxiv.org/abs/2404.05509",
        "https://proceedings.mlr.press/v267/mundinger25a.html"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary preprint full text or abstract reviewed; evidence scope is specified in source_checks."
    }
  },
  {
    "id": "2025-01-ai-assisted-magl-inhibitor-optimization",
    "title": "Deep learning helps prioritize potent MAGL inhibitor candidates",
    "summary": "Deep graph neural networks trained on 13,490 Minisci reaction outcomes predicted reaction feasibility for a virtual library of 26,375 compounds derived from moderate MAGL inhibitors. Combined with physicochemical and structure-based scoring, this workflow prioritized 212 candidates; 14 were synthesized and showed subnanomolar MAGL inhibition, with reported potency improvements up to 4,500-fold over the starting hit. This is an AI-assisted, human-selected laboratory optimization result, not an autonomous workflow or a therapeutic outcome.",
    "field": "medicine",
    "ai_technology": "Deep graph neural networks for reaction outcome prediction; multi-objective molecular optimization",
    "organization": "F. Hoffmann-La Roche",
    "source_url": "https://www.cambridge.org/engage/chemrxiv/article-details/6798ae2881d2151a024590d5",
    "source_type": "paper",
    "publication_date": "2025-01-29",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-66324-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Fourteen selected synthesized compounds were reported with subnanomolar MAGL inhibition and co-crystal structures for three ligands.",
      "unknown": "Independent replication, clinical activity and version 1 claim correspondence were not established.",
      "ai_role": "Reaction-prediction models trained on high-throughput Minisci experiments contributed to candidate feasibility scoring; other computational scores and human selection also shaped the experimental set.",
      "evidence_urls": [
        "https://www.cambridge.org/engage/chemrxiv/article-details/6798ae2881d2151a024590d5"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary ChemRxiv v2 abstract and version metadata reviewed; full manuscript and supplements not inspected."
    }
  },
  {
    "id": "2024-07-transfer-learning-oer-perovskites",
    "title": "Transfer learning guides experimental discovery of oxygen-evolution perovskites",
    "summary": "A pretrained model, ensemble learning and active learning screened 16,050 compositions. The paper reports synthesis of 36 predicted oxides, including 13 pure perovskite structures. Two selected Pr–Sr–Co–Fe(-Mn) materials showed overpotentials of 327 and 315 mV at 10 mA/cm² under the authors’ alkaline three-electrode protocol. These values are assay-specific, not device performance.",
    "field": "materials",
    "ai_technology": "Pretrained model; ensemble learning; active learning",
    "organization": "Xiamen University; University of Liverpool; Shenzhen University",
    "source_url": "https://www.nature.com/articles/s41467-024-50605-5",
    "source_type": "paper",
    "publication_date": "2024-07-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary article reports the study-specific material and measurement summarized here.",
      "unknown": "Independent replication and commercial deployment are not established.",
      "ai_role": "Transfer learning and active-learning selection guide candidate synthesis; authors measured OER activity and characterized the catalysts.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-50605-5"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher indexed full text/abstract and version metadata reviewed."
    }
  },
  {
    "id": "2025-06-ferroai-ferroelectric-morphotropic-boundary",
    "title": "FerroAI guides discovery of a ferroelectric composition with high dielectric response",
    "summary": "FerroAI, trained on text-mined phase-transition data, predicted composition–temperature phase behavior and a morphotropic phase boundary in a Zr/Hf co-doped barium titanate–BCT system. Researchers synthesized and characterized compositions near the predicted boundary. The x = 0.3 sample showed dielectric constant 9,535 at 1 kHz and 11,051 at 100 Hz at the transition temperature. The frequency-specific measured values should not be interchanged.",
    "field": "materials",
    "ai_technology": "FerroAI deep learning model; NLP-extracted phase-transition dataset",
    "organization": "Tongji University; Hong Kong University of Science and Technology",
    "source_url": "https://www.nature.com/articles/s41524-025-01778-0",
    "source_type": "paper",
    "publication_date": "2025-06-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary article reports the study-specific material and measurement summarized here.",
      "unknown": "Independent replication and commercial deployment are not established.",
      "ai_role": "A deep model trained on literature-mined phase-transition data predicts phase boundaries; those predictions guide synthesis and dielectric/thermal characterization.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01778-0",
        "https://arxiv.org/abs/2506.10970"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher indexed full text/abstract and version metadata reviewed."
    }
  },
  {
    "id": "2026-02-perovskite-r1-solar-cell-additives",
    "title": "Perovskite-R1 suggests precursor additives tested in solar-cell devices",
    "summary": "A domain-tuned QwQ-32B model generated precursor-additive suggestions. Researchers tested selected recommendations in perovskite solar-cell devices against controls and manually selected additives and report improved device performance and material stability in their tested conditions. Two tested suggestions were 3,5-difluoropyridine-2-carboxylic acid and 5-hydroxy-2-methylbenzoic acid. This does not establish industrial operating lifetime or broad superiority over expert-selected additives.",
    "field": "materials",
    "ai_technology": "Perovskite-R1, instruction-tuned QwQ-32B large language model",
    "organization": "Renmin University of China",
    "source_url": "https://www.nature.com/articles/s43246-026-01099-9",
    "source_type": "paper",
    "publication_date": "2026-02-10",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary article reports the study-specific material and measurement summarized here.",
      "unknown": "Independent replication and commercial deployment are not established.",
      "ai_role": "The domain-tuned LLM proposes precursor additives; researchers design, synthesize and test selected additives in devices and stability experiments.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43246-026-01099-9"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher indexed full text/abstract and version metadata reviewed."
    }
  },
  {
    "id": "2026-01-cleavenet-mmp13-substrates",
    "title": "CleaveNet designs peptide substrates with MMP13 cleavage profiles",
    "summary": "CleaveNet combines a generative model conditioned on protease cleavage profiles with a predictor that ranks candidate sequences. In an in-vitro screen of 95 peptide substrates against 12 matrix metalloproteinases, the authors report that designed substrates were cleaved by MMP13 and identify substrates with different efficiency/selectivity trade-offs. The panel includes training controls and baseline candidates; it is not 95 novel AI discoveries. These are laboratory substrates, not validated diagnostic or therapeutic products.",
    "field": "biology",
    "ai_technology": "CleaveNet generator and predictor; deep learning",
    "organization": "Microsoft Research",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12775482/",
    "source_type": "paper",
    "publication_date": "2026-01-06",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports the specific in-vitro protein assay summarized here.",
      "unknown": "Clinical or diagnostic deployment and independent replication are not established.",
      "ai_role": "A generator proposes peptide sequences conditioned on cleavage profiles; a model ranks candidates and researchers test synthetic substrates against recombinant enzymes.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12775482/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Full primary journal article text, results and experimental assays reviewed. The preprint version-specific claim correspondence was not checked."
    }
  },
  {
    "id": "2026-04-aggreprot-linb-solubility",
    "title": "AggreProt-guided mutations improve LinB solubility and yield",
    "summary": "AggreProt is a deep neural-network predictor of residue-level protein aggregation propensity. Researchers used its predicted aggregation-prone regions to guide mutations in haloalkane dehalogenase LinB and report reduced aggregation, improved soluble yield and up to a 100% increase in yield over wild type under their assay. This is one engineered-enzyme result; model validation on selected peptides/proteins is ancillary, and the result does not establish industrial biodegradation performance.",
    "field": "biology",
    "ai_technology": "AggreProt deep neural-network predictor",
    "organization": "Masaryk University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13328301/",
    "source_type": "paper",
    "publication_date": "2026-04-28",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports the specific in-vitro protein assay summarized here.",
      "unknown": "Clinical or diagnostic deployment and independent replication are not established.",
      "ai_role": "The model identifies residue-level aggregation regions; predictions inform mutations that are experimentally assessed for aggregation, solubility and yield.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13328301/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary paper indexed full text and publication metadata reviewed."
    }
  },
  {
    "id": "2026-06-spufz1-deep-learning-engineering",
    "title": "Deep learning guides higher-activity SpuFz1 nuclease variants",
    "summary": "A DeepSequence-based predictor ranked SpuFz1 mutations without experimental activity labels; experimentally confirmed substitutions were combined into SpuFz1-v3. At the B2M locus in HEK293T cells, the article reports 35.52% indel formation versus 3.07% for wild type. Comparative sequence analysis, not this AI model, guided the separate ultrashort RNA design. The reported activity is assay- and locus-specific; off-target and clinical safety are not established.",
    "field": "biology",
    "ai_technology": "DeepSequence-based Fanzor-Fitness Predictor",
    "organization": "Chinese Academy of Sciences",
    "source_url": "https://www.nature.com/articles/s41467-026-74624-6",
    "source_type": "paper",
    "publication_date": "2026-06-18",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific measured result summarized here.",
      "unknown": "Independent replication and clinical or industrial efficacy are not established.",
      "ai_role": "A learned sequence fitness predictor ranks candidate mutations; researchers assay single substitutions and combine successful variants.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-74624-6"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source indexed full text or abstract and publication metadata reviewed; scope stated in source_checks."
    }
  },
  {
    "id": "2025-11-thermophilic-metagenomic-carbonic-anhydrases",
    "title": "Machine learning-guided discovery of thermophilic carbonic anhydrases",
    "summary": "AdaBoost and LightGBM classifiers applied to sequence descriptors prioritized three gamma carbonic anhydrases from Fang Hot Spring metagenomic data. The recombinant enzymes showed carbonic anhydrase activity and melting temperatures from 97.0 to 109.1 °C in the reported assays. Some candidates are similar to annotated proteins; the contribution is discovery and testing of thermostable enzymes, not new protein families or demonstrated industrial performance.",
    "field": "biology",
    "ai_technology": "AdaBoost; LightGBM",
    "organization": "National Center for Genetic Engineering and Biotechnology; Mahidol University; Thammasat University",
    "source_url": "https://www.nature.com/articles/s41598-025-24713-1",
    "source_type": "paper",
    "publication_date": "2025-11-20",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific measured result summarized here.",
      "unknown": "Independent replication and clinical or industrial efficacy are not established.",
      "ai_role": "Classifiers rank candidate metagenomic enzyme sequences for expression; researchers express, purify and assay carbonic anhydrase activity and thermal stability.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41598-025-24713-1"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source indexed full text or abstract and publication metadata reviewed; scope stated in source_checks."
    }
  },
  {
    "id": "2026-03-parallel-bayesian-optimization-enzyme-nanohybrids",
    "title": "Bayesian optimization identifies enzyme nanohybrids with high activity recovery",
    "summary": "Parallelized hybrid-space Bayesian optimization prioritized metal/ligand carriers and conditions for enzyme immobilization. The primary abstract reports activity recovery of 100% for glucose oxidase, 90% for catalase and 79% for Candida antarctica lipase B across the demonstrated workflow. These are enzyme-specific assay outcomes, not discovery of new enzymes or proof of industrial stability.",
    "field": "chemistry",
    "ai_technology": "Parallelized hybrid-space Bayesian optimization (PHBO)",
    "organization": "Tsinghua University; Synthera Biotechnology Co. Ltd.",
    "source_url": "https://www.nature.com/articles/s41467-026-70251-3",
    "source_type": "paper",
    "publication_date": "2026-03-07",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific measured result summarized here.",
      "unknown": "Independent replication and clinical or industrial efficacy are not established.",
      "ai_role": "The optimizer proposes carrier compositions and experimental conditions, then experimental activity feedback informs later proposals.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-70251-3"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source indexed full text or abstract and publication metadata reviewed; scope stated in source_checks."
    }
  },
  {
    "id": "2025-07-venusmine-kbpetase",
    "title": "VenusMine identifies PET hydrolase KbPETase",
    "summary": "VenusMine combines structure-based search with protein-language-model representations and learned property filters to prioritize candidate PET hydrolases from large sequence collections. The authors tested 34 selected proteins and report PET-hydrolyzing activity for 14; KbPETase had an optimum activity temperature 32 °C above the IsPETase comparator. This is laboratory enzyme evidence, not demonstrated waste-processing performance. A 2024 preprint reportedly contains an earlier version; its exact version-to-result chronology was not freshly resolved, so this draft uses the journal version date and does not claim first discovery priority.",
    "field": "biology",
    "ai_technology": "VenusMine; ProstT5; ESM2",
    "organization": "Shanghai Jiao Tong University; Shanghai Artificial Intelligence Laboratory",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12228687/",
    "source_type": "paper",
    "publication_date": "2025-07-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-61599-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "34 selected proteins were tested; the source reports PET-hydrolyzing activity for 14 and characterizes KbPETase experimentally.",
      "unknown": "Scale-up, real-world plastic-waste processing, independent replication and exact relation of the earlier preprint’s experiments remain outside this bounded review.",
      "ai_role": "VenusMine uses structure search, protein-language-model embeddings and learned property predictions to select enzyme candidates for experimental testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12228687/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Indexed primary article text including title, methods, results and publication metadata; direct PMC open triggered CAPTCHA."
    }
  },
  {
    "id": "2026-02-etopt-thermophilic-beta-agarases",
    "title": "ETopt-guided screening identifies thermophilic beta-agarases",
    "summary": "ETopt combines sequence, structural and protein-language-model features in an ensemble XGBoost model to predict enzyme optimum temperatures. The authors report wet-lab screening that identified two novel thermophilic beta-agarases. The draft is limited to those source-identified enzyme findings; model benchmarks and other screened enzyme classes are not counted as separate discoveries. Industrial process performance and independent replication are not established.",
    "field": "biology",
    "ai_technology": "ETopt multimodal model; ensemble XGBoost; ProtBERT and Prime embeddings",
    "organization": "Ocean University of China",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S0141813026009943",
    "source_type": "paper",
    "publication_date": "2026-02-24",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.ijbiomac.2026.151068",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The source highlights experimental identification of two novel thermophilic beta-agarases; detailed enzyme activity and comparator data were not available in the reviewed abstract.",
      "unknown": "Full article, exact assay results, sequences, prior enzyme overlap and independent replication were not reviewed.",
      "ai_role": "ETopt integrates learned sequence and structural representations with selected features to predict optimum temperature and prioritize candidate enzymes for wet-lab validation.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S0141813026009943",
        "https://pubmed.ncbi.nlm.nih.gov/41747980/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher highlights and abstract; PubMed primary metadata and abstract."
    }
  },
  {
    "id": "2023-09-extrapolative-ml-rwgs-catalyst",
    "title": "Extrapolative machine learning identifies a multielement RWGS catalyst",
    "summary": "In a closed loop combining machine-learning prediction with experiments, researchers tested 300 reverse water-gas shift catalyst formulations over 44 cycles, starting from 45 data points. They report Pt(3)/Rb(1)-Ba(1)-Mo(0.6)-Nb(0.2)/TiO₂ as the optimal tested composition; Nb was absent from the initial dataset. The result is specific to the study’s catalyst tests and conditions, not commercial CO₂ conversion. The earlier ChemRxiv version is not used to claim priority here.",
    "field": "chemistry",
    "ai_technology": "Extrapolative machine learning; closed-loop prediction and experiment",
    "organization": "Hokkaido University; Kogakuin University; RIKEN; ENEOS Corporation",
    "source_url": "https://www.nature.com/articles/s41467-023-41341-3.pdf",
    "source_type": "paper",
    "publication_date": "2023-09-21",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-41341-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The source reports 300 experimentally tested catalysts over 44 ML-plus-experiment cycles and identifies the best tested formulation.",
      "unknown": "Commercial operation, independent replication and performance beyond reported experimental conditions are not established.",
      "ai_role": "Machine-learning predictions guide sequential catalyst selection; experimental results update the closed-loop search.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-023-41341-3.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher PDF indexed primary full text and publication metadata; direct HTML endpoint blocked by cookie redirect."
    }
  },
  {
    "id": "2023-11-deep-learning-cell-free-antimicrobial-peptides",
    "title": "Deep-learning-guided cell-free screening identifies antimicrobial peptides",
    "summary": "Generative and predictive deep-learning models proposed and prioritized peptide sequences for cell-free synthesis and screening. The authors report 30 functional antimicrobial peptides among 500 candidates tested across five design–build–test rounds. The assay tested peptides in unpurified cell-free mixtures, a limitation for interpreting activity and potency. The reported effects do not establish clinical efficacy or antibiotic approval.",
    "field": "biology",
    "ai_technology": "Variational autoencoders; convolutional and recurrent neural-network regressors",
    "organization": "Max Planck Institute for Terrestrial Microbiology; Goethe University Frankfurt; Philipps University of Marburg",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/37938588/",
    "source_type": "paper",
    "publication_date": "2023-11-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-42434-9",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "500 model-prioritized candidates were produced and screened; 30 showed functional antimicrobial activity in the reported assays.",
      "unknown": "Unpurified CFPS assay mixtures, animal/human efficacy and independent replication limit interpretation.",
      "ai_role": "Generative models proposed peptide sequences and learned regressors prioritized candidates before cell-free synthesis and activity tests.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/37938588/",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10632401/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PubMed record, abstract and figure captions; indexed full-text results for assay conditions."
    }
  },
  {
    "id": "2026-03-m3cad-qlx3dv-antimicrobial-peptides",
    "title": "M3-CAD designs antimicrobial QLX-3DV peptides",
    "summary": "M3-CAD combines sequence, predicted structural and physicochemical features in a multimodal peptide-design pipeline. The journal study reports in-vitro activity for selected QLX-3DV peptides, including untemplated candidates, against tested bacterial strains; selected leads also received additional host-toxicity and mouse wound-infection assays. These are selected preclinical results, not evidence of clinical efficacy or safety. The article cites an earlier M3-CAD/QL-AMP-1 preprint; this entry is restricted to the reported QLX-3DV study result and does not claim the pipeline itself as new.",
    "field": "biology",
    "ai_technology": "M3-CAD multimodal deep-learning pipeline",
    "organization": "Qilu Hospital of Shandong University; Chinese University of Hong Kong, Shenzhen",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13185819/",
    "source_type": "paper",
    "publication_date": "2026-03-09",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202515835",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Selected designed peptides were synthesized and tested in vitro; selected leads were also examined in mouse wound infection and host-toxicity assays.",
      "unknown": "Clinical efficacy and safety, independent replication and full sequence-level comparison with the earlier M3-CAD preprint are not established here.",
      "ai_role": "M3-CAD uses multimodal learned sequence and predicted-structure representations with activity-related objectives to generate and rank peptides for experimental testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13185819/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PMC full text including publication metadata, methods and experimental results."
    }
  },
  {
    "id": "2025-11-tarsa-breast-cancer-active-peptides",
    "title": "TARSA prioritizes peptides active against breast cancer cells",
    "summary": "TARSA combines reinforcement learning and posterior sampling to search a 36-million-sequence library of experimentally resolved helical peptides. The study reports that 15 of 105 selected candidates showed cytotoxic activity against triple-negative breast cancer cells in vitro; the three leads were further tested against other breast cancer subtypes and peripheral blood cells. The source also reports toxicity against normal breast cells, so these findings do not establish therapeutic safety or clinical efficacy.",
    "field": "medicine",
    "ai_technology": "TARSA reinforcement learning; MCMC posterior sampling; learned cytotoxicity models",
    "organization": "University of British Columbia; University of Ottawa; Mila, Université de Montréal; Simon Fraser University",
    "source_url": "https://www.nature.com/articles/s41467-025-66748-y",
    "source_type": "paper",
    "publication_date": "2025-11-27",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-66748-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "105 selected peptides were experimentally evaluated; 15 showed activity in the reported breast-cancer-cell assays.",
      "unknown": "Cell-culture results do not establish therapeutic efficacy; the source reports normal-breast-cell toxicity, and independent replication remains unknown.",
      "ai_role": "TARSA reinforcement learning and posterior sampling guide exploration of a very large peptide library, with learned potency/toxicity models used to select candidates for assays.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-66748-y"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher full article including publication date, model description, methods and experimental results."
    }
  },
  {
    "id": "2024-05-ml-designed-lead-free-photocatalyst",
    "title": "ML-guided mechanosynthesis yields a lead-free photocatalyst",
    "summary": "Machine-learning-guided synthesis and testing identified a lead-free metal-halide perovskite for oxidative alkene cleavage. The publisher reports an 80% benzoic-acid yield under the study’s conditions, compared with 6% for the initial Cs2AgBiBr6 benchmark. This is a formulation- and reaction-specific laboratory result, not universal photocatalytic performance or industrial deployment.",
    "field": "chemistry",
    "ai_technology": "Machine-learning-assisted materials selection and iterative synthesis",
    "organization": "Nanyang Technological University",
    "source_url": "https://advanced.onlinelibrary.wiley.com/doi/abs/10.1002/advs.202309714",
    "source_type": "paper",
    "publication_date": "2024-05-28",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202309714",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher article reports an experimentally tested ML-selected perovskite yielding 80% benzoic acid versus 6% for the initial benchmark in the specified reaction.",
      "unknown": "Independent replication and broader application are not established by this bounded review.",
      "ai_role": "ML ranks perovskite candidates using experimental feedback across synthesis rounds; researchers synthesize and test the selected photocatalysts.",
      "evidence_urls": [
        "https://advanced.onlinelibrary.wiley.com/doi/abs/10.1002/advs.202309714"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher indexed primary full text and metadata; specific formulation and selected result reviewed. Earlier disclosure not established in this pass."
    }
  },
  {
    "id": "2024-11-ml-selected-silver-co2-electrolysis-conditions",
    "title": "Machine learning selects silver-electrode CO₂ electrolysis conditions",
    "summary": "A model trained on 56 experimental observations selected electrode-fabrication and electrolysis conditions for silver-catalyzed CO₂ reduction. Researchers experimentally validated the proposed conditions against a desired hydrogen-to-carbon-monoxide ratio and CO efficiency. This is an assay-specific process-optimization result for a known silver catalyst, not a newly discovered catalyst or evidence of industrial carbon-neutral operation.",
    "field": "chemistry",
    "ai_technology": "Random-forest regression; kernel ridge regression",
    "organization": "Osaka University; Tohoku University",
    "source_url": "https://d-nb.info/1349018996/34",
    "source_type": "paper",
    "publication_date": "2024-11-21",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/celc.202400518",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports 56 experimental observations and experimental validation of selected conditions for the target product ratio and CO faradaic efficiency.",
      "unknown": "Independent replication and performance outside stated assay conditions are not established.",
      "ai_role": "ML models trained on experimental outcomes estimate selectivity across fabrication and electrolysis parameters; inverse search proposes conditions for follow-up testing.",
      "evidence_urls": [
        "https://d-nb.info/1349018996/34"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary author manuscript PDF read through abstract, results and conditions; exact online date from source ledger, issue metadata says 2024."
    }
  },
  {
    "id": "2026-02-ml-guided-fecomcrmncu-ldh-oer-catalyst",
    "title": "ML-guided FeCoCrMnCu hydroxide for oxygen evolution",
    "summary": "An XGBoost model trained on measurements from 70 Fe–Co–Cr–Mn–Cu compositions selected a layered double hydroxide for oxygen evolution. The reported composition had a measured overpotential of 270 mV at 10 mA cm⁻², versus a model prediction of 261.3 mV; the paper also reports a 72-hour stability test under its protocol. This comparison does not establish universal catalyst superiority or full-cell performance.",
    "field": "materials",
    "ai_technology": "XGBoost regression",
    "organization": "National Chung Hsing University",
    "source_url": "https://doi.org/10.1021/acscatal.5c07303.s001",
    "source_type": "paper",
    "publication_date": "2026-02-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acscatal.5c07303",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher supporting information and indexed abstract identify the optimal Fe0.15Co0.10Cr0.30Mn0.30Cu0.15 composition, experimentally measured 270 mV overpotential, and 72 h operation.",
      "unknown": "Independent replication and broader application are not established by this bounded review.",
      "ai_role": "XGBoost trained on 70 experimental compositions guides selection from a much larger composition space; researchers synthesize and characterize the selected catalyst.",
      "evidence_urls": [
        "https://doi.org/10.1021/acscatal.5c07303.s001"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher indexed article abstract and publisher-hosted supporting information; source identity/DOI and results checked. Full article text not accessed."
    }
  },
  {
    "id": "2026-08-reap-guided-p450-bm3-engineering",
    "title": "REAP-guided engineering improves P450 BM3 activity",
    "summary": "REAP combines a protein-language-model ranking predictor with active selection and automated build-and-test cycles. The study reports a P450 BM3 quintuple variant with 57-fold higher product yield than the engineered FL#62 starting scaffold in its high-throughput assay. This is a variant-specific comparison with an already engineered scaffold, not a 57-fold gain over wild type. A separate sortase result is not included in this bounded draft.",
    "field": "biology",
    "ai_technology": "REAP; PLM-RankReg; automated build-and-test",
    "organization": "Tianjin University; Arizona State University",
    "source_url": "https://www.nature.com/articles/s41467-026-76264-2",
    "source_type": "paper",
    "publication_date": "2026-08-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-76264-2",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher indexed article reports a quintuple variant and 57-fold product-yield increase over FL#62 in the high-throughput assay; kinetic characterization is also described.",
      "unknown": "Independent replication and broader application are not established by this bounded review.",
      "ai_role": "Learned ranking and active selection nominate variants for automated synthesis and testing; the reported gain is measured against the engineered FL#62 scaffold.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-76264-2"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher indexed primary article text, title/DOI and result; PMC direct open CAPTCHA challenged. Journal/preprint chronology and sequence overlap not audited."
    }
  },
  {
    "id": "2026-05-ml-ranked-sesquiterpene-synthase-variants",
    "title": "ML ranking identifies active sesquiterpene-synthase variants",
    "summary": "A composite score combines Boltz2-predicted binding and DLKcat-predicted turnover to rank variants in a focused Agr5 library. Of 15 top-ranked variants tested, four produced more viridiflorol than wild-type Agr5; the paper reports 11.05-fold higher production for Y374F and a 952.79 mg/L titer in its yeast assay. The initial V314G design involved mechanism-guided human analysis, and the measured values are assay-specific.",
    "field": "biology",
    "ai_technology": "DeEnzyme_score combining Boltz2 and DLKcat predictions",
    "organization": "Queen’s University",
    "source_url": "https://pubs.acs.org/accacs/article/16/14/13022/5114435/Customizing-Natural-Products-of-Sesquiterpene",
    "source_type": "paper",
    "publication_date": "2026-05-28",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article reports four activity-enhancing variants among 15 tested and specifies the Y374F production ratio and titer in a yeast assay.",
      "unknown": "Prior sequence-level overlap, earlier disclosure and independent replication remain unestablished; the proposed allosteric explanation partly relies on simulation.",
      "ai_role": "Boltz2 and DLKcat predictions are combined to rank variants selected for experimental testing; the initial V314G design is attributed to mechanism-guided analysis, not AI.",
      "evidence_urls": [
        "https://pubs.acs.org/accacs/article/16/14/13022/5114435/Customizing-Natural-Products-of-Sesquiterpene"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher indexed full article text including title, methods, results, article history and author-affiliation block; direct ACS endpoint returned 403."
    }
  },
  {
    "id": "2025-07-ml-selected-light-rare-earth-laves-magnetocaloric-phases",
    "title": "Machine learning guides magnetocaloric Laves-phase synthesis",
    "summary": "Three machine-learning models predicted Curie temperatures to select Nd–Pr–Co–Ni and Pr–Ce–Co–Ni Laves-phase compositions, which were synthesized by arc melting and characterized. The Nd–Pr compositions showed magnetic ordering between 20 and 36 K and reported magnetic entropy changes of 6.0–7.2 J kg⁻¹ K⁻¹ at 5 T. The authors caution that predictions lack precision and do not show superiority over other light-rare-earth compounds; no working hydrogen liquefier is demonstrated.",
    "field": "materials",
    "ai_technology": "Random forest; gradient boosting; neural-network regression",
    "organization": "Institute for Energy Technology (IFE)",
    "source_url": "https://backend.orbit.dtu.dk/ws/portalfiles/portal/408621365/1-s2.0-S1359645425005695-main.pdf",
    "source_type": "paper",
    "publication_date": "2025-07-16",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.actamat.2025.121282",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary full text reports synthesis and characterization, magnetic ordering and magnetocaloric measurements for ML-selected compositions.",
      "unknown": "Independent replication and broader application are not established by this bounded review.",
      "ai_role": "Models estimate Curie temperature to guide composition selection; researchers synthesize and measure selected Laves phases.",
      "evidence_urls": [
        "https://backend.orbit.dtu.dk/ws/portalfiles/portal/408621365/1-s2.0-S1359645425005695-main.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Author-institution repository PDF full text and publisher metadata; composition-level previous-work comparison is bounded, not exhaustive."
    }
  },
  {
    "id": "2025-03-logirx-escitalopram-cardiomyocyte-hypertrophy-pathway",
    "title": "LogiRx predicts and tests an escitalopram pathway in cardiac hypertrophy",
    "summary": "LogiRx, a mechanistic machine-learning model, predicted an off-target serotonin-receptor/PI3Kγ pathway for escitalopram’s antihypertrophic effect. The authors report mechanistic experiments in cultured cardiomyocytes, reduced hypertrophy in a mouse model, and retrospective patient-database associations. The drug and initial compound screen predated LogiRx; patient associations do not establish causation or clinical benefit.",
    "field": "medicine",
    "ai_technology": "LogiRx logic-based mechanistic machine learning",
    "organization": "University of Virginia; University of Colorado Anschutz Medical Campus",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/40035765/",
    "source_type": "paper",
    "publication_date": "2025-03-04",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2420499122",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary abstract reports experimental pathway validation, cardiomyocyte and mouse outcomes, and patient database associations.",
      "unknown": "Independent replication and broader application are not established by this bounded review.",
      "ai_role": "LogiRx combines a mechanistic network and drug-target data to predict drug-induced pathways for experimental follow-up; the compound screen and drugs were prior inputs.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40035765/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PubMed metadata/abstract and publisher full-text PDF indexed result; retrospective human association caveat preserved."
    }
  },
  {
    "id": "2025-09-ml-guided-cysteine-arylation",
    "title": "ML-guided optimization improves photochemical cysteine arylation",
    "summary": "The ChemRxiv v1 reports a two-stage optimization for nickel-catalyzed photochemical cysteine arylation: an embedding-based ranker screens categorical reagent choices, followed by Bayesian optimization of continuous conditions. The journal version reports 88% yield for both optimized conditions; the preprint reports 89% for the shorter second condition. The ranker was initially a weak yield predictor, with better separation after experimental iteration. The study expands a reaction class with earlier precedents; it does not establish invention of cysteine arylation or clinical use.",
    "field": "chemistry",
    "ai_technology": "Embedding-based neural ranking; Gaussian-process Bayesian optimization",
    "organization": "University of Amsterdam",
    "source_url": "https://www.cambridge.org/engage/chemrxiv/article-details/68cc1a7323be8e43d6f4b2fb",
    "source_type": "paper",
    "publication_date": "2025-09-23",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acscatal.6c06459",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The two-stage workflow iteratively tests candidates and optimizes conditions; journal reports 88% yields for both conditions, while v1 reports 89% for the shorter condition.",
      "unknown": "Exact prior conditions and independent replication remain unreviewed; preprint and journal yield discrepancy retained.",
      "ai_role": "Learned embeddings rank categorical reagent conditions; later Bayesian optimization selects continuous conditions using feedback from laboratory experiments.",
      "evidence_urls": [
        "https://www.cambridge.org/engage/chemrxiv/article-details/68cc1a7323be8e43d6f4b2fb"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary ChemRxiv v1 record/date and version content verified in continuation audit; publisher indexed article Methods/Results freshly checked; ACS direct fetch returned 403. Journal exact online-first day remains unverified."
    }
  },
  {
    "id": "2025-12-diffusion-designed-bimetallic-ammonia-decomposition-catalyst",
    "title": "Diffusion-guided design tests bimetallic ammonia-decomposition catalysts",
    "summary": "A diffusion model guided by a learned nitrogen-adsorption predictor generated bimetallic candidates near the target adsorption energy. Co75Mo25 was among the generated candidates and maps exactly to the tested Co3Mo1 ratio; under the reported supported-catalyst assay it reached 18.6 μmol gmetal⁻¹ s⁻¹ at 520 °C, about twice the paper’s Ru reference under its stated test. Co–Mo catalysts are established prior art, so this entry does not claim the first Co–Mo ammonia-decomposition catalyst or global composition novelty. Ex-situ XRD showed oxidized phases after air exposure and does not establish the working active phase.",
    "field": "chemistry",
    "ai_technology": "Diffusion model guided by random-forest nitrogen-adsorption predictor",
    "organization": "Worcester Polytechnic Institute; University of California, Riverside; University of Massachusetts Lowell",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12814173/",
    "source_type": "paper",
    "publication_date": "2025-12-19",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/jacs.5c14652",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Authors report Co3Mo1 catalyst performance under a defined gas-flow, temperature, support, loading and pretreatment protocol; Co75Mo25 is the corresponding generated candidate.",
      "unknown": "Exact composition/formulation prior art, long-term stability, energy efficiency and scale-up remain unresolved; this draft makes no global-priority claim.",
      "ai_role": "A random-forest adsorption-energy predictor guides diffusion denoising; DFT and laboratory testing evaluate generated candidates.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12814173/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PMC full text opened; article metadata and candidate/experimental mapping reconciled using continuation audit dated 2026-11-16."
    }
  },
  {
    "id": "2026-06-fungai-antifungal-peptides",
    "title": "Fung-AI prioritizes peptides with measured antifungal activity",
    "summary": "Fung-AI combines a generative adversarial network with learned antifungal and hemolysis classifiers to select peptides for testing. The study first tested five peptides from separate clusters; activity of peptide 40 prompted testing of eight more members from cluster 17. Five of the nine tested cluster-17 peptides inhibited Fusarium graminearum, a post-selection cluster-specific outcome—not a general 5/13 model hit rate. Four peptides showed activity against Candida albicans, while none inhibited C. auris up to 500 μg/mL. The paper’s HepG2 cytotoxicity statements are internally inconsistent for peptide 48 and peptide 65, and peptide 12 and 40 have reported cytotoxicity; these assays establish neither human safety nor clinical efficacy.",
    "field": "biology",
    "ai_technology": "Fung-AI GAN; learned antifungal and hemolysis classifiers",
    "organization": "Johns Hopkins University Applied Physics Laboratory; Johns Hopkins University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13289945/",
    "source_type": "paper",
    "publication_date": "2026-06-15",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1371/journal.pcbi.1014105",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports staged testing, fungal MICs and cell-line cytotoxicity; adaptive denominator and conflicting LC50 statements were reconciled in the continuation audit.",
      "unknown": "Sequence-level prior art and independent replication not assessed; HepG2 assays do not establish human safety.",
      "ai_role": "GAN generates candidates; antifungal and hemolysis classifiers narrow the candidate set before synthesis and fungal assays.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13289945/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PLOS article and indexed PMC full text checked; follow-up audit read full article and resolves staged denominator and LC50 inconsistency, not by remeasurement."
    }
  },
  {
    "id": "2025-07-random-forest-candida-screening",
    "title": "Machine-learning screening selects compounds active against Candida",
    "summary": "Random-forest prediction and chemical filters helped narrow commercial compounds for experimental testing. Of 17 selected compounds, the paper reports 11 with inhibitory activity against Candida albicans at tested concentrations. Compound 17 had an IC50 of 0.071 μM and MIC of 4.5 μM, compared with fluconazole values of 0.39 μM and 0.78 μM, respectively; its lower IC50 but higher MIC does not establish superiority. The reported selectivity-index definition appears inconsistent with tabulated values, so this draft omits that index. Cell assays do not establish clinical efficacy or safety.",
    "field": "biology",
    "ai_technology": "Random-forest classifiers and regressors; chemical similarity filtering",
    "organization": "Federal University of Goiás",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12320878/",
    "source_type": "paper",
    "publication_date": "2025-07-02",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1080/17460913.2025.2525717",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary full text reports 17 experimentally tested compounds and 11 active; compound-specific IC50/MIC values are stated with different fluconazole comparators.",
      "unknown": "Molecule-level previous activity, exact compound identities and independent replication remain to review.",
      "ai_role": "Random-forest models predict activity and toxicity to prioritize commercially available compounds; selected compounds are experimentally tested.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12320878/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary full-text article on PMC indexed text and methods/results read; online date, title and DOI verified."
    }
  },
  {
    "id": "2024-10-minerva-nickel-suzuki-optimization",
    "title": "Minerva identifies nickel-catalyzed Suzuki reaction conditions",
    "summary": "Minerva combines Gaussian-process optimization with automated high-throughput experiments to search reaction conditions. In the nickel-catalyzed Suzuki campaign, the preprint reports a condition with 76 area-percent yield and 92% selectivity, while two chemist-designed screening plates did not find successful conditions. Area percent is a chromatographic assay measure, not isolated yield. The preprint and expanded journal report are one study; this does not establish invention of Suzuki coupling or clinical utility.",
    "field": "chemistry",
    "ai_technology": "Minerva Gaussian-process surrogate; Bayesian acquisition; automated HTE",
    "organization": "F. Hoffmann-La Roche AG; EPFL",
    "source_url": "https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/66f976b5cec5d6c142981e4e/original/highly-parallel-optimisation-of-nickel-catalysed-suzuki-reactions-through-automation-and-machine-intelligence.pdf",
    "source_type": "paper",
    "publication_date": "2024-10-01",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-61803-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Experimental 96-well optimization reports 76 AP yield and 92% selectivity; two expert-designed screening plates did not find successful conditions.",
      "unknown": "Product assay area-percent is not isolated yield; pharmaceutical structures and independent replication remain unreviewed.",
      "ai_role": "A GP surrogate and Bayesian acquisition choose batches for automated experiments and use measured outcomes to update condition search.",
      "evidence_urls": [
        "https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/66f976b5cec5d6c142981e4e/original/highly-parallel-optimisation-of-nickel-catalysed-suzuki-reactions-through-automation-and-machine-intelligence.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary ChemRxiv v1 full-text PDF and primary PubMed/journal metadata and abstract checked; journal full PMC direct open CAPTCHA challenged. Exact first-version content for later API case studies is excluded from draft."
    }
  },
  {
    "id": "2026-09-deep-learning-fluid-ferroelectric-molecules",
    "title": "Deep learning-guided synthesis yields fluid ferroelectric molecules",
    "summary": "A graph variational autoencoder proposes molecules while learned classifiers and regressors, retrosynthesis and stock-inventory constraints guide selection. The journal version reports 11 first-generation materials characterized through mixture-based extrapolation and two additional materials selected in a second iteration that exhibited the ferroelectric-nematic phase in their neat state. The first arXiv version (2025-12-18) contained the first-generation mixture-based work, not the later neat-state confirmation; this draft uses the journal date for that result. Predictions have material errors, and the work does not establish device performance.",
    "field": "materials",
    "ai_technology": "Graph variational autoencoder; graph neural-network ensembles; XGBoost",
    "organization": "University of Leeds",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13569083/",
    "source_type": "paper",
    "publication_date": "2026-09-11",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/d6mh01249c",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Journal full text reports two second-round synthesized materials with neat-state ferroelectric-nematic phase; first 11 compounds were assessed through mixture-based extrapolation.",
      "unknown": "Transition predictions may differ substantially from observed values; independent replication and device performance not established.",
      "ai_role": "Generative and predictive graph models rank and propose molecules; inventory and retrosynthetic filters guide laboratory synthesis and characterization.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13569083/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary arXiv v1 PDF directly read and primary journal full text indexed on PMC; version-specific first-cohort versus second-round evidence reconciled."
    }
  },
  {
    "id": "2026-04-pepic-bhet-hydrolyzing-enzyme",
    "title": "PEPIC identifies a BHET-hydrolyzing enzyme candidate",
    "summary": "PEPIC is a sequence-based convolutional model that predicts PET-hydrolyzing potential. The study reports prioritizing W2061_PET9 and experimentally showing that purified protein converts soluble BHET into MHET, confirmed by HPLC and mass spectrometry. The experiment used a soluble PET intermediate; it does not establish degradation of polymer PET. Homology screening and human selection also contributed.",
    "field": "biology",
    "ai_technology": "PEPIC sequence convolutional neural network",
    "organization": "SunMoon University",
    "source_url": "https://journals.asm.org/doi/pdf/10.1128/aem.01586-25",
    "source_type": "paper",
    "publication_date": "2026-04-30",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1128/aem.01586-25",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Purified W2061_PET9 converted BHET to MHET in an in-vitro assay confirmed chromatographically and by mass spectrometry.",
      "unknown": "Polymer PET degradation, exact sequence-level prior characterization and independent replication are not established.",
      "ai_role": "PEPIC scores sequences for predicted PET-hydrolytic activity to prioritize a candidate after genomic/homology selection; direct assay validates soluble BHET turnover.",
      "evidence_urls": [
        "https://journals.asm.org/doi/pdf/10.1128/aem.01586-25"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher PDF indexed full text confirms title/DOI, publication date, AI role and Fig. 6 experimental validation."
    }
  },
  {
    "id": "2024-04-reinforcement-learning-sars-cov-2-mpro-inhibitors",
    "title": "Reinforcement learning and hit expansion identify SARS-CoV-2 Mpro inhibitor series",
    "summary": "A mixed workflow using deep reinforcement learning, pharmacophore and shape filters, docking and hit expansion reports three SARS-CoV-2 main-protease inhibitor series with biochemical IC50 values of 1.3–2.3 μM. The paper combines generated and commercially sourced hits; not every final inhibitor is claimed as directly model-generated. PDB entry 8R11 was released before the article and already disclosed a bound compound structure, so no priority claim is made for that structure. These biochemical findings do not establish antiviral efficacy or clinical safety.",
    "field": "medicine",
    "ai_technology": "Deep reinforcement learning; pharmacophore/shape filtering; docking",
    "organization": "Idorsia Pharmaceuticals; Johannes Gutenberg University Mainz",
    "source_url": "https://pubs.rsc.org/en/content/articlepdf/2024/md/d4md00106k",
    "source_type": "paper",
    "publication_date": "2024-04-29",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D4MD00106K",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher abstract reports three experimentally validated Mpro inhibitor series with IC50 1.3–2.3 μM; PDB 8R11 pre-dates the paper.",
      "unknown": "Exact assignment of each final inhibitor to generated versus commercially sourced material and prior chemotype overlap not fully audited; no antiviral or clinical claims.",
      "ai_role": "Reinforcement learning generates candidate structures; computational filters and hit expansion guide follow-up, but commercially sourced hits also contributed.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlepdf/2024/md/d4md00106k"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary RSC article landing page and full-text PDF abstract; first-publication date verified. PDB identity/release was reported in input ledger and should be retained as earlier structure disclosure."
    }
  },
  {
    "id": "2025-02-ml-guided-ce-phosphor",
    "title": "Machine learning identifies a Ce-doped green-emitting phosphor",
    "summary": "XGBoost screening of measured Ce substitution sites guided synthesis of Ca2SrSc6O12:Ce3+, reported to show blue excitation and green emission. This is a laboratory phosphor result, not a commercial LED demonstration. The 2026 journal version describes revised data; its exact mapping to this preprint result is unresolved, so this record is bounded to the arXiv v1 claim.",
    "field": "materials",
    "ai_technology": "XGBoost regression; crystal-chemical descriptors",
    "organization": "University of Houston",
    "source_url": "https://arxiv.org/abs/2502.18859",
    "source_type": "paper",
    "publication_date": "2025-02-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Ca2SrSc6O12:Ce3+ was synthesized and its excitation/emission measured.",
      "unknown": "Journal-version result mapping and earlier conference disclosure are unresolved.",
      "ai_role": "Model predicted Ce excitation energy and guided human selection for synthesis.",
      "evidence_urls": [
        "https://arxiv.org/abs/2502.18859"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2026-07-active-learning-protonic-cell-electrode",
    "title": "Active learning selects an oxygen electrode for protonic ceramic cells",
    "summary": "A supervised DFT-trained model and Gaussian-process experimental loop guided selection of BaCo0.8Nb0.1Zr0.1O3−δ. The publisher abstract reports 2.68 W cm−2 peak fuel-cell power at 600 °C and a separate 500-hour electrolysis test. These are specified laboratory cell tests; the Research Square v1 and journal versions differ in candidate counts and are not conflated here.",
    "field": "materials",
    "ai_technology": "DFT-trained supervised model; Gaussian-process Bayesian optimization",
    "organization": "Georgia Institute of Technology",
    "source_url": "https://doi.org/10.1002/adma.74212",
    "source_type": "paper",
    "publication_date": "2026-07-21",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher primary abstract reports BCNZ20 and specified fuel-cell and electrolysis measurements.",
      "unknown": "Version-specific candidate counts and detailed test protocols require full-text comparison; no commercial performance claim.",
      "ai_role": "Supervised composition/property prediction narrows candidates; Bayesian experimental optimization iteratively guides measurement.",
      "evidence_urls": [
        "https://doi.org/10.1002/adma.74212"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2025-06-ml-cmn-diatomic-orr-catalyst",
    "title": "Machine-learning-guided Co–Mn diatomic oxygen-reduction catalyst",
    "summary": "A machine-learning-derived theoretical activity map guided a nitrogen-bridged Co–Mn catalyst proposal. The study reports synthesized Co-N-Mn/NC with a 0.90 V half-wave potential and 271 mW cm−2 peak power density in its alkaline oxygen-reduction and zinc–air tests. These are study-specific laboratory results, not commercial battery performance.",
    "field": "chemistry",
    "ai_technology": "Machine-learning activity map; theoretical descriptor modeling",
    "organization": "Anhui Normal University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12134119/",
    "source_type": "paper",
    "publication_date": "2025-06-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Reported Co-N-Mn/NC catalyst was tested for ORR and in a zinc–air battery; paper reports 0.90 V and 271 mW cm−2.",
      "unknown": "Atom-specific structure attribution, training overlap, prior composition art and external replication remain open.",
      "ai_role": "Calculated activity map informed candidate geometry and pair selection before synthesis.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12134119/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2024-04-graph-ml-gas-separation-membranes",
    "title": "Graph machine learning identifies polymer gas-separation membranes",
    "summary": "A graph-based machine-learning model ranked polymer candidates; two were synthesized and measured above selected empirical permeability–selectivity upper bounds for O2/N2, H2/CH4 and H2/N2. The bounds are historical empirical comparisons, not physical limits; the study does not demonstrate industrial mixed-gas performance. The claim here is bounded to the primary arXiv v1, which reports the synthesized candidates.",
    "field": "materials",
    "ai_technology": "Explainable graph machine learning",
    "organization": "University of Notre Dame",
    "source_url": "https://arxiv.org/abs/2404.10903",
    "source_type": "paper",
    "publication_date": "2024-04-16",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Two synthesized candidates reported above selected empirical upper bounds in specified gas pairs.",
      "unknown": "Exact journal/preprint result correspondence and prior structure-level art not exhaustively checked.",
      "ai_role": "Graph ML predicts permeability/selectivity from polymer structure and prioritizes candidates for synthesis and testing.",
      "evidence_urls": [
        "https://arxiv.org/abs/2404.10903"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2026-03-ml-screened-w-doped-nifeooh",
    "title": "Machine learning selects W-doped nickel–iron oxyhydroxide for electrolysis",
    "summary": "EquiformerV2-based interatomic predictions screened 3,976 configurations and guided W1–NiFeOOH selection. The journal reports 13.1 A cm−2 at 2.0 V and 80 °C in one anion-exchange-membrane electrolyzer test, and a separate >500-hour test at 1 A cm−2 and 60 °C. Conditions are distinct laboratory tests and are not evidence of commercial scale.",
    "field": "materials",
    "ai_technology": "EquiformerV2 interatomic machine-learning potential",
    "organization": "Seoul National University",
    "source_url": "https://www.nature.com/articles/s41467-026-68735-3",
    "source_type": "paper",
    "publication_date": "2026-03-11",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher primary indexed article reports tested W1–NiFeOOH, high-current-density and separate durability results.",
      "unknown": "A dataset deposited 2026-01-10 predates the paper; its preview was inaccessible, so earliest disclosure of the specific result is unresolved. Training overlap and independent replication remain open.",
      "ai_role": "Interatomic model predicts stability/adsorption properties to screen catalyst configurations before synthesis and electrolysis tests.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-68735-3"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2024-11-nn-guided-ruti-mn-oer-electrode",
    "title": "Neural network guides a Ru–Ti–Mn oxygen-evolution electrode",
    "summary": "A backpropagation neural network fitted composition/activity and deactivation data and suggested a Ru:Ti:Mn ratio of 0.26:0.26:0.48. The synthesized electrode had measured composition near 0.24:0.28:0.48 and reported 165.2 mV overpotential at 10 mA cm−2 in the acidic assay. Model optimum and measured composition are distinct; reported long-duration tests apply only under the study’s specific conditions.",
    "field": "chemistry",
    "ai_technology": "Backpropagation neural network",
    "organization": "Tsinghua University",
    "source_url": "https://doi.org/10.21203/rs.3.rs-5208867/v1",
    "source_type": "paper",
    "publication_date": "2024-11-27",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Research Square v1 reports measured Ru/Ti/Mn electrode and acidic OER result.",
      "unknown": "Exact prior composition overlap and whether candidate was included in training data remain open.",
      "ai_role": "Neural network fit experimental composition/activity/durability data to suggest a candidate composition, which was synthesized and tested.",
      "evidence_urls": [
        "https://doi.org/10.21203/rs.3.rs-5208867/v1"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2024-09-descriptor-guided-dual-atom-catalysts",
    "title": "Descriptor-guided Co–Co and Co–Ir oxygen electrocatalysts",
    "summary": "Physically informed feature selection derived an ARSC descriptor from calculated catalyst properties to guide oxygen-reaction candidate design. The study reports synthesis and alkaline testing of two carbon-supported samples: ORR half-wave potentials of 0.941 and 0.937 V and OER overpotentials of 330 and 340 mV at 10 mA cm−2. These are laboratory measurements; structural analyses support but do not prove every proposed active-site configuration.",
    "field": "chemistry",
    "ai_technology": "ARSC descriptor; feature selection",
    "organization": "Tianjin University",
    "source_url": "https://www.nature.com/articles/s41467-024-52519-8",
    "source_type": "paper",
    "publication_date": "2024-09-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher PDF reports synthesized Co–Co and Co–Ir samples with stated ORR and OER tests.",
      "unknown": "Prior composition overlap and independent replication remain open; specific active-site configuration is not directly resolved for every site.",
      "ai_role": "Calculated dual-atom catalyst descriptors selected candidates followed by synthesis and electrochemical testing.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-52519-8"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2025-03-ml-guided-vitrimer-design",
    "title": "Machine-learning models guide design of two vitrimers",
    "summary": "An ensemble trained on molecular-dynamics data predicted glass-transition temperatures and screened candidates. The primary preprint reports two synthesized vitrimers with higher measured glass-transition temperatures than the compared bifunctional transesterification vitrimers and demonstrates healability. This is a bounded laboratory comparison, not proof of broad industrial recyclability or lifetime.",
    "field": "materials",
    "ai_technology": "Ensemble machine-learning models trained on molecular-dynamics data",
    "organization": "University of Washington",
    "source_url": "https://arxiv.org/abs/2503.20956",
    "source_type": "paper",
    "publication_date": "2025-03-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Two nominated vitrimers were synthesized and measured; the paper reports higher Tg than the selected comparator set.",
      "unknown": "Exact composition novelty, healing conditions and independent replication not established.",
      "ai_role": "Molecular-dynamics-informed ML predicted Tg and ranked candidate chemistries for synthesis and measurement.",
      "evidence_urls": [
        "https://arxiv.org/abs/2503.20956"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2025-08-ml-screened-redox-perovskites",
    "title": "Machine learning guides screening of redox perovskites for solar hydrogen",
    "summary": "Random-forest models predicted oxygen-vacancy enthalpy and screened stability across candidate redox perovskites; four selected materials were synthesized and tested thermochemically. The paper’s solar-hydrogen system outcomes are simulations, not a demonstrated plant. The full article was available only as publisher-indexed primary text in this review.",
    "field": "materials",
    "ai_technology": "Random-forest vacancy-enthalpy model and stability classifier",
    "organization": "University of Newcastle",
    "source_url": "https://www.nature.com/articles/s41524-025-01726-y",
    "source_type": "paper",
    "publication_date": "2025-08-01",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher indexed primary text confirms model screens and synthesis/thermochemical experiments on four compositions.",
      "unknown": "Direct full article and supplement were blocked; full experimental yields, prior composition art, and independent replication not audited.",
      "ai_role": "ML screening of vacancy enthalpy and stability narrowed computational candidates for synthesis and thermochemical testing.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01726-y"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2025-09-ml-selected-ntimp2-mmp-specificity",
    "title": "Machine learning selects an N-TIMP2 variant with altered MMP specificity",
    "summary": "Models trained on binding data for MMP-1, MMP-3 and MMP-9 guided selection of an N-TIMP2 variant. The paper reports high affinity for MMP-9, moderate affinity for MMP-3 and low affinity for MMP-1 relative to wild type. This is an in-vitro protein specificity result, not clinical efficacy. Earlier N-TIMP2 specificity engineering exists; this draft claims only the study’s selected variant and assay result.",
    "field": "biology",
    "ai_technology": "Machine-learning binding predictors for MMP-1, MMP-3 and MMP-9",
    "organization": "Ben-Gurion University of the Negev",
    "source_url": "https://doi.org/10.1111/febs.70249",
    "source_type": "paper",
    "publication_date": "2025-09-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports an experimentally tested selected N-TIMP2 variant with differential MMP binding profile.",
      "unknown": "Conference abstract posting/result chronology and exact sequence-level overlap remain unresolved; no clinical claim.",
      "ai_role": "Binding predictors scored mutation candidates and guided selection of variant for experimental binding assays.",
      "evidence_urls": [
        "https://febs.onlinelibrary.wiley.com/doi/10.1111/febs.70249",
        "https://pubmed.ncbi.nlm.nih.gov/40912922/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2024-03-lscnn-antibacterial-ugi-products",
    "title": "Feature-fusion model prioritizes antibacterial Ugi products",
    "summary": "LSCNN ranked a combinatorial Ugi library after model development on focused activity data. The paper reports six active products among ten selected combinations against MRSA; purified H4–H6 had MIC and MBC of 12 μM. Different screening phases and cutoffs mean these figures are not a randomized general hit-rate comparison.",
    "field": "chemistry",
    "ai_technology": "LSCNN feature-fusion neural network",
    "organization": "Zhejiang University",
    "source_url": "https://doi.org/10.1039/D3SC06441G",
    "source_type": "paper",
    "publication_date": "2024-03-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports 6/10 selected combinations active against MRSA and purified H4–H6 MIC/MBC results.",
      "unknown": "Prior chemical-structure overlap, full independent replication, and clinical efficacy are not established.",
      "ai_role": "Model ranks products from combinatorial Ugi space for synthesis and testing.",
      "evidence_urls": [
        "https://pubs.rsc.org/en/content/articlepdf/2024/sc/d3sc06441g",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11041243/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2024-03-ml-ga-lbd-antimicrobial-peptides",
    "title": "Machine learning and genetic search design antimicrobial peptides",
    "summary": "Regression and k-nearest-neighbor models with a genetic algorithm guided optimization from a shrimp lipopolysaccharide-binding peptide to LBD_A–D. The article reports in-vitro tests, NMR structure for LBD_B and improved survival in a kuruma prawn co-injection experiment. This is not human therapy and does not establish treatment efficacy after infection.",
    "field": "biology",
    "ai_technology": "Regression and k-nearest-neighbor models; genetic algorithm",
    "organization": "Beijing Normal University",
    "source_url": "https://doi.org/10.1016/j.jare.2024.02.016",
    "source_type": "paper",
    "publication_date": "2024-03-01",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Article reports synthesized/tested peptide variants, NMR characterization and a prawn co-injection assay.",
      "unknown": "Full exact assay/organism scope and peptide sequence novelty/independent replication not exhaustively compared.",
      "ai_role": "Activity models and genetic search guided sequence variants which were synthesized and biologically tested.",
      "evidence_urls": [
        "https://doi.org/10.1016/j.jare.2024.02.016",
        "https://pubmed.ncbi.nlm.nih.gov/38431124/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2024-09-co-training-nir-cr-phosphor",
    "title": "Co-training identifies a high-efficiency Cr-activated NIR phosphor",
    "summary": "Semi-supervised co-training with kernel ridge regression and support-vector regression expanded the labeled phosphor dataset and selected Lu3Y2Ga3O12:Cr3+. The study reports emission near 750 nm with internal and external quantum efficiencies of 97.1% and 38.8%, respectively, and a fabricated laboratory phosphor-converted LED. The abstract’s values are author-reported laboratory measurements, not commercial product performance.",
    "field": "materials",
    "ai_technology": "Semi-supervised KRR/SVR co-training",
    "organization": "Yanshan University",
    "source_url": "https://www.nature.com/articles/s41524-024-01395-3",
    "source_type": "paper",
    "publication_date": "2024-09-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary publisher article reports synthesized candidate, emission/efficiencies and laboratory pc-LED.",
      "unknown": "Prior composition-level disclosures and independent device durability are not established.",
      "ai_role": "Co-training predicts emission peak and screens candidate garnet compositions for experimental synthesis.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-024-01395-3"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2026-01-text-mining-eu-phosphor-wavelengths",
    "title": "Text mining and machine learning guide Eu-activated phosphor wavelength prediction",
    "summary": "A literature text-mining pipeline extracted phosphor/emission examples for model development; four Eu2+-activated candidates were synthesized and characterized. In the author manuscript, CaGd2S4:Eu2+ emitted at 440 nm versus 451 nm predicted under 365 nm excitation. Formula Li2SiN2 in one abstract conflicts with Li2SiN3 in results; this record does not claim that candidate formula.",
    "field": "materials",
    "ai_technology": "Literature text-mining pipeline; genetic-algorithm feature selection; ML wavelength regression",
    "organization": "Xiamen University",
    "source_url": "https://www.nature.com/articles/s41524-026-01967-5",
    "source_type": "paper",
    "publication_date": "2026-01-23",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Author manuscript reports synthesized and measured CaGd2S4:Eu2+ and three other candidates; specific comparison retained.",
      "unknown": "One formula conflict is unresolved; author manuscript/date/version details and composition-level prior art not fully checked.",
      "ai_role": "Text mining builds experimental training set; ML predicts emission wavelength and guides candidate selection.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-026-01967-5"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2022-09-isa-ml-oxygen-reduction-electrodes",
    "title": "Machine learning screens perovskite oxygen-reduction electrodes",
    "summary": "The study introduced ionic Lewis-acid strength as a physical descriptor and used machine learning to screen 6,871 perovskite compositions. Four oxides were synthesized and reported to have superior oxygen-reduction activity metrics in specified ceramic-fuel-cell tests. The record counts one paper-level study and does not imply 6,871 physical syntheses or commercial deployment.",
    "field": "materials",
    "ai_technology": "Machine-learning composition screening; ionic Lewis-acid-strength descriptor",
    "organization": "Shenzhen University",
    "source_url": "https://www.nature.com/articles/s41560-022-01098-3",
    "source_type": "paper",
    "publication_date": "2022-09-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary publisher abstract reports four synthesized oxides screened from 6,871 and experimental activity results.",
      "unknown": "Exact formulation-level overlap and independent replication not exhaustively assessed.",
      "ai_role": "ML plus physical descriptor screened composition candidates followed by synthesis and electrochemical tests.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41560-022-01098-3"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2024-09-ml-discovery-alkaline-laccases",
    "title": "Machine learning identifies alkaline laccases from Lepista nuda",
    "summary": "Machine-learning pH-optimum predictors screened metagenomic sequences, and two predicted Lepista nuda laccases were produced and profiled. The authors report LnLccA and LnLccB optimum activity at pH 10 and 9 for substrate 2,6-DMP, respectively, while both had pH 4 optima for ABTS. The substrate-dependent contrast is retained; it is not a general alkaline optimum or industrial demonstration.",
    "field": "biology",
    "ai_technology": "Supervised, semi-supervised and unsupervised pH-optimum prediction models",
    "organization": "University of Helsinki; KTH Royal Institute of Technology",
    "source_url": "https://link.springer.com/article/10.1186/s13068-024-02566-6",
    "source_type": "paper",
    "publication_date": "2024-09-11",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Open primary full article reports recombinant LnLccA and LnLccB pH profiles, substrate-specific optima and validation of ML predictions.",
      "unknown": "Prior characterization and novelty of sequence/activity remain open; industrial applicability not demonstrated.",
      "ai_role": "Models trained on curated laccase pH data predicted candidates from metagenomic sequences that were expressed and profiled.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s13068-024-02566-6"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2024-07-esmif1-antibody-variant-selection",
    "title": "Structure-informed model selects antibody variants active against SARS-CoV-2 escape variants",
    "summary": "ESM-IF1 scored substitutions in antibody–antigen complexes, and researchers experimentally screened variants of two clinical antibodies. The journal article reports up to 25-fold neutralization improvement against BQ.1.1 and up to 37-fold affinity improvement against XBB.1.5; these are distinct assays/variants. The preprint abstract reports a different neutralization figure, so this record uses journal values and does not backdate them to preprint v1.",
    "field": "medicine",
    "ai_technology": "ESM-IF1 inverse-folding protein model",
    "organization": "Stanford University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11616794/",
    "source_type": "paper",
    "publication_date": "2024-07-25",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "PMC-hosted primary article reports experimental screening and improved neutralization/affinity results for antibody variants.",
      "unknown": "Exact v1 result content and sequence-level overlap with earlier group studies not completely established; no clinical efficacy.",
      "ai_role": "Structure-informed inverse-folding scores guided substitutions followed by antibody binding and neutralization tests.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11616794/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2025-11-lightgbm-argyrodite-conductors",
    "title": "LightGBM guides discovery of argyrodite lithium-ion conductors",
    "summary": "A composition-based LightGBM model trained on about 4,000 compounds guided synthesis of co-substituted argyrodite families. The primary paper reports Li6.7Ge0.595Si0.105P0.3S5I conductivity of 7.2×10−3 S cm−1 and a separate hot-pressed conductivity above 10−2 S cm−1. These are material measurements, not a demonstrated full battery; the model did not originate the argyrodite family.",
    "field": "materials",
    "ai_technology": "Elements-To-Ionics LightGBM",
    "organization": "Institute of Science Tokyo",
    "source_url": "https://doi.org/10.1002/smll.202509918",
    "source_type": "paper",
    "publication_date": "2025-11-18",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher full-text indexed article identifies LightGBM screening and specific synthesized conductor conductivity results.",
      "unknown": "Prior composition reports and detailed processing/test comparisons remain open.",
      "ai_role": "Composition-based model estimates ionic conductivity and guides synthesis of co-substituted candidates.",
      "evidence_urls": [
        "https://doi.org/10.1002/smll.202509918"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2025-02-ml-fluoroether-electrolytes",
    "title": "Machine learning screens fluoroether electrolytes for lithium-metal batteries",
    "summary": "The study applies machine learning to a virtual library of 5,576 electrolyte combinations, then synthesizes and tests two candidate fluoroethers. The paper reports nearly 99.5% Coulombic efficiency in specified Li||NMC811 testing. This is a bounded cell result, not evidence that all virtual candidates were made or that either is commercially deployable.",
    "field": "chemistry",
    "ai_technology": "Machine-learning electrolyte property and synthesizability screening",
    "organization": "Nanjing University",
    "source_url": "https://doi.org/10.1002/anie.202424493",
    "source_type": "paper",
    "publication_date": "2025-02-25",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary paper and PubMed abstract support virtual screening, two synthesized candidates and cell testing.",
      "unknown": "Candidate-specific assignment and exact cycling conditions behind near-99.5% metric need supplementary mapping; no commercial cell claim.",
      "ai_role": "ML screens electrolyte property/synthesizability combinations for synthesis and Li-metal cell assessment.",
      "evidence_urls": [
        "https://doi.org/10.1002/anie.202424493"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2025-02-euclid-cnn-strong-lens-confirmation",
    "title": "CNN selection leads to spectroscopic confirmation of a Euclid strong lens",
    "summary": "Five convolutional models ranked Euclid early-release images, followed by human visual grading. Spectroscopy and lens modeling support EUCL J081705.61+702348.8 as a strong lens, with foreground redshift 0.335 and lensed source redshift 1.475. The 97 visually graded candidates are not 97 confirmed lenses, and model ranking alone did not establish the lens.",
    "field": "physics",
    "ai_technology": "Ensemble of five convolutional neural networks",
    "organization": "University of Groningen",
    "source_url": "https://arxiv.org/abs/2502.09802v1",
    "source_type": "paper",
    "publication_date": "2025-02-13",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Primary v1 and journal paper report spectroscopically supported lens with redshifts and lens modeling.",
      "unknown": "Earlier object reporting and aliases remain open; human visual grading and spectroscopy contributed to confirmation.",
      "ai_role": "CNN ensemble ranked candidates for human grading and spectroscopic follow-up.",
      "evidence_urls": [
        "https://arxiv.org/abs/2502.09802v1"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2024-04-deepevo-g3pdh-thermotolerant-variants",
    "title": "DeepEvo yields heat-tolerant G3PDH variants",
    "summary": "A generative model and protein-language-model-based temperature selector were combined in an iterative design procedure. The article reports eight heat-tolerant glyceraldehyde-3-phosphate dehydrogenase variants among 30 generated and tested sequences. This is an experimentally screened enzyme set, not proof of industrial deployment.",
    "field": "biology",
    "ai_technology": "DeepEvo generative model; protein language model temperature selector",
    "organization": "Tianjin Institute of Industrial Biotechnology, Chinese Academy of Sciences",
    "source_url": "https://spj.science.org/doi/10.34133/bdr.0031",
    "source_type": "paper",
    "publication_date": "2024-04-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher article reports eight heat-tolerant variants among 30 generated sequences tested.",
      "unknown": "Exact assay definition, earlier disclosure and sequence-level prior overlap remain open.",
      "ai_role": "Generative model proposes functional sequences and PLM-based selector ranks high-temperature candidates in iterative design/test loop.",
      "evidence_urls": [
        "https://spj.science.org/doi/10.34133/bdr.0031"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2024-12-plm-guided-creatinase-stability",
    "title": "Protein language model guides creatinase stability optimization",
    "summary": "A deep protein language model used prior single- and multiple-mutant measurements to guide combinations across two design/test rounds. The 13-substitution variant 13M4 reportedly retained nearly wild-type catalytic activity and increased apparent melting temperature by 10.19 °C. This assay does not establish long-term stability across production conditions, and the model recombined previously measured mutations.",
    "field": "biology",
    "ai_technology": "Deep protein language model",
    "organization": "Shanghai Jiao Tong University",
    "source_url": "https://doi.org/10.1002/mlf2.12151",
    "source_type": "paper",
    "publication_date": "2024-12-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher article and PubMed metadata report 13M4, near-WT catalytic activity and apparent melting-temperature increase.",
      "unknown": "Prior exact-sequence reporting and independent replication not established.",
      "ai_role": "Protein language model guided combinations of previously tested mutations for experimental stability and activity assays.",
      "evidence_urls": [
        "https://doi.org/10.1002/mlf2.12151"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, date, model role and bounded reported result."
    }
  },
  {
    "id": "2026-02-ml-guided-low-pgm-rwgs-catalyst",
    "title": "Machine learning guides a low-precious-metal RWGS catalyst",
    "summary": "An iterative machine-learning-guided campaign optimized precious-metal and additive compositions and support over 35 prediction–experiment cycles. The authors report a low-loading Pt–Ru–Pd catalyst that outperformed a commercial benchmark and their earlier high-PGM design in the reported reverse water–gas shift assay. This is a laboratory catalyst result, not demonstrated industrial performance.",
    "field": "chemistry",
    "ai_technology": "Machine-learning-guided iterative catalyst composition and support optimization",
    "organization": "Hokkaido University",
    "source_url": "https://pubs.acs.org/doi/10.1021/acscatal.5c09278",
    "source_type": "paper",
    "publication_date": "2026-02-25",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The publisher article reports the 35-cycle campaign and its best low-PGM catalyst outperforming the stated commercial and earlier high-PGM comparators.",
      "unknown": "The result is specific to the reported assay; scale-up and industrial operation are not established.",
      "ai_role": "Machine-learning-guided predictions prioritize PGM/additive compositions and support for iterative experimental testing.",
      "evidence_urls": [
        "https://pubs.acs.org/doi/10.1021/acscatal.5c09278"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher indexed primary article text including author affiliations, article history and abstract/result; full-text access remains publisher-limited."
    }
  },
  {
    "id": "2022-07-semi-supervised-lithium-thioborate",
    "title": "Semi-supervised learning selects a lithium thioborate for vacancy engineering",
    "summary": "Semi-supervised learning screened lithium-containing structures and selected Li3BS3 for experimental characterization. The parent material was a poor room-temperature conductor. Researchers then introduced vacancies by Si substitution and amorphized the substituted material; the paper reports approximately four orders of magnitude higher ionic conductivity for a-Li2.95B0.95Si0.05S3. The model selected the framework, while human-led defect engineering produced the high-conductivity sample. This does not establish that pristine Li3BS3 is superionic.",
    "field": "materials",
    "ai_technology": "Semi-supervised learning; agglomerative clustering",
    "organization": "California Institute of Technology",
    "source_url": "https://authors.library.caltech.edu/records/evkz6-5jd40",
    "source_type": "paper",
    "publication_date": "2022-07-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D2EE03499A",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study selected Li3BS3 for characterization and reports increased ionic conductivity after Si substitution and amorphization, with the combined vacancy-engineering treatment improving conductivity by roughly four orders of magnitude.",
      "unknown": "The ML model selected a framework candidate; vacancy engineering and synthesis were subsequent researcher-led steps. The primary page was checked through indexed full-text results because direct article opening returned 403.",
      "ai_role": "Semi-supervised learning used literature-derived conductivity labels and structural descriptors to rank lithium-containing structures and select Li3BS3 for experimental characterization.",
      "evidence_urls": [
        "https://authors.library.caltech.edu/records/evkz6-5jd40",
        "https://pubs.rsc.org/en/content/articlehtml/2023/ee/d2ee03499a"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "CaltechAUTHORS primary manuscript record and deposited manuscript metadata; RSC indexed primary full-text results and methods."
    }
  },
  {
    "id": "2020-02-ml-ca-nb-ta-bi-oxide-conductors",
    "title": "Materials informatics and high-throughput synthesis identify durable oxide-ion conductors",
    "summary": "A machine-learning model trained on literature conductivity data screened inorganic crystal structures and narrowed the chemical space explored by high-throughput synthesis. The study reports high conductivity and thermal durability in selected Ca-(Nb,Ta)-Bi-O compositions; oxygen concentration-cell measurements gave oxide-ion transport numbers around 70–80% at 973 K. The result came from the combined informatics and experimental workflow. It does not establish a pure ionic conductor or room-temperature electrolyte.",
    "field": "materials",
    "ai_technology": "Machine-learning materials informatics; partial least-squares regression",
    "organization": "Toyota Central R&D Labs, Inc.",
    "source_url": "https://www.nature.com/articles/s43246-019-0004-7",
    "source_type": "paper",
    "publication_date": "2020-02-04",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43246-019-0004-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "High-throughput film libraries and bulk samples were synthesized and characterized; oxygen concentration cells measured oxide-ion transport numbers around 70–80% at 973 K.",
      "unknown": "The high-temperature transport result does not establish a pure ionic conductor or room-temperature electrolyte; supplementary details were not checked in this source review.",
      "ai_role": "A machine-learning model trained on literature oxide-ion conductivities screened crystal structures and helped narrow element combinations for high-throughput experimental synthesis and evaluation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43246-019-0004-7"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "publisher primary full text, methods/results and publication metadata"
    }
  },
  {
    "id": "2026-03-ml-cof-hydrogen-peroxide",
    "title": "Machine learning identifies covalent organic frameworks for hydrogen peroxide photosynthesis",
    "summary": "An ensemble machine-learning framework combined data augmentation, multilevel structural descriptors and physics-informed features to screen more than 10,000 covalent organic framework candidates. The researchers synthesized and tested eight selected COFs. The paper reports COF-343 at 12,978.7 μmol H2O2 h−1 g−1, with additional measured results for COF-2449, COF-2835 and COF-8484. This is one study-level materials result, not four discoveries. Detailed standard-test conditions in the supplement were not reviewed, and the results do not establish industrial performance. A data/code deposit may predate the journal article, but its date and result content were not checked here.",
    "field": "chemistry",
    "ai_technology": "Deep neural network ensemble; random-forest band-position model",
    "organization": "Sichuan University; Technische Universität Berlin",
    "source_url": "https://www.nature.com/articles/s44160-026-01037-0",
    "source_type": "paper",
    "publication_date": "2026-03-25",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary Nature Synthesis article page, abstract, methods/results text and author metadata as indexed by publisher."
    },
    "paper_url": "https://doi.org/10.1038/s44160-026-01037-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The indexed primary article reports synthesis and photocatalytic testing of eight selected COFs, including measured H2O2 production rates for COF-343, COF-2449, COF-2835 and COF-8484.",
      "unknown": "Detailed standard-test conditions in the supplementary information were not reviewed; the reported laboratory rates do not establish industrial performance. Zenodo deposit chronology/content was not checked.",
      "ai_role": "A DNN ensemble predicted photocatalytic yields and a random-forest model estimated band positions to guide COF screening.",
      "evidence_urls": [
        "https://www.nature.com/articles/s44160-026-01037-0"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-ml-cobalt-gamma-prime-alloys",
    "title": "Machine learning helps identify experimentally synthesized cobalt superalloys",
    "summary": "A random-forest model trained on first-principles formation and decomposition energies screened more than 150,000 proposed gamma-prime compositions. Domain screening and CALPHAD narrowed the set; researchers synthesized two selected Co-based gamma/gamma-prime alloy systems and characterized their microstructures. The measured compositions differ from nominal targets, and preliminary hardness values were below the cited conventional comparator. Engine performance was not demonstrated. Count the study once.",
    "field": "materials",
    "ai_technology": "Random-forest regression; density-functional-theory energy screening",
    "organization": "Xiamen University; Harbin Institute of Technology, Shenzhen",
    "source_url": "https://www.nature.com/articles/s41524-024-01455-8",
    "source_type": "paper",
    "publication_date": "2024-11-14",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary npj Computational Materials article page, abstract, indexed results and author metadata."
    },
    "paper_url": "https://doi.org/10.1038/s41524-024-01455-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers synthesized and characterized two screened Co-based alloy systems; XRD and electron microscopy support reported microstructures and preliminary hardness was measured.",
      "unknown": "Nominal and measured compositions differ; preliminary hardness was below the cited conventional comparator; no engine performance or independent replication was reported.",
      "ai_role": "A random-forest model trained on DFT energies screened candidate gamma-prime compositions; domain experts and CALPHAD then narrowed candidates for synthesis.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-024-01455-8"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-01-active-learning-lead-free-solder",
    "title": "Active learning guides experimental design of lead-free solder alloys",
    "summary": "Uncertainty-aware Gaussian-process models and multi-objective Bayesian acquisition guided seven design–experiment iterations. The study reports two selected alloys, Si-4 and Se-5, tested for tensile strength and fracture elongation. Reported averages follow the authors’ stated one-sigma selection of repeated tensile tests. Performance was mixed: the tested samples had poorer wettability and solder-joint shear strength than the commercial comparator. This is one laboratory alloy-design campaign, not evidence of industrial deployment or across-the-board superiority.",
    "field": "materials",
    "ai_technology": "Gaussian-process models; multi-objective Bayesian active learning",
    "organization": "Shanghai University; Hong Kong University of Science and Technology (Guangzhou)",
    "source_url": "https://www.nature.com/articles/s41524-024-01480-7",
    "source_type": "paper",
    "publication_date": "2025-01-10",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary npj Computational Materials indexed full text, publication history, experimental characterization and author affiliations."
    },
    "paper_url": "https://doi.org/10.1038/s41524-024-01480-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Seven design–experiment iterations included tensile testing of selected Si-4 and Se-5 alloys and evaluation of additional solder properties against a commercial comparator.",
      "unknown": "Averages follow a stated one-sigma selection of repeated tensile tests; wettability and joint shear strength were worse than the comparator. No industrial deployment or independent replication was established.",
      "ai_role": "Gaussian-process models represented experimental uncertainty and Bayesian multi-objective acquisition selected compositions for successive synthesis and testing rounds.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-024-01480-7"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-ml-low-sn-lead-free-relaxor-window",
    "title": "Machine learning identifies a low-tin lead-free relaxor composition window",
    "summary": "A multimodal literature-mining workflow and a physics-guided deep ensemble screened more than 150 million hypothetical compositions. The study reports that the x=0.01 and x=0.02 members of (Sr0.48Na0.26Bi0.26)(Ti1−xSnx)O3 met the authors’ predefined dielectric-stability targets while retaining room-temperature permittivity above 3300. The parent composition family and Sn substitution were already known; the result is a constrained low-Sn composition window, not a new parent chemistry. The reported upper-temperature stability criteria do not establish device certification or deployment.",
    "field": "materials",
    "ai_technology": "Multimodal literature mining; deep neural-network ensemble",
    "organization": "Seoul National University; Korea University; Yeungnam University",
    "source_url": "https://www.nature.com/articles/s41467-026-74376-3",
    "source_type": "paper",
    "publication_date": "2026-06-15",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PubMed record, article abstract, first-online date and affiliations."
    },
    "paper_url": "https://doi.org/10.1038/s41467-026-74376-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors report experimental validation of the x=0.01 and 0.02 Sn compositions against predefined dielectric-stability targets and room-temperature permittivity above 3300.",
      "unknown": "The parent family and substitution were known; the stated criteria do not establish full device certification or deployment. Direct publisher page was inaccessible, so evidence is limited to PubMed metadata and abstract in this check.",
      "ai_role": "Multimodal literature mining curated prior measurements and a deep neural-network ensemble screened candidate dielectric compositions and guided refinement.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42297784/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-11-multiomics-flowering-time-genes",
    "title": "Multi-omics models guide experimental validation of flowering-time genes",
    "summary": "Researchers trained models on genomic, transcriptomic and methylomic data for Arabidopsis traits and used feature interpretation to select flowering-time gene candidates. The preprint reports nine experimentally validated genes inferred to contribute to flowering time, tested using mutant phenotypes. Random-forest and ridge-regression models were used, so do not assign every gene to the random forest alone. Count one study, not nine discoveries. This draft follows the primary 2023 preprint version; the journal paper appeared in 2024.",
    "field": "biology",
    "ai_technology": "Random-forest models; ridge regression; multi-omics integration",
    "organization": "DOE Great Lakes Bioenergy Research Center, Michigan State University; Chinese Academy of Agricultural Sciences",
    "source_url": "https://biorxiv.org/content/10.1101/2023.11.14.566971v1",
    "source_type": "paper",
    "publication_date": "2023-11-14",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary bioRxiv v1 full text, posted-date line, abstract, methods and author affiliations."
    },
    "paper_url": "https://www.nature.com/articles/s41467-024-50701-6",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The preprint reports experimental validation of nine flowering-time gene candidates using single- and/or double-mutant phenotypes.",
      "unknown": "Individual candidate-selection routes are not all attributed to one model; the result is from a preprint version and does not establish independent replication.",
      "ai_role": "Random-forest and ridge-regression models used multi-omics data to predict plant traits and identify features associated with flowering-time predictions.",
      "evidence_urls": [
        "https://storage.prod.researchhub.com/uploads/papers/2023/11/15/2023.11.14.566971.full.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-08-deep-learning-ribozyme-neutral-network",
    "title": "Deep learning helps experimentally map a ribozyme neutral network",
    "summary": "After earlier experimental evolutionary rounds, the authors trained a deep classifier to guide the final in-silico generation and experimentally tested candidates. They measured all 2^16 intermediates between two active RNA ligase ribozymes separated by 16 mutations and report an extensive network of neutral paths. The result applies to this ribozyme and assay, not all RNA evolution. The paper also notes possible model overfitting and the restricted classifier’s accuracy.",
    "field": "biology",
    "ai_technology": "Deep-learning classifier; evolutionary algorithm",
    "organization": "Okinawa Institute of Science and Technology Graduate University",
    "source_url": "https://www.nature.com/articles/s41467-022-32538-z",
    "source_type": "paper",
    "publication_date": "2022-08-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary Nature Communications indexed full text, abstract, main text, publication metadata and author affiliation."
    },
    "paper_url": "https://doi.org/10.1038/s41467-022-32538-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reports experimental activity measurements for all 2^16 connecting sequence variants and a screen of more than 120,000 ribozyme sequences.",
      "unknown": "The neutral network is bounded to one RNA ligase ribozyme and assay; generality across other landscapes is not established, and the paper notes potential classifier overfitting.",
      "ai_role": "A deep-learning classifier trained on prior experimental generations guided in-silico selection, mutation and recombination in the final generation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-022-32538-z"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-10-ml-fatty-acyl-acp-reductase",
    "title": "Machine learning guides fatty acyl-ACP reductase engineering",
    "summary": "A machine-learning-guided iterative design–test–learn workflow engineered fatty acyl-ACP reductase variants for microbial fatty alcohol production. Across ten rounds, the study reports over twice the production of the starting natural sequences under tested conditions; the top sequence also had enhanced catalytic rate on palmitoyl-ACP. These are laboratory results, not industrial-scale production. The journal article date is used here; the earlier preprint’s exact posting and result-version differences were not checked.",
    "field": "biology",
    "ai_technology": "Machine-learning-guided protein engineering; Gaussian-process optimization",
    "organization": "University of Wisconsin-Madison",
    "source_url": "https://www.nature.com/articles/s41467-021-25831-w",
    "source_type": "paper",
    "publication_date": "2021-10-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary Nature Communications indexed full text, abstract, article date, author affiliations; preprint chronology not fully checked."
    },
    "paper_url": "https://doi.org/10.1038/s41467-021-25831-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports ten design–test–learn rounds, in-vivo microbial fatty-alcohol production and catalytic-rate characterization on palmitoyl-ACP.",
      "unknown": "Results are under the study’s microbial test conditions and do not establish industrial production; exact preprint-version results remain unchecked.",
      "ai_role": "Machine-learning models iteratively selected reductase sequences for experimental testing and subsequent model updates.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-021-25831-w"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-08-ml-gcamp-calcium-indicators",
    "title": "Machine learning guides engineering of fluorescent calcium indicators",
    "summary": "An ensemble of three regression models trained on experimentally derived GCaMP mutation libraries screened 1,423 uncharacterized variants. The study reports eGCaMP and eGCaMP+ with faster kinetics and larger fluorescence responses than previously published fast variants, and eGCaMP2+ with greater dynamic range than the tested sixth-, seventh- and eighth-generation indicators. The preprint reports cell and neuron assays. These are laboratory sensor results, not human or clinical validation. This is one study; the journal article is a later version.",
    "field": "biology",
    "ai_technology": "Ensemble of three regression models",
    "organization": "University of Washington",
    "source_url": "https://doi.org/10.21203/rs.3.rs-3146778/v1",
    "source_type": "paper",
    "publication_date": "2023-08-07",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary Research Square v1 full text, posted-date header, abstract/results, author affiliations and version-of-record note."
    },
    "paper_url": "https://www.nature.com/articles/s43588-024-00611-w",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The study reports experimental evaluation in vitro, HEK293 cells and primary neurons for selected GCaMP variants.",
      "unknown": "Results are laboratory sensor assays, not evidence of human or clinical use; the record uses the 2023 preprint version while the journal paper appeared in 2024.",
      "ai_role": "An ensemble of three regression models predicted mutation effects and screened 1,423 GCaMP variants for experimental follow-up.",
      "evidence_urls": [
        "https://assets-eu.researchsquare.com/files/rs-3146778/v1_covered_50aaf3d0-2f80-4222-91d1-d14c7e34a5f6.pdf?c=1711091168"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-04-knowledge-graph-antibiotic-resistance",
    "title": "Knowledge-graph-guided discovery of antibiotic-resistance associations",
    "summary": "KIDS combined an expert-curated knowledge graph with learned link-prediction models to rank gene–antibiotic resistance hypotheses for experiments. The authors report 93 validated associations involving 83 E. coli genes and 15 antibiotics, including six genes they describe as previously unknown resistance genes; five tested Salmonella homologs also showed resistance. These are laboratory gene–drug associations, not clinical validation. The article’s second-round percentage conflicts with its counts (29/90 is 32.2%, although the text states 28.8%). Count one study.",
    "field": "biology",
    "ai_technology": "Knowledge graph with PRA, multilayer perceptron, AdaBoost-stacked model, TransE and TransD",
    "organization": "University of California, Davis",
    "source_url": "https://www.nature.com/articles/s41467-022-29993-z",
    "source_type": "paper",
    "publication_date": "2022-04-29",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary Nature indexed full text, article dates, affiliations, methods/results, contributions and stated counting discrepancy."
    },
    "paper_url": "https://doi.org/10.1038/s41467-022-29993-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors report wet-lab validation of 93 gene–antibiotic associations, including five tested Salmonella homologs.",
      "unknown": "Reported percentages are internally inconsistent for 29 of 90 second-round tests; these assays do not establish clinical relevance or treatment outcomes.",
      "ai_role": "KIDS ranked gene–antibiotic links from an expert-curated graph with learned link-prediction models, directing experiments to candidate resistance associations.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-022-29993-z"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-11-deep-learning-cascadia-slow-slip",
    "title": "Deep learning detects slow slip events in Cascadia GNSS data",
    "summary": "A detector combining convolutional and attention-based neural networks, trained on synthetic geodetic time series, was applied to raw GNSS data in Cascadia from 2007 to 2022. The authors report 78 detections, including 35 of 40 events in the 2007–2017 comparison catalog and 43 additional events. Detection patterns align with tremor activity, but the paper says independent acquisition-data validation is needed, and shorter/low-confidence events remain uncertain. This method is specific to this station network and does not predict earthquakes.",
    "field": "physics",
    "ai_technology": "Convolutional neural network and Transformer attention model",
    "organization": "ISTerre, Université Grenoble Alpes; Université Savoie Mont Blanc; CNRS; IRD; Université Gustave Eiffel; GIPSA-lab, Université Grenoble Alpes; Grenoble INP; Institut Universitaire de France",
    "source_url": "https://www.nature.com/articles/s43247-023-01107-7",
    "source_type": "paper",
    "publication_date": "2023-11-24",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher full text, abstract, methods/results, affiliations and publication history."
    },
    "paper_url": "https://doi.org/10.1038/s43247-023-01107-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Applied detector outputs were compared with a prior event catalog and tremor activity in real Cascadia GNSS data.",
      "unknown": "Some additional/short detections remain uncertain; article states validation with an independent acquisition dataset is needed. This is not earthquake prediction.",
      "ai_role": "A CNN plus Transformer trained on synthetic GNSS series classified slow-slip windows in raw multistation GNSS data and produced event detections.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43247-023-01107-7"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-02-deep-clustering-turkiye-tremor",
    "title": "Deep clustering surfaces tremor-like episodes before the 2023 Türkiye earthquake",
    "summary": "An autoencoder and deep embedded clustering surfaced low-frequency tremor-like episodes in seismic records preceding the 2023 Türkiye earthquake. The authors then used source-location and particle-motion analyses to suggest an association with nearby cement-quarry activity. They explicitly state that evidence is insufficient to establish that industrial activity caused or influenced the earthquake; the paper is a retrospective signal interpretation, not an earthquake forecast or causal finding.",
    "field": "physics",
    "ai_technology": "Autoencoder and deep embedded clustering",
    "organization": "Helmholtz Centre Potsdam, GFZ German Research Centre for Geosciences; Stanford University",
    "source_url": "https://www.nature.com/articles/s41598-025-88381-x",
    "source_type": "paper",
    "publication_date": "2025-02-21",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PubMed metadata/abstract plus full primary article PDF from institutional repository; methods, conclusion and affiliations."
    },
    "paper_url": "https://doi.org/10.1038/s41598-025-88381-x",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study applies clustering to seismic records and follows with source-location and particle-motion analyses.",
      "unknown": "The authors state they cannot establish a causal relation between quarry activity and the earthquake; this is retrospective and not a forecast.",
      "ai_role": "Deep embedded clustering of autoencoder representations identified low-frequency tremor-like episodes for subsequent geophysical source analysis.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/39984511/",
        "https://refubium.fu-berlin.de/bitstream/handle/fub188/47316/s41598-025-88381-x.pdf?isAllowed=y&save=y&sequence=3"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-05-ai-generated-cell-penetrating-peptide",
    "title": "AI-generated cell-penetrating peptide selected with molecular-dynamics simulations",
    "summary": "An RNN generated peptide sequences and molecular-dynamics-based scoring prioritized Pep-MD for synthesis. In cell assays, the authors report less toxicity than TAT and stronger cell permeability under the tested conditions. The paper proposes a membrane-entry mechanism based partly on simulated structures; these cell results do not establish therapeutic efficacy.",
    "field": "biology",
    "ai_technology": "Recurrent neural network sequence generator with molecular-dynamics-based selection score",
    "organization": "Tokyo Institute of Technology; RIKEN; The University of Tokyo; National Institute for Materials Science",
    "source_url": "https://www.nature.com/articles/s41598-021-90245-z",
    "source_type": "paper",
    "publication_date": "2021-05-20",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher full PDF, methods/results, publication information and author affiliations."
    },
    "paper_url": "https://doi.org/10.1038/s41598-021-90245-z",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The selected synthesized peptide was tested in cell-viability, confocal-imaging and FACS permeability assays against TAT.",
      "unknown": "Cell-based assays and a single comparator; no clinical or therapeutic efficacy evidence; mechanistic interpretation partly depends on modeled structures.",
      "ai_role": "An RNN generated candidate peptides; a simulation-derived score prioritized one sequence for synthesis and experimental cell testing.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41598-021-90245-z.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-09-machine-learning-fano-dimension",
    "title": "Machine learning helps recover dimension from quantum periods for a class of Fano varieties",
    "summary": "A feed-forward neural network classified the dimension of Fano varieties from quantum-period coefficients with 98% accuracy on the authors’ dataset. The authors report that machine learning helped formulate rigorous asymptotic results, which prove dimension recovery for specified low-Picard-rank toric Fano classes. The broader conjecture that the quantum period uniquely determines every Fano variety remains open; the empirical classification accuracy is not itself a proof. The journal article appeared 2023-09-08, before the same work’s 2023-09-11 arXiv posting.",
    "field": "mathematics",
    "ai_technology": "Feed-forward neural network",
    "organization": "Imperial College London; University of Nottingham",
    "source_url": "https://www.nature.com/articles/s41467-023-41157-1",
    "source_type": "paper",
    "publication_date": "2023-09-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary author-hosted full manuscript plus primary publisher article metadata; checked authors, method, theorem scope and chronology."
    },
    "paper_url": "https://doi.org/10.1038/s41467-023-41157-1",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article combines a dataset classification result with rigorous asymptotic proofs for specified classes.",
      "unknown": "98% is empirical accuracy on the paper’s dataset; the broad uniqueness conjecture remains open. Journal publication preceded arXiv submission.",
      "ai_role": "A neural network analyzed quantum-period coefficients and helped identify dimension-related patterns used to formulate asymptotic results, which the authors prove for restricted classes.",
      "evidence_urls": [
        "https://kasprzyk.work/research/pdf/ml_dimension.pdf",
        "https://www.nature.com/articles/s41467-023-41157-1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-08-ml-affine-deligne-lusztig-dimension",
    "title": "Machine learning assists a proved dimension bound for affine Deligne–Lusztig varieties",
    "summary": "The authors used computational data and machine-learning pattern exploration to study affine Deligne–Lusztig varieties. They report rediscovering the previously known virtual-dimension formula, then using the exploration to formulate and prove a new lower bound on actual dimension. The mathematical theorem, not model accuracy or rediscovery of the known formula, is the specific result. The 2024 journal article and 2023 arXiv v1 are versions of one study.",
    "field": "mathematics",
    "ai_technology": "Linear models, support-vector machines and neural networks",
    "organization": "Beijing International Center for Mathematical Research, Peking University; Center for Machine Learning Research, Peking University; University of Hong Kong",
    "source_url": "https://arxiv.org/html/2308.11355v1",
    "source_type": "paper",
    "publication_date": "2023-08-22",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary arXiv v1 full text and primary journal indexed full text, including methods, theorem, affiliations and version dates."
    },
    "paper_url": "https://doi.org/10.1007/s42543-024-00086-8",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The paper reports mathematical proof of a lower bound prompted by machine-learning exploration.",
      "unknown": "The virtual-dimension formula was previously known and is not the new theorem; the work is not a formal automated proof.",
      "ai_role": "Models explored generated mathematical data and highlighted patterns that led mathematicians to formulate and prove a new lower bound.",
      "evidence_urls": [
        "https://arxiv.org/html/2308.11355v1",
        "https://link.springer.com/article/10.1007/s42543-024-00086-8"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-ml-galois-criteria",
    "title": "Machine-learning analysis motivates proved criteria for Galois groups",
    "summary": "Interpretable decision trees applied to Dedekind-zeta coefficients helped the authors identify patterns in Galois extensions of degrees 4, 6, 8, 9 and 10. They formulated and proved mathematical criteria, including a nonic result: for the stated class, a₁₀₀₀(K) ≤ 4.5 implies the extension is cyclic. This theorem is a specific result; dataset classification accuracy is not a universal guarantee, and the paper reports a counterexample to an apparently perfect classifier branch. The arXiv v1 of 2025-08-08 and the 2026 journal version are one study.",
    "field": "mathematics",
    "ai_technology": "Interpretable decision trees and logistic regression",
    "organization": "Department of Mathematics, University of Connecticut; Korea Institute for Advanced Study; Department of Mathematics, University of California, Berkeley",
    "source_url": "https://arxiv.org/html/2508.06670v1",
    "source_type": "paper",
    "publication_date": "2025-08-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary arXiv v1 and publisher full text, including result, proof scope, dates and affiliations."
    },
    "paper_url": "https://doi.org/10.1007/s40687-026-00620-3",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors report conventional proofs of criteria motivated by model interpretation.",
      "unknown": "The criteria apply to stated classes; model accuracy is data-dependent and a branch with apparent perfect performance has a stated out-of-sample counterexample.",
      "ai_role": "Decision-tree patterns in number-field coefficients motivated mathematical criteria that the authors then proved using conventional reasoning.",
      "evidence_urls": [
        "https://arxiv.org/html/2508.06670v1",
        "https://link.springer.com/article/10.1007/s40687-026-00620-3"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-01-ml-high-throughput-fe-nb-b-glass",
    "title": "Machine learning guides discovery of hard Fe–Nb–B metallic-glass films",
    "summary": "Random-forest models for glass formation and hardness helped select Fe–Nb–B composition regions for sputtered-film experiments. The study reports a maximum measured hardness of 29 GPa for Fe₃Nb₂₅B₇₂ and an H/E ratio of 0.09 for Fe₂₀Nb₁₂B₆₈. H/E is used as an estimate of wear resistance, not a direct wear test. Samples were 500–800 nm films with substrate-effect corrections; bulk properties and industrial production were not demonstrated. This extends earlier machine-learning/high-throughput work on metallic glasses and is one study.",
    "field": "materials",
    "ai_technology": "Random-forest models for glass formation and hardness",
    "organization": "Stanford Synchrotron Radiation Lightsource, SLAC National Accelerator Laboratory; Colorado School of Mines; National Renewable Energy Laboratory; Argonne National Laboratory; University of Maryland",
    "source_url": "https://www.nrel.gov/docs/fy22osti/80853.pdf",
    "source_type": "paper",
    "publication_date": "2022-01-10",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary full article PDF, title/DOI/date, affiliations, methods, composition/result and measurement limitations."
    },
    "paper_url": "https://doi.org/10.1063/5.0068207",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Nanoindentation and synchrotron characterization were used to test sputtered films and measure hardness and structure.",
      "unknown": "Thin-film results; H/E is an estimated wear-resistance metric, not direct wear testing or evidence of bulk/industrial performance.",
      "ai_role": "Random-forest predictions helped select Fe–Nb–B composition regions for high-throughput sputtered-film synthesis and measurement.",
      "evidence_urls": [
        "https://www.nrel.gov/docs/fy22osti/80853.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2020-09-active-deep-learning-lamost-emission-objects",
    "title": "Active deep learning identifies emission-line object candidates in LAMOST spectra",
    "summary": "An active-learning workflow with a 12-layer convolutional network analyzed about 4.1 million LAMOST DR2 spectra and used human labeling of uncertain cases. After visual inspection and rejection of 58 poor spectra, the paper reports 4,321 spectra from about 3,788 objects, including 1,013 spectra of 948 objects not matched to SIMBAD or the specified Hou et al. 2016 catalog. Visual inspection supported emission signatures, but catalog non-match does not prove universal prior absence and proposed subtypes remain hypotheses. Count the catalog study once.",
    "field": "physics",
    "ai_technology": "12-layer one-dimensional convolutional neural network with uncertainty-sampling active learning",
    "organization": "Astronomical Institute of the Czech Academy of Sciences; Faculty of Information Technology, Czech Technical University in Prague",
    "source_url": "https://arxiv.org/abs/2009.03219",
    "source_type": "paper",
    "publication_date": "2020-09-07",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary arXiv metadata/abstract and journal full indexed results, publication dates, affiliations, and stated catalog limits."
    },
    "paper_url": "https://doi.org/10.1051/0004-6361/201936090",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The paper reports human-oracle labeling and visual inspection of candidates, confirming emission signatures in retained spectra.",
      "unknown": "“New” means not matched to the named SIMBAD/Hou catalogs at the study date; proposed object classes require further classification.",
      "ai_role": "An active-learning CNN prioritized emission-line spectra in a large archive and asked human annotators to label uncertain examples.",
      "evidence_urls": [
        "https://arxiv.org/abs/2009.03219",
        "https://doi.org/10.1051/0004-6361/201936090"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-08-hybrid-ml-metallic-glasses",
    "title": "Hybrid machine learning guides experimental discovery of multicomponent metallic glasses",
    "summary": "Hybrid classification and regression models with physically motivated descriptors, trained on about 5,000 metallic-glass compositions, guided exploration of multicomponent alloy compositions. Researchers synthesized 23 selected compositions: 12 predicted glass-formers produced glass in bulk, ribbon or both, while 11 predicted negative comparisons did not form glass under the tested preparation conditions. Six of the positive designs formed bulk glass; six formed glassy ribbons only. These selected experiments are not an unbiased estimate of global predictive accuracy. Some authors disclosed inventorship on a filed patent related to the new bulk compositions; composition-level prior art and independent replication remain open.",
    "field": "materials",
    "ai_technology": "Hybrid classification/regression models including adaptive boosting, SVM, KNN and rational-quadratic Gaussian-process regression",
    "organization": "City University of Hong Kong",
    "source_url": "https://www.nature.com/articles/s41524-021-00607-4",
    "source_type": "paper",
    "publication_date": "2021-08-23",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher full indexed article and institutional full-text copy, methods, results, date, affiliation and patent disclosure."
    },
    "paper_url": "https://doi.org/10.1038/s41524-021-00607-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reports synthesis and diffraction/thermal characterization of selected predicted positive and negative compositions.",
      "unknown": "Selected validation set, not unbiased global accuracy; patent inventorship disclosed and composition-level novelty/replication remain open.",
      "ai_role": "Hybrid models trained on prior metallic-glass data prioritized compositions for synthesis and measurement.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-021-00607-4",
        "https://scholars.cityu.edu.hk/ws/portalfiles/portal/86074146/82269007.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-11-ml-selected-eu2-phosphors",
    "title": "Machine learning selects Eu²⁺-activated green and blue-green phosphors",
    "summary": "A bagged gradient-boosted regression model trained on 129 Eu²⁺-activated phosphors predicted emission wavelengths and helped select 20 host candidates for synthesis. The authors report three Eu²⁺-activated phosphors with green or blue-green emissions: Li₂Ca₄Si₄O₁₃:Eu²⁺, Na₂Ca₂Si₂O₇:Eu²⁺, and SrLaGaO₄:Eu²⁺. The host compounds were known; novelty concerns their activator-specific luminescence. Some powders contained impurities, and selected-particle diffraction/spectroscopy characterization does not establish phase-pure bulk yields or device suitability. This is one campaign.",
    "field": "materials",
    "ai_technology": "Bagged gradient-boosted regression trees",
    "organization": "Research and Services Division of Materials Data and Integrated System, National Institute for Materials Science; Osaka Metropolitan University; Research Center for Functional Materials, National Institute for Materials Science",
    "source_url": "https://pubs.rsc.org/en/content/articlelanding/2023/ma/d2ma00881e",
    "source_type": "paper",
    "publication_date": "2022-11-29",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary full article/PDF, date, affiliations, methods/results, candidate measurements and stated limitations."
    },
    "paper_url": "https://doi.org/10.1039/D2MA00881E",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Twenty candidates were synthesized and characterized; three showed designed green or blue-green Eu²⁺ emission.",
      "unknown": "Some samples contain impurity phases; selected-particle evidence and host-specific results do not establish phase-pure bulk/device performance or complete prior-art priority.",
      "ai_role": "A model trained on known phosphors ranked host candidates by predicted emission wavelength for experimental synthesis and characterization.",
      "evidence_urls": [
        "https://mdr.nims.go.jp/filesets/fcaeedc2-0180-470c-af42-ec87e7356505/download"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-07-ml-guided-fe-mn-ni-nox-catalyst",
    "title": "Machine learning guides Fe–Mn–Ni catalysts for NOx reduction",
    "summary": "An artificial neural network trained on literature data and a genetic algorithm helped identify Fe–Mn–Ni catalyst compositions for ammonia selective catalytic reduction. Researchers synthesized and tested model-guided candidates and report good NOx conversion for Fe₀.₃₁₉Mn₀.₂₀₆Ni₀.₄₇₅ over the tested temperature range. The work combines model selection, human screening and controlled experiments; the ratio is not a complete oxide formula and does not demonstrate industrial exhaust performance. The article text differs on whether the workflow involved two or four iterations, so this summary avoids an iteration count.",
    "field": "chemistry",
    "ai_technology": "Artificial neural network and genetic algorithm",
    "organization": "State Environmental Protection Key Laboratory of Integrated Surface Water-Groundwater Pollution Control, Southern University of Science and Technology",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10666265/",
    "source_type": "paper",
    "publication_date": "2023-07-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PMC full text, PubMed metadata/affiliation and publisher date chronology."
    },
    "paper_url": "https://doi.org/10.1021/acs.est.3c00293",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports synthesis and catalytic conversion testing of selected Fe–Mn–Ni formulations.",
      "unknown": "Article reports inconsistent iteration counts; test range and lab results do not establish durability or practical flue-gas performance.",
      "ai_role": "An ANN and genetic search ranked candidate Fe–Mn–Ni compositions; experimental results were used in the iterative optimization workflow.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10666265/",
        "https://pubmed.ncbi.nlm.nih.gov/37393584/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-11-ml-guided-sortase-a-evolution",
    "title": "Machine-learning-guided evolution identifies higher-activity sortase A variants",
    "summary": "Two machine-learning-guided directed-evolution series for sortase A were run with and without the known high-activity 5M variant in the initial training set. The authors report new variants with higher activity than 5M in a GFP/YFP cross-linking assay. This is not a 2.2–2.5-fold improvement over wild type; the known 5M comparator is already engineered. Feature selection used ability to find 5M and may bias exploration. The related GB1 analysis uses an existing dataset, not a second laboratory discovery. Sequence-level comparison with later sortase work remains open.",
    "field": "biology",
    "ai_technology": "Gaussian-process Bayesian selection model",
    "organization": "Tohoku University",
    "source_url": "https://pubs.acs.org/accacs/article/11/23/14615/417029/Machine-Learning-Guided-Library-Design-Cycle-for",
    "source_type": "paper",
    "publication_date": "2021-11-19",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary ACS indexed full article and Tohoku research announcement; methods/results and journal timing."
    },
    "paper_url": "https://doi.org/10.1021/acscatal.1c03753",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Constructed variants were assayed for SrtA-catalyzed GFP/YFP cross-linking; resulting variants exceeded the 5M comparator.",
      "unknown": "The assay is specific; 5M is an already high-activity engineered comparator; feature selection may bias to 5M; independent sequence-level overlap/replication open.",
      "ai_role": "A Gaussian-process ML model ranked variants from initial and second-round data to design subsequent SrtA mutant libraries.",
      "evidence_urls": [
        "https://pubs.acs.org/accacs/article/11/23/14615/417029/Machine-Learning-Guided-Library-Design-Cycle-for",
        "https://www.eng.tohoku.ac.jp/news/detail-%2C-id%2C2033.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-05-ml-guided-high-silica-faujasite",
    "title": "Machine learning guides direct synthesis of higher-silica faujasite",
    "summary": "Machine-learning models learned relationships between crystallization composition/conditions and zeolite characteristics from prior synthesis records, guiding human selection of conditions for faujasite synthesis. The team reports directly synthesized, unseeded, organic-structure-directing-agent-free FAU with Si/Al 3.5, compared with the prior reported maximum 2.8. Following ion exchange, laboratory tests report improved propane cracking and dehydrogenation rates per proton site relative to the lower-silica comparator. The FAU framework was already known; this result concerns composition and route, not a new framework or industrial performance.",
    "field": "chemistry",
    "ai_technology": "Geometric harmonics, neural networks and Gaussian-process regression",
    "organization": "University of Minnesota; Dalian University of Technology; Johns Hopkins University; Argonne National Laboratory; California Institute of Technology; Johns Hopkins University Applied Physics Laboratory",
    "source_url": "https://authors.library.caltech.edu/records/makq9-r3046",
    "source_type": "paper",
    "publication_date": "2023-05-31",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary Caltech institutional record and deposited published article/supplement; date, affiliations, methods and abstract/results."
    },
    "paper_url": "https://doi.org/10.1038/s41467-023-38738-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Researchers synthesized and characterized FAU and tested acid catalysts in propane cracking/dehydrogenation under laboratory conditions.",
      "unknown": "Known framework; activity comparison is assay-specific and does not establish industrial gains.",
      "ai_role": "Models trained on prior synthesis records identified composition/condition settings that guided direct high-silica FAU synthesis.",
      "evidence_urls": [
        "https://authors.library.caltech.edu/records/makq9-r3046"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-04-ml-guided-ngal-dna-aptamer",
    "title": "Machine learning guides refinement of DNA aptamers binding NGAL",
    "summary": "Neural models trained on particle-display affinity-screening data guided sequence mutation and candidate nomination, followed by synthesis and experimental affinity testing. The authors report a 23-nucleotide truncation of candidate G13 with measured dissociation constant 1.5 nM, compared with about 8 nM for the full-length parent under their assay. This is a binding result, not demonstrated diagnostic or therapeutic efficacy. The screening set is not 187,499 independently validated agents; sequence-level prior art remains open.",
    "field": "biology",
    "ai_technology": "Neural-network models trained on particle-display affinity data",
    "organization": "Google Research; Aptitude Medical Systems",
    "source_url": "https://www.nature.com/articles/s41467-021-22555-9",
    "source_type": "paper",
    "publication_date": "2021-04-22",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher indexed full text, article dates, methods/results, affiliations and disclosed limitations."
    },
    "paper_url": "https://doi.org/10.1038/s41467-021-22555-9",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reports particle-display screening and measured NGAL binding, including full Kd for truncated and parent G13.",
      "unknown": "Binding assay only; no diagnostic/therapeutic outcome; one holdout set used for tuning and sequence novelty requires further review.",
      "ai_role": "Trained neural models extrapolated from experimentally measured sequences to nominate and iteratively refine higher-affinity DNA aptamers and truncations.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-021-22555-9"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-05-ml-guided-mo-sb-nir-phosphor",
    "title": "Machine learning guides a near-infrared Mo/Sb halide phosphor",
    "summary": "A regression model trained on experimental and theoretical data guided Mo/Sb/Br co-doping for a lead-free near-infrared phosphor. The authors report synthesizing Cs₂Zr(Cl₀.₄₆Br₀.₅₄)₆:12%Mo⁴⁺/3.6%Sb³⁺, with emission near 920 nm, internal quantum efficiency of 92.4%, and external quantum efficiency of 65.9% under the reported conditions. Its measured intensity was about 20% higher than the authors’ orthogonal-design comparator. A separate diffusion model was used for image reconstruction demonstrations; this does not establish clinical diagnostic performance. Earlier composition-level prior art and independent replication remain open.",
    "field": "materials",
    "ai_technology": "Transformer-based regression model",
    "organization": "Guangxi University; South China University of Technology",
    "source_url": "https://www.nature.com/articles/s41467-026-73105-0",
    "source_type": "paper",
    "publication_date": "2026-05-15",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary indexed publisher article and NLM publication metadata; abstract/results, model roles, online date and measured outcomes."
    },
    "paper_url": "https://doi.org/10.1038/s41467-026-73105-0",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Authors report synthesis and optical characterization of the predicted composition, LED device testing, and separate imaging demonstration.",
      "unknown": "Study-specific efficiency/intensity results; earlier independent reports and composition-level priority remain unverified; imaging is not clinical diagnostic validation.",
      "ai_role": "A model trained on experimental/theoretical data predicted co-dopant concentrations for researchers to synthesize and test; a separate diffusion model reconstructed imaging outputs.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-73105-0",
        "https://pubmed.ncbi.nlm.nih.gov/42140992/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-ml-selected-nickel-one-phenyl-complexes",
    "title": "Machine learning helps select nickel(I)–phenyl complexes that insert CO₂ at room temperature",
    "summary": "The researchers calculated descriptors for 132 bidentate phosphine ligands, used k-means clustering to identify 21 candidate ligands, then applied DFT barrier calculations and human selection before synthesis. They isolated and characterized Ni(I)–phenyl complexes, including an X-ray-confirmed i-Pr-Xantphos complex. The i-Pr-Xantphos and dtbpf complexes underwent CO₂ insertion at room temperature in the reported experiments; nickel carboxylation was also tested. The study does not establish CO₂ insertion or Ni(I) chemistry generally as new; its result concerns bounded ligand selection/speciation.",
    "field": "chemistry",
    "ai_technology": "Unsupervised k-means clustering of DFT-derived ligand descriptors",
    "organization": "Institute of Organic Chemistry, RWTH Aachen University; University of Jyväskylä",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12395408/",
    "source_type": "paper",
    "publication_date": "2025-07-18",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary JACS full text, abstract, methods/results, chronology and affiliations."
    },
    "paper_url": "https://doi.org/10.1021/jacs.5c00441",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Synthesis, X-ray characterization, CO₂ insertion assays and catalytic carboxylation tests reported for selected Ni complexes.",
      "unknown": "Only representative ligands from predicted groups were experimentally evaluated; prior Ni(I)-CO₂ chemistry exists; results do not imply commercial carbon capture.",
      "ai_role": "Unsupervised clustering of computed descriptors identified a set of ligands to evaluate; DFT and researcher selection further narrowed candidates before synthesis and testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12395408/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-11-ml-guided-yadh-nonenal",
    "title": "Machine learning guides yeast alcohol dehydrogenase specificity for trans-2-nonenal",
    "summary": "Researchers measured a training library of yeast alcohol dehydrogenase I variants, used support-vector activity regression/classification and a thermostability model to rank combinations, then tested 22 prioritized variants. Four showed higher specific activity toward trans-2-nonenal than wild type; LWFMY also showed improved thermostability under the study’s tests. The result is laboratory enzyme engineering, not an industrial process or validated product. The primary journal version was published 2026-11-01; a search result for an earlier-hosted copy had ambiguous date/provenance, so this entry does not claim it was the first public version.",
    "field": "biology",
    "ai_technology": "Support-vector regression/classification models and docking-informed mutation selection",
    "organization": "RIKEN Center for Sustainable Resource Science; Graduate School of Science, Technology, and Innovation, Kobe University",
    "source_url": "https://link.springer.com/article/10.1186/s13068-026-02839-2",
    "source_type": "paper",
    "publication_date": "2026-11-01",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary Springer full article and PubMed metadata; methods, dates, affiliations and outcomes."
    },
    "paper_url": "https://doi.org/10.1186/s13068-026-02839-2",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The top 22 ranked variants were experimentally evaluated; four outperformed wild type on trans-2-nonenal, with one also improved in thermal stability.",
      "unknown": "Small selected validation set; specific activity differs from catalytic efficiency across substrate concentrations; ambiguous earlier-hosted copy means first-public date is not asserted.",
      "ai_role": "Models trained on measured variant data ranked combinations in unexplored sequence space for experimental testing.",
      "evidence_urls": [
        "https://link.springer.com/article/10.1186/s13068-026-02839-2",
        "https://pubmed.ncbi.nlm.nih.gov/42972448/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-02-gp-guided-ketoreductase-ipatasertib-intermediate",
    "title": "Gaussian-process-assisted engineering of a ketoreductase for an ipatasertib precursor",
    "summary": "A six-round engineering campaign combined mutational scans, structure-guided libraries and Gaussian-process models. The model-aided library yielded M3, reported with 22-fold higher apparent activity than wild type in the UV assay. Three subsequent conventional mutagenesis rounds produced M6, a 10-substitution descendant with 64-fold higher apparent kcat, but also about 12-fold higher apparent KM; do not attribute M6 as a direct model output or treat kcat as production yield. The authors report a 100 g/L laboratory process test with ≥98% conversion and 99.7% diastereomeric excess after 30 hours. These are process-relevant laboratory results, not a commercial production run or drug efficacy.",
    "field": "biology",
    "ai_technology": "Gaussian-process regression",
    "organization": "Institute of Chemistry and Biotechnology, Zurich University of Applied Sciences; F. Hoffmann-La Roche Ltd.",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10902378/",
    "source_type": "paper",
    "publication_date": "2024-02-28",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PMC full article, methods/results, publication history, affiliations and chronology of model and conventional rounds."
    },
    "paper_url": "https://doi.org/10.1038/s42004-024-01130-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "M3 and later M6 were tested for activity; article reports preparative reactions at 100 g/L substrate and conversion/stereoselectivity under specified laboratory conditions.",
      "unknown": "M6 follows conventional site-mutagenesis rounds after the AI-informed step; apparent kcat differs from process yield and KM changed; no commercial production/clinical efficacy claim.",
      "ai_role": "Gaussian-process predictions informed smaller follow-up variant libraries within a broader workflow that also used mutational scanning, rational design and conventional evolution.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10902378/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-ml-guided-emt-zeolite-synthesis",
    "title": "ML-guided synthesis conditions produce EMT zeolite",
    "summary": "Researchers trained models on 174 prior synthesis attempts and selected six EMT synthesis conditions for testing. Five yielded EMT by X-ray diffraction; two used starting-mixture Si/Al ratios outside the training range. EMT is an established framework, and the small, researcher-selected study does not establish a general success rate or independent replication.",
    "field": "materials",
    "ai_technology": "XGBoost classifier, PCA and pretrained foundation models",
    "organization": "University of Houston",
    "source_url": "https://arxiv.org/abs/2608.03760",
    "source_type": "paper",
    "publication_date": "2026-08-04",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "As described in the primary source",
      "unknown": "Independent replication and global priority not established by this review.",
      "ai_role": "XGBoost classifier, PCA and pretrained foundation models",
      "evidence_urls": [
        "https://arxiv.org/abs/2608.03760"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher/preprint material inspected for identity, AI role and bounded result."
    }
  },
  {
    "id": "2023-12-bo-nonnoble-oer-catalysts",
    "title": "Bayesian optimization identifies active non-noble oxygen-evolution oxides",
    "summary": "Multiobjective Bayesian optimization guided robotic synthesis and electrochemical testing in a researcher-defined Co–Mn–Sb–Sn–Ti oxide space. The study identifies Mn-rich candidates and reports improved activity after Ti incorporation into MnOx following short accelerated-stress testing. Its 17-fold acceleration compares measurement time with a hypothetical exhaustive triplicated grid; it does not establish commercial electrolyzer lifetime or universal superiority.",
    "field": "materials",
    "ai_technology": "Multiobjective Bayesian optimization with multitask Gaussian-process surrogate",
    "organization": "Forschungszentrum Jülich; Friedrich-Alexander University Erlangen-Nürnberg; Helmholtz Institute Erlangen-Nürnberg for Renewable Energy",
    "source_url": "https://publikationen.bibliothek.kit.edu/1000167576",
    "source_type": "paper",
    "publication_date": "2023-12-13",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "As described in the primary source",
      "unknown": "Independent replication and global priority not established by this review.",
      "ai_role": "Multiobjective Bayesian optimization with multitask Gaussian-process surrogate",
      "evidence_urls": [
        "https://publikationen.bibliothek.kit.edu/1000167576"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher/preprint material inspected for identity, AI role and bounded result."
    }
  },
  {
    "id": "2025-11-calcamp-gdst-antimicrobial-peptides",
    "title": "ML prioritization identifies two antimicrobial peptide candidates",
    "summary": "CalcAMP ranked 16,384 14-residue sequences in a researcher-defined LL-37-derived design space bounded by patent WO2015088344. Researchers selected GDST-038 and GDST-045 for laboratory tests; the paper reports activity against tested bacteria and elimination of detectable adherent S. aureus in a 3D human epidermal infection model at concentrations of at least 15 µM. Retro-inverso variants were prepared and tested after selection. These results do not establish human safety or clinical efficacy, and exact sequence-level patent priority remains open.",
    "field": "biology",
    "ai_technology": "CalcAMP machine-learning activity ranking",
    "organization": "Amsterdam University Medical Centre; Leiden University; Leiden University Medical Center",
    "source_url": "https://www.mdpi.com/2079-6382/14/11/1172",
    "source_type": "paper",
    "publication_date": "2025-11-20",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary MDPI full article, abstract/results and author-affiliation block reviewed."
    },
    "paper_url": "https://doi.org/10.3390/antibiotics14111172",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary article reports in-vitro antimicrobial testing, biofilm assays and a 3D human epidermal infection model for selected parent/retro-inverso peptides.",
      "unknown": "These are in-vitro and tissue-model findings, not clinical efficacy or human safety. Sequence-level patent scope and independent replication are not established.",
      "ai_role": "CalcAMP ranked peptide sequences by predicted antimicrobial activity; researchers selected and experimentally tested two lead sequences and retro-inverso variants.",
      "evidence_urls": [
        "https://www.mdpi.com/2079-6382/14/11/1172"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-arcadiamp-arcinin",
    "title": "Diffusion-guided peptide design yields antimicrobial candidate Arcinin",
    "summary": "ARCADIAMP combines iterative discrete diffusion generation, ESM2-based activity classifiers and sequence novelty filtering. Eight of ten experimentally screened candidates met the paper’s MIC threshold; Arcinin showed activity against tested pathogens and reduced bacterial burden in a mouse wound infection model. These are selected preclinical findings, not evidence of human safety or treatment efficacy; formal resistance-evolution testing remains future work.",
    "field": "biology",
    "ai_technology": "Iterative discrete diffusion model with ESM2 classifiers and novelty filter",
    "organization": "University of California, Davis",
    "source_url": "https://www.nature.com/articles/s41467-026-75030-8",
    "source_type": "paper",
    "publication_date": "2026-07-07",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "As described in the primary source",
      "unknown": "Independent replication and global priority not established by this review.",
      "ai_role": "Iterative discrete diffusion model with ESM2 classifiers and novelty filter",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-75030-8"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher/preprint material inspected for identity, AI role and bounded result."
    }
  },
  {
    "id": "2025-10-dlfea4ampgen-antimicrobial-peptides",
    "title": "Deep-learning feature-guided design yields antimicrobial peptides D1 and D2",
    "summary": "DLFea4AMPGen uses deep-learning activity models and SHAP feature attribution to define peptide fragments used in a candidate sequence space. Sixteen selected peptides were synthesized; the primary article reports 12 with at least two tested activities and identifies D1 and D2 as active against drug-resistant bacterial isolates. D1 also reduced bacterial load in a reported mouse infection experiment. This is preclinical evidence, not human treatment or safety; the study does not establish that every generated sequence or all three activity classes were validated.",
    "field": "biology",
    "ai_technology": "Deep-learning activity prediction and SHAP-guided feature-fragment generation",
    "organization": "Chinese Academy of Agricultural Sciences; Inner Mongolia University; Academy of Military Medical Sciences; Western Medical Branch of PLA General Hospital",
    "source_url": "https://www.nature.com/articles/s41467-025-64378-y.pdf",
    "source_type": "paper",
    "publication_date": "2025-10-15",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the described assays.",
      "unknown": "Independent replication and exhaustive sequence-priority review not established.",
      "ai_role": "Deep-learning activity prediction and SHAP-guided feature-fragment generation",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-64378-y.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher article/indexed full text and bibliographic metadata reviewed for identity, AI role, dates, affiliations, and bounded result."
    }
  },
  {
    "id": "2025-08-zeobind-osda-zeolite-compositions",
    "title": "Machine-learning screening identifies zeolite structure-directing agents",
    "summary": "ZeoBind screened nearly 500 million zeolite–molecule pairs from a library of 2.3 million hypothetical, synthetically accessible organic structure-directing agents. The primary article reports two experimentally validated molecules that template ERI and CHA zeolites with novel compositions. The claim is limited to these tested framework/material outcomes; it does not establish exhaustive coverage of OSDA space or independent replication. Journal publication date is used because a prior preprint exists and its version chronology is not resolved here.",
    "field": "materials",
    "ai_technology": "Machine-learning predictive models for zeolite–molecule matching and screening",
    "organization": "Massachusetts Institute of Technology; Universitat Politècnica de València; Spanish National Research Council",
    "source_url": "https://www.nature.com/articles/s43588-025-00842-5",
    "source_type": "paper",
    "publication_date": "2025-08-07",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Reported experimental findings as stated in article.",
      "unknown": "Independent replication and broader priority not established.",
      "ai_role": "Machine-learning predictive models for zeolite–molecule matching and screening",
      "evidence_urls": [
        "https://www.nature.com/articles/s43588-025-00842-5"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher article reviewed for identity, affiliations, dates, AI role and bounded experimental result."
    }
  },
  {
    "id": "2025-12-ml-guided-crispri-isoprenol",
    "title": "ML-guided CRISPRi combinations improve isoprenol production",
    "summary": "An ensemble of machine-learning regressors recommended combinations of up to four CRISPRi guide RNAs during six design–build–test–learn cycles in engineered Pseudomonas putida. The paper reports an approximately fivefold isoprenol-titer improvement over its control for highlighted strains. It reports 347 unique arrays for the highlighted optimization and 472 unique strains overall; these are different denominators. A 2024 DOE meeting abstract describes an earlier campaign but does not quantify the later result; the journal article date is used without claiming first public disclosure. The finding is specific to this strain and laboratory workflow, not industrial fuel production.",
    "field": "biology",
    "ai_technology": "Automated Recommendation Tool ensemble of machine-learning regressors",
    "organization": "Lawrence Berkeley National Laboratory; Joint BioEnergy Institute; Lawrence Livermore National Laboratory; Sandia National Laboratories; University of California, Berkeley",
    "source_url": "https://www.nature.com/articles/s41467-025-66304-8",
    "source_type": "paper",
    "publication_date": "2025-12-13",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher article/PDF and institutional preprint metadata reviewed for identity, bounded outcomes, AI role, affiliations and chronology."
    },
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article reports six optimization cycles and about fivefold higher isoprenol titer versus the stated control, with distinct counts of 347 unique arrays and 472 unique strains overall.",
      "unknown": "The result is study-specific; industrial scale-up and independent replication are not established. The earlier DOE abstract establishes a related campaign, not the later quantitative result.",
      "ai_role": "The Automated Recommendation Tool, an ensemble of ML regressors, recommended guide-RNA combinations after initial genes and training data were supplied by researchers.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-66304-8_reference.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-04-active-learning-flaviolin-media",
    "title": "Active learning identifies salt-rich media for flaviolin production",
    "summary": "An automated recommendation tool proposed culture-medium mixtures for engineered Pseudomonas putida across three semi-automated campaigns. The paper reports direct flaviolin-titer improvements of about 70% and 60% in two campaigns and a 350% process-yield improvement in a third, each versus its campaign starting medium. The spectroscopic proxy changed by different amounts, so proxy values are not substituted for direct quantification. Results are specific to the tested strain, media and objective and do not establish industrial performance. A preprint preceded the journal article, but its exact posting date and version-specific outcome set are unresolved; the journal date is used without asserting first disclosure.",
    "field": "biology",
    "ai_technology": "Automated Recommendation Tool for active-learning medium optimization",
    "organization": "Lawrence Berkeley National Laboratory; Department of Energy Agile BioFoundry; Joint BioEnergy Institute; Sandia National Laboratories; University of California, Berkeley; BCAM, Basque Center for Applied Mathematics",
    "source_url": "https://www.nature.com/articles/s42003-025-08039-2",
    "source_type": "paper",
    "publication_date": "2025-04-18",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher article/PDF and institutional preprint metadata reviewed for identity, bounded outcomes, AI role, affiliations and chronology."
    },
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports direct titer changes of about 70% and 60% in two campaigns and a 350% yield change in a third, with each comparison tied to its own starting medium.",
      "unknown": "Outcomes are specific to the engineered strain and the tested campaigns; the earlier preprint date/version contents and independent replication are not established.",
      "ai_role": "The Automated Recommendation Tool proposed media formulations iteratively from experimental measurements; feature-importance analysis highlighted salt effects, but does not prove a unique causal mechanism.",
      "evidence_urls": [
        "https://www.nature.com/articles/s42003-025-08039-2.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-06-ml-guided-salt-hydration-measurements",
    "title": "Machine-learning screening guides thermochemical measurements of salt hydrates",
    "summary": "Composition-based machine-learning models screened salt-hydration reactions, after which researchers synthesized and measured selected candidates by differential scanning calorimetry. The article reports volumetric energy densities of 1.74 ± 0.02, 1.73 ± 0.02 and 1.70 ± 0.02 GJ/m³ for three rubidium Tutton salt hydrates. These compounds were already known; the bounded result is AI-guided thermochemical characterization, not new compound synthesis or demonstrated heat-storage deployment. An earlier preprint contains the measurements, but its precise public date/version chronology is unresolved, so no first-public claim is made.",
    "field": "materials",
    "ai_technology": "Stoichiometry-based machine-learning models of salt hydration properties",
    "organization": "University of Amsterdam; Radboud University",
    "source_url": "https://www.nature.com/articles/s43246-026-01200-2",
    "source_type": "paper",
    "publication_date": "2026-06-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports the described measurements.",
      "unknown": "Independent replication and global priority are not established.",
      "ai_role": "Stoichiometry-based machine-learning models of salt hydration properties",
      "evidence_urls": [
        "https://www.nature.com/articles/s43246-026-01200-2"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher full text and article metadata reviewed for stated result, AI contribution, date and affiliation."
    }
  },
  {
    "id": "2024-10-ml-guided-amidase-enantioselectivity",
    "title": "Machine learning-guided amidase engineering improves enantioselectivity",
    "summary": "Random-forest models using chemistry and geometry descriptors were trained on 240 prior amidase datasets and used to guide substrate and variant selection. The primary article reports that variant I198F/W328F achieved an enantioselectivity E-value of 303.5, compared with 5.7 for wild type, for the reported substrate assay. Most experimental substrate predictions agreed, with identified exceptions. This bounded laboratory result does not establish broad performance across substrates or industrial application.",
    "field": "biology",
    "ai_technology": "Random-forest classification models and descriptor-guided variant design",
    "organization": "Chinese Academy of Sciences; Beijing Normal University; University of Chinese Academy of Sciences",
    "source_url": "https://www.nature.com/articles/s41467-024-53048-0",
    "source_type": "paper",
    "publication_date": "2024-10-09",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary paper reports the stated assay results.",
      "unknown": "Independent replication and exhaustive prior-art review not established.",
      "ai_role": "Random-forest classification models and descriptor-guided variant design",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-53048-0"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher full text and bibliographic metadata reviewed for identity, AI role, dates, affiliations and bounded results."
    }
  },
  {
    "id": "2025-11-ai-screened-polydioxanone-packaging",
    "title": "AI screening identifies poly-PDO for recyclable barrier packaging",
    "summary": "Machine-learning models screened approximately 7.4 million virtual ring-opening polymerization polymers for properties relevant to recyclable food packaging. Researchers experimentally evaluated existing poly(p-dioxanone), whose barrier performance had not previously been reported in the reviewed source, and reported water-barrier properties and 95% monomer recovery. The result is a specific material characterization and packaging candidate, not a newly invented polymer or demonstrated commercial package. The arXiv v1 already reports the measured outcome; later journal publication is the same study.",
    "field": "materials",
    "ai_technology": "Polymer informatics models predicting material and recyclability properties",
    "organization": "Georgia Institute of Technology",
    "source_url": "https://arxiv.org/abs/2511.04704",
    "source_type": "paper",
    "publication_date": "2025-11-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary paper reports the stated assay results.",
      "unknown": "Independent replication and exhaustive prior-art review not established.",
      "ai_role": "Polymer informatics models predicting material and recyclability properties",
      "evidence_urls": [
        "https://arxiv.org/abs/2511.04704"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher full text and bibliographic metadata reviewed for identity, AI role, dates, affiliations and bounded results."
    }
  },
  {
    "id": "2024-05-ml-designed-dual-glp1-gcgr-agonists",
    "title": "Machine-learning-designed peptides show dual GCGR and GLP-1R activity",
    "summary": "An ensemble of multitask convolutional neural networks trained on 125 peptide analogues guided optimization of 15 synthesized variants. The study reports three potent dual agonists; P3 improved potency at both glucagon and GLP-1 receptors compared with the best dual agonist in the training set. These are cell-based receptor assays only. The paper reports weak performance for designing GCGR-selective peptides and establishes no clinical safety or therapeutic efficacy.",
    "field": "biology",
    "ai_technology": "Ensemble multitask convolutional neural networks for peptide sequence optimization",
    "organization": "AstraZeneca; University of Cambridge; Google DeepMind",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11374683/",
    "source_type": "paper",
    "publication_date": "2024-05-16",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary paper reports the stated assay results.",
      "unknown": "Independent replication and exhaustive prior-art review not established.",
      "ai_role": "Ensemble multitask convolutional neural networks for peptide sequence optimization",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11374683/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher/PMC article and journal metadata reviewed; author affiliation confirmed from publisher article PDF. The PMC full-text open displayed a reCAPTCHA."
    }
  },
  {
    "id": "2024-05-hydrogelfinder-machine-learning-peptides",
    "title": "HydrogelFinder: A Foundation Model for Efficient Self-Assembling Peptide Discovery Guided by Non-Peptidal Small Molecules",
    "summary": "HydrogelFinder used a generative model and learned prediction to prioritize 111 candidate peptides; 17 were synthesized and nine self-assembled under tested conditions. Cell assays covered two selected materials and do not establish general safety or clinical suitability.",
    "field": "biology",
    "ai_technology": "HydrogelFinder generative language model and learned gelation prediction",
    "organization": "Hunan University; Zhejiang University; King Abdullah University of Science and Technology; Central South University; Shandong University",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38704695/",
    "source_type": "paper",
    "publication_date": "2024-05-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary abstract reports 111 candidates, 17 synthesized and nine experimentally self-assembling; cell compatibility tests were limited to two selected materials.",
      "unknown": "Independent replication and exhaustive prior-art review are not established.",
      "ai_role": "HydrogelFinder generative language model and learned gelation prediction",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38704695/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for study identity, AI operation, dates, affiliations and the bounded result."
    }
  },
  {
    "id": "2024-07-machine-learning-guided-peptide-drug",
    "title": "Machine-Learning-Guided Peptide Drug Discovery: Development of GLP-1 Receptor Agonists with Improved Drug Properties",
    "summary": "A machine-learning-guided peptide optimization workflow selected GUB021794, reported at 0.018 nM GLP-1R potency in the paper assay and dose-dependent weight loss in diet-induced obese mice. These preclinical results do not establish human efficacy, safety or a dosing regimen.",
    "field": "biology",
    "ai_technology": "streaMLine machine-learning QSAR analysis integrated with peptide synthesis and screening",
    "organization": "Gubra",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/38977267/",
    "source_type": "paper",
    "publication_date": "2024-07-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary abstract reports the named candidate and mouse outcomes; no human clinical result is established.",
      "unknown": "Independent replication and exhaustive prior-art review are not established.",
      "ai_role": "streaMLine machine-learning QSAR analysis integrated with peptide synthesis and screening",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38977267/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for study identity, AI operation, dates, affiliations and the bounded result. PubMed affiliation block names Gubra."
    }
  },
  {
    "id": "2025-04-active-learning-oxygen-evolution-catalysts",
    "title": "Leveraging Data Mining, Active Learning, and Domain Adaptation for Efficient Discovery of Advanced Oxygen Evolution Electrocatalysts",
    "summary": "A multistage ML and active-learning workflow guided experimental catalyst batches. The journal article reports the best measured overpotential at 10 mA/cm² falling from 209 to 154 mV between batches one and five for sample B; stability rankings differ and mechanism calculations are not independent experimental proof.",
    "field": "physics",
    "ai_technology": "Supervised/unsupervised materials models, active learning and domain adaptation",
    "organization": "University of Chicago; North China Electric Power University; Nanjing University; The Hong Kong University of Science and Technology",
    "source_url": "https://knowledge.uchicago.edu/records/424pk-bp381",
    "source_type": "paper",
    "publication_date": "2025-04-04",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary journal record and full-text link identify the reported campaign and experimental catalyst result; arXiv v1 submitted 2024-07-05 already names the Ru–Mn–Ca–Pr catalyst, so journal date is not first disclosure.",
      "unknown": "Independent replication and exhaustive prior-art review are not established.",
      "ai_role": "Supervised/unsupervised materials models, active learning and domain adaptation",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/40184453/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for study identity, AI operation, dates, affiliations and the bounded result."
    }
  },
  {
    "id": "2026-01-iterative-discovery-of-potent-polym",
    "title": "Iterative discovery of potent polymeric antibiotics via multi-stage and multi-task learning against antimicrobial resistance",
    "summary": "PolyCLOVER integrated multi-stage learning, active learning and high-throughput experiments across a ~100,000-member poly(β-amino ester) space, identifying three lead nanoparticles with measured antibacterial activity and in-vivo mouse outcomes. Mouse groups were small; this does not establish human efficacy or safety.",
    "field": "biology",
    "ai_technology": "Multi-stage self-supervised learning and active learning coupled to high-throughput experimentation",
    "organization": "Zhejiang University; Sir Run Run Shaw Hospital, Zhejiang University School of Medicine",
    "source_url": "https://www.nature.com/articles/s41467-026-68682-z",
    "source_type": "paper",
    "publication_date": "2026-01-21",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary Nature abstract/full text reports experimental MICs of 4 and 8 μg/mL against named resistant bacteria and mouse infection models; this is one study, not three discoveries.",
      "unknown": "Independent replication and exhaustive prior-art review are not established.",
      "ai_role": "Multi-stage self-supervised learning and active learning coupled to high-throughput experimentation",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-68682-z"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for study identity, stated AI contribution, date and bounded result."
    }
  },
  {
    "id": "2022-10-ligand-based-discovery-of-coronavir",
    "title": "Ligand-based discovery of coronavirus main protease inhibitors using MACAW molecular embeddings",
    "summary": "MACAW molecular embeddings and ligand-based models prioritized purchased compounds for assays. The paper reports inhibition of recombinant coronavirus main protease for eight of ten selected molecules; compound 7 also reduced hCoV-229E infectivity in a cell assay. This is not a clinical COVID-19 result or proof of SARS-CoV-2 patient benefit.",
    "field": "biology",
    "ai_technology": "MACAW molecular embeddings used in ligand-based machine learning virtual screening",
    "organization": "Central South University; Université de Lorraine; University Hospital of Nancy; University of California, Santa Cruz",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9621234/",
    "source_type": "paper",
    "publication_date": "2022-10-28",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary article reports 8/10 selected compounds inhibiting recombinant Mpro and follow-up cellular testing of compound 7 against HCoV-229E.",
      "unknown": "Independent replication and exhaustive prior-art review are not established.",
      "ai_role": "MACAW molecular embeddings used in ligand-based machine learning virtual screening",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9621234/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for study identity, stated AI contribution, date and bounded result."
    }
  },
  {
    "id": "2023-03-alphaflow-rl-guided-cds-shell-growth",
    "title": "Reinforcement-learning-guided CdS shell-growth route using AlphaFlow",
    "summary": "AlphaFlow used reinforcement learning with an automated droplet reactor to select and optimize reagent sequences for CdS-shell growth on supplied CdSe quantum dots. Under the authors’ aged, diluted starting-material comparison, the selected route achieved a 26 nm larger absorption-peak shift and 450% higher photoluminescence intensity after six cycles than the conventional sequence; freshly diluted starting material showed a different comparison. This is a lab route to a known nanomaterial, not discovery of a new compound or proof of universal superiority.",
    "field": "physics",
    "ai_technology": "Reinforcement-learning agent with neural reward regression and boosted-tree viability classification in an automated fluidic lab",
    "organization": "North Carolina State University; University at Buffalo",
    "source_url": "https://www.nature.com/articles/s41467-023-37139-y",
    "source_type": "paper",
    "publication_date": "2023-03-14",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary Nature Communications article reports a six-cycle experimental comparison on aged diluted starting material; conditions and starting materials constrain result.",
      "unknown": "Earlier route disclosures and independent replication are not established.",
      "ai_role": "Reinforcement-learning-guided reaction sequence selection and optimization linked to measured shell-growth signals.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-023-37139-y"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary publisher full text checked for identity, dates, organization affiliations, AI method and bounded experimental comparison."
    }
  },
  {
    "id": "2026-04-al-mg-zn-transfer-learning-alloys",
    "title": "CALPHAD-based transfer learning designs high-strength Al–Mg–Zn alloys",
    "summary": "A CALPHAD-informed transfer-learning and multi-objective optimization workflow selected two laboratory alloys. The primary abstract reports tensile strengths of 472 ± 7 MPa and 569 ± 12 MPa, with elongations of 23.5 ± 0.5% and 14.9 ± 0.3%. These are measured alloy properties, not industrial qualification.",
    "field": "materials",
    "ai_technology": "CALPHAD-informed transfer learning; ensemble regression; NSGA-II",
    "organization": "Central South University; Changsha University of Science and Technology; China Three Gorges University",
    "source_url": "https://www.nature.com/articles/s41524-026-02073-2",
    "source_type": "paper",
    "publication_date": "2026-04-10",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broad deployment are not established by this review.",
      "ai_role": "A thermodynamics-informed representation transfers knowledge from established alloy systems; ML predictions and NSGA-II guide experiments, and selected alloys are synthesized and tested.",
      "evidence_urls": [
        "https://doi.org/10.1038/s41524-026-02073-2"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks."
    }
  },
  {
    "id": "2025-04-tbir3-puni3-structure",
    "title": "Machine-learning-guided synthesis identifies a TbIr3 structural phase",
    "summary": "Unsupervised clustering and a recommendation workflow guided synthesis and characterization of TbIr3 in a PuNi3-type structure. The authors report two syntheses and structural refinement; the source describes secondary phases, so this is not a phase-pure material claim. The formula was known in other structures; the result is the structure-type assignment.",
    "field": "materials",
    "ai_technology": "PCA; K-means clustering; recommendation engine",
    "organization": "Indian Institute of Science Education and Research Thiruvananthapuram",
    "source_url": "https://doi.org/10.1021/jacs.5c03510",
    "source_type": "paper",
    "publication_date": "2025-04-18",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broad deployment are not established by this review. The exact earlier preprint posting day was not established.",
      "ai_role": "Unsupervised structure-type clustering and recommendation nominate TbIr3 for synthesis; diffraction and compositional analysis test the structure assignment.",
      "evidence_urls": [
        "https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/67bc9edbfa469535b9bd0d65/original/TbIr3_final.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks."
    }
  },
  {
    "id": "2025-01-orgmol-immersion-coolants",
    "title": "Org-Mol prioritizes two experimental immersion-coolant candidates",
    "summary": "A 3D transformer pretrained on molecular structures and fine-tuned for physical-property prediction screened millions of ester candidates. Researchers synthesized and characterized two selected candidates. One did not meet the preferred thermal-conductivity threshold. This is experimental candidate evaluation, not industrial deployment or environmental-safety validation.",
    "field": "chemistry",
    "ai_technology": "Org-Mol 3D molecular transformer",
    "organization": "Chinese Academy of Sciences",
    "source_url": "https://arxiv.org/abs/2501.09896",
    "source_type": "paper",
    "publication_date": "2025-01-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The primary source reports the study-specific result summarized here.",
      "unknown": "Independent replication and broad deployment are not established by this review.",
      "ai_role": "A pretrained molecular transformer predicts bulk properties to prioritize candidate esters; researchers assess feasibility, synthesize and measure two candidates.",
      "evidence_urls": [
        "https://arxiv.org/abs/2501.09896"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks."
    }
  },
  {
    "id": "2024-06-ml-rational-lysine-dioxygenase-pentamutant",
    "title": "Machine-learning-guided lysine dioxygenase variant improves turnover",
    "summary": "A structure-based machine-learning mutation-ranking method, combined with molecular dynamics and human rational design, guided variant selection for a lysine dioxygenase. The authors report a five-substitution variant with total turnover number 218 ± 3, compared with 160 ± 2 for wild type in the reported assay. The improvement is specific to this enzyme and assay, not a claim of industrial or therapeutic performance. The source uses conflicting variant abbreviations: its abstract says LPNYI and its Results say LPYNI.",
    "field": "biology",
    "ai_technology": "MutComputeX; molecular dynamics",
    "organization": "University of Minnesota Twin Cities; University of Texas at Austin",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.06.04.597480v1",
    "source_type": "paper",
    "publication_date": "2024-06-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/cbic.202400495",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Primary abstract reports the designed pentamutant’s TTN and wild-type comparator.",
      "unknown": "Independent replication and broader application are not established by this bounded review. The source uses conflicting abbreviations for the five-substitution variant.",
      "ai_role": "Structure-based ML prioritized amino-acid substitutions; MD and rational selection also contributed to the final combined variant.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/38895203/",
        "https://www.biorxiv.org/content/10.1101/2024.06.04.597480v1"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary preprint version and journal methods/results; the variant abbreviation differs between abstract and Results."
    }
  },
  {
    "id": "2026-07-unikip-sulfotransferase-efst1a1",
    "title": "UniKP prioritizes a sulfotransferase for sulfotyrosine production",
    "summary": "UniKP was used to rank sulfotransferase homologues by predicted kinetics. The study reports experimental testing of five prioritized candidates and identifies EfST1A1, with about 6.5-fold higher in-vitro activity than the initial hit under the reported assay. Later directed evolution is not attributed to the model. This is a bounded enzyme-discovery result, not proof of industrial production or broad enzyme generality.",
    "field": "biology",
    "ai_technology": "UniKP kinetic prediction",
    "organization": "Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences",
    "source_url": "https://www.sciencedirect.com/science/article/abs/pii/S1385894726047108",
    "source_type": "paper",
    "publication_date": "2026-07-15",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.cej.2026.177249",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Publisher abstract/indexed article text reports five tested AI-ranked enzymes and identifies EfST1A1; institute research announcement summarizes experimental activity and later evolution.",
      "unknown": "Independent replication and broader application are not established by this bounded review.",
      "ai_role": "UniKP ranks enzyme homologues by predicted kinetic properties; researchers tested top candidates and separately applied directed evolution to the selected hit.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/abs/pii/S1385894726047108"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Publisher article preview/indexed text and institute source; full publisher text inaccessible. Issue date is recorded; exact online-first and earlier version not established."
    }
  },
  {
    "id": "2026-07-aaepre-xylanase-engineering",
    "title": "AI-guided mining identifies and engineers an alkaline xylanase",
    "summary": "A fine-tuned sequence model and generative workflow guided identification of xylanase 8E20 and mutant 8E20-178. The primary abstract reports an optimum-pH shift from 8 to 10 and a 1.9-fold improvement, but calls the metric catalytic activity while indexed conclusion text calls it catalytic efficiency; this draft preserves the ambiguity. It does not infer unreported assay conditions or industrial-scale performance.",
    "field": "biology",
    "ai_technology": "AAEPre sequence model; xynODM generative model",
    "organization": "Institute of Animal Science, Chinese Academy of Agricultural Sciences",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/42398548/",
    "source_type": "paper",
    "publication_date": "2026-07-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary abstract reports 8E20-178, pH optimum shift and 1.9-fold improvement.",
      "unknown": "Publisher full text inaccessible; exact improvement metric and assay conditions unresolved. Prior sequence characterization and independent replication not assessed.",
      "ai_role": "Sequence models and composite selection index prioritized candidate enzyme; generative model guided variant engineering.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/42398548/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated."
    }
  },
  {
    "id": "2025-06-multiscale-topological-lithium-conductors",
    "title": "Topological learning and simulations predict lithium superionic-conductor candidates",
    "summary": "A multiscale topological-learning workflow combined algebraic-topology descriptors, unsupervised clustering and ab initio molecular-dynamics simulations to screen lithium-containing structures. The study reports 14 candidate conductors meeting computational thresholds for diffusion barriers and electrochemical stability. The reported conductivities are simulated, with ambient values estimated from elevated-temperature trajectories. Some candidates had prior experimental reports or patents cited by the authors; those are not experimental validation produced by this AI workflow. Many candidates remain untested.",
    "field": "materials",
    "ai_technology": "Multiscale topological learning; persistent homology; affinity propagation clustering; ab initio molecular dynamics",
    "organization": "Peking University; Michigan State University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12186519/",
    "source_type": "paper",
    "publication_date": "2025-06-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary PMC full text and publication metadata; article identity, methods, bounded computational outcomes and affiliation block."
    },
    "paper_url": "https://doi.org/10.1021/jacs.5c04828",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study used ab initio molecular-dynamics simulations to estimate conductivity, diffusion barriers and electrochemical stability for 45 selected candidate materials and reports 14 meeting its computational thresholds.",
      "unknown": "Conductivities are computational estimates, including Arrhenius extrapolation from elevated-temperature simulations; source-cited experimental reports for some materials predate or are independent of this AI workflow, and other candidates remain untested.",
      "ai_role": "Topological feature extraction and unsupervised learning screened structures and prioritized candidate clusters; AIMD simulations evaluated specific candidate transport claims.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12186519/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-06-ml-genetic-refractory-alloy-design",
    "title": "Machine learning and genetic search predict stronger refractory alloys",
    "summary": "A random-forest model trained on published refractory high-entropy-alloy data was coupled to a stochastic genetic algorithm to optimize alloy compositions for yield strength at selected temperatures. For an AlMo0.5NbTa0.5TiZr base composition, the authors report modeled gains of 90 MPa at 25 °C and 13% at 1,000 °C for optimized compositions. Other base-alloy optimizations are also reported. The proposed alloys were not experimentally fabricated or mechanically measured in this study; these are computational design predictions.",
    "field": "materials",
    "ai_technology": "Random-forest regression; stochastic genetic algorithm",
    "organization": "CFD Research Corporation; University at Buffalo",
    "source_url": "https://doi.org/10.21203/rs.3.rs-1666028/v1",
    "source_type": "paper",
    "publication_date": "2022-06-13",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-022-00926-0",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The model was validated against published experimental yield-strength data, including previously unseen datasets; the proposed optimized compositions and their target-temperature improvements were computational predictions.",
      "unknown": "The proposed optimized alloys were not experimentally fabricated or measured in this study; predicted yield strength depends on the source dataset and model assumptions.",
      "ai_role": "A random-forest model predicted yield strength and a stochastic genetic algorithm searched compositions to maximize strength at specified temperatures.",
      "evidence_urls": [
        "https://assets-eu.researchsquare.com/files/rs-1666028/v1_covered.pdf"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-16",
      "scope": "Research Square primary v1 full text and posted-date metadata"
    }
  },
  {
    "id": "2022-10-ml-predicted-bioactive-peptide-nhpd50",
    "title": "Mass spectrometry and machine learning identify bioactive peptide NHPD-50",
    "summary": "A predicted-peptide-variant model trained using mass-spectrometry observations from seven mouse tissues helped prioritize candidate peptides. The authors report that NHPD-50, a secretogranin-1-derived peptide, lowered blood glucose and increased plasma insulin in specified diabetic-mouse experiments. The authors state that the data do not establish the peptide’s exact mechanism. These are early preclinical findings, not a human diabetes treatment; other predicted peptides had different or absent outcomes.",
    "field": "biology",
    "ai_technology": "Predicted-peptide-variant machine-learning prioritization from mass-spectrometry data",
    "organization": "Novo Nordisk A/S",
    "source_url": "https://www.nature.com/articles/s41467-022-34031-z",
    "source_type": "paper",
    "publication_date": "2022-10-20",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-16",
      "scope": "Primary Nature full article and bibliographic metadata reviewed for identity, AI role, date, affiliation and bounded mouse results."
    },
    "paper_url": "https://doi.org/10.1038/s41467-022-34031-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Primary paper reports the stated mouse assay results.",
      "unknown": "Independent replication and exhaustive prior-art review are not established; the paper states that the mechanism is undetermined.",
      "ai_role": "A predicted-peptide-variant model prioritized candidate sequences from mass-spectrometry data; biological activity was established through subsequent assays.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-022-34031-z"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-nonequilibrium-chemical-short-range-order-metallic-alloys",
    "title": "Machine learning simulations reveal nonequilibrium short-range order in metallic alloys",
    "summary": "A moment-tensor machine-learning interatomic potential enabled large-scale simulations of CrCoNi processing. The authors report that thermomechanical deformation and solidification can generate steady-state chemical short-range order outside the equilibrium-annealing spectrum, and propose an ordering-bias mechanism. This is a simulation-based computational finding, not direct experimental validation of the predicted states in manufactured alloys. The result and AI role were already present in arXiv v1; journal publication followed in 2025.",
    "field": "materials",
    "ai_technology": "Moment-tensor machine-learning interatomic potential; E(3)-equivariant graph neural network for local chemical-motif identification",
    "organization": "Massachusetts Institute of Technology",
    "source_url": "https://arxiv.org/pdf/2409.15474v1",
    "source_type": "paper",
    "publication_date": "2024-09-23",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-64733-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The primary v1 reports molecular-dynamics and nonequilibrium Monte Carlo simulations of CrCoNi thermomechanical deformation and solidification; these simulations yield steady-state SRO states outside the equilibrium-annealing spectrum.",
      "unknown": "This is a simulation-based result, not direct experimental confirmation of every predicted state or industrial deployment. Broad prior-art screening is bounded; independent reproduction is not established.",
      "ai_role": "A moment-tensor machine-learning interatomic potential (TS-f) supplies energies and forces for simulations; an E(3)-equivariant graph neural network separately identifies local motifs and supports SRO quantification.",
      "evidence_urls": [
        "https://arxiv.org/pdf/2409.15474v1",
        "https://arxiv.org/abs/2409.15474",
        "https://www.nature.com/articles/s41467-025-64733-z",
        "https://pubmed.ncbi.nlm.nih.gov/41062492/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 full text and submission history; indexed Nature Communications full text including Methods/contributions and affiliation block; PubMed metadata/abstract."
    }
  },
  {
    "id": "2026-06-mutexagpt-cold-adapted-amylase-variants",
    "title": "MutexaGPT-guided cold-adapted amylase variants",
    "summary": "The MutexaGPT multi-agent large language model system translated a cold-adaptation design request into molecular simulations and linker-variant proposals for bidomain amylase. The authors report experimentally validated variants with 1.7-fold and 3.7-fold activity enhancement at 0 °C. The preprint appeared on 19 June 2026; the peer-reviewed journal version followed on 10 September 2026.",
    "field": "biology",
    "ai_technology": "Multi-agent large language model system coordinating high-throughput molecular dynamics simulations and mutation design",
    "organization": "Vanderbilt University",
    "source_url": "https://doi.org/10.21203/rs.3.rs-10072540/v1",
    "source_type": "paper",
    "publication_date": "2026-06-19",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43588-026-01049-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The primary article reports two MutexaGPT-guided bidomain amylase linker variants with experimentally measured 1.7-fold and 3.7-fold activity enhancement at 0 °C. The Methods define relative activity using the 0 °C to 45 °C activity ratio.",
      "unknown": "The experimental result is specific to the tested bidomain amylase and variants; broader enzyme-family transfer and independent reproduction are not established.",
      "ai_role": "LLM agents clarify the design request, translate it into computational plans and mutation patterns, and explain simulation outputs. High-throughput molecular dynamics metrics guide the proposed variants; experimental assays test their activity.",
      "evidence_urls": [
        "https://assets-eu.researchsquare.com/files/rs-10072540/v1_covered_65be6e89-30da-488a-a7d7-0310d3a07d5f.pdf",
        "https://www.nature.com/articles/s43588-026-01049-y"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Research Square v1 full-text indexed content and metadata; primary Nature Computational Science indexed article text, including abstract, Methods, experimental result, date and affiliation block."
    }
  },
  {
    "id": "2025-01-evogradient-antimicrobial-peptides",
    "title": "EvoGradient evolves antimicrobial peptides active against resistant pathogens",
    "summary": "EvoGradient uses an explainable deep-learning potency model and virtual sequence evolution to optimize candidate antimicrobial peptides. Applied to peptides encoded by low-abundance human oral bacteria, it virtually evolved 32 candidates; the six most effective were synthesized and tested against multidrug-resistant pathogens. The authors report activity across carbapenem-resistant E. coli, K. pneumoniae and A. baumannii and vancomycin-resistant E. faecium; pep-19-mod reduced bacterial loads by more than 95% in mouse thigh-infection models under the reported systemic and local administrations. These are preclinical findings and do not establish human efficacy or safety.",
    "field": "biology",
    "ai_technology": "Explainable deep-learning antimicrobial-peptide potency model with gradient-based virtual sequence evolution",
    "organization": "Southeast University",
    "source_url": "https://www.nature.com/articles/s41564-024-01907-3",
    "source_type": "paper",
    "publication_date": "2025-01-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41564-024-01907-3",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Six synthesized peptides were tested against multidrug-resistant pathogens; the authors also report pep-19-mod in-vivo testing in mouse thigh-infection models, with over 95% bacterial-load reduction under the described systemic and local administration.",
      "unknown": "Mouse infection outcomes do not establish human safety, dosing, efficacy or clinical translation; only six candidates from the 32 virtual evolutions were synthesized and tested.",
      "ai_role": "EvoGradient predicts peptide potency and virtually modifies peptide sequences through gradient-based evolution; the workflow selected optimized candidates for synthesis and experimental testing.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41564-024-01907-3",
        "https://pubmed.ncbi.nlm.nih.gov/39825096/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Microbiology indexed article text and primary PubMed abstract/metadata/affiliations; conclusions limited to reported preclinical assays."
    }
  },
  {
    "id": "2022-01-hydramp-antimicrobial-peptide-design",
    "title": "HydrAMP generates antimicrobial peptide analogues validated in bacterial assays",
    "summary": "HydrAMP, a conditional variational autoencoder with antimicrobial activity classifiers, generated peptide analogues and de novo candidates. The 2022 preprint reports wet-lab confirmation of increased activity for a generated Pexiganan-based analogue; the 2023 journal version reports testing across five bacterial strains, including nine analogues of clinical prototypes and six analogues of an inactive peptide. The tested panel was small and model/human filtered; it does not establish therapeutic efficacy.",
    "field": "biology",
    "ai_technology": "Conditional variational autoencoder with classifier-guided peptide generation and molecular-dynamics-based candidate preselection",
    "organization": "University of Warsaw",
    "source_url": "https://www.biorxiv.org/content/10.1101/2022.01.27.478054v1",
    "source_type": "paper",
    "publication_date": "2022-01-28",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-36994-z",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Generated peptide analogues were experimentally evaluated for antimicrobial activity across five bacterial strains; the journal abstract reports high activity for nine generated analogues of clinical prototypes and six analogues of an inactive peptide.",
      "unknown": "The tested panel is selected and small; these are in-vitro assays and do not establish safety, therapeutic efficacy or clinical benefit. The 2023 correction replaced a duplicated generation-statistics figure, not the wet-lab result stated here.",
      "ai_role": "A conditional VAE generated peptide analogues and unconstrained candidates from antimicrobial activity conditions; ranking and molecular-dynamics preselection helped choose candidates for experimental validation.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.1101/2022.01.27.478054v1",
        "https://www.nature.com/articles/s41467-023-36994-z",
        "https://www.nature.com/articles/s41467-023-40879-6.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary bioRxiv v1 abstract/metadata; primary Nature Communications indexed full article including methods/results; primary publisher author correction."
    }
  },
  {
    "id": "2025-07-offline-rl-agbii4-film-microstructure",
    "title": "Offline reinforcement learning identifies an AgBiI4 film-processing condition",
    "summary": "The Daisy framework combines SEM-image interpretation with offline reinforcement learning trained on historical Ag-Bi-I thin-film conditions and microstructure measurements. For AgBiI4, the planner proposed a condition outside the recorded combinations, and a newly fabricated film showed no visible pinholes and mean grain size 14.5% above the historical mean for defect-free samples. Its grain distribution was comparable to the best historical sample, so the study does not show that the new condition outperformed the previous best or improved device performance.",
    "field": "materials",
    "ai_technology": "Daisy image-analysis models and offline reinforcement-learning synthesis planner trained on historical microstructure data",
    "organization": "University of Washington",
    "source_url": "https://www.nature.com/articles/s41524-025-01701-7",
    "source_type": "paper",
    "publication_date": "2025-07-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-025-01701-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "A newly fabricated AgBiI4 film under the RL-proposed processing condition showed 0% visible defect coverage and mean grain size 14.5% above the historical mean for defect-free samples; its grain distribution was comparable to the best historical sample.",
      "unknown": "This is a thin-film morphology result; the new condition did not outperform the best historical sample and device-level optoelectronic performance was not established.",
      "ai_role": "Daisy uses ML models to interpret SEM morphology and an offline reinforcement-learning planner trained on historical synthesis conditions to propose a processing condition that was then fabricated and measured.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01701-7"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher full text as indexed, including article date, methods, results, author contributions and affiliations."
    }
  },
  {
    "id": "2025-08-foundation-model-electrolyte-formulation",
    "title": "A chemical foundation model guides high-conductivity electrolyte formulation",
    "summary": "Researchers fine-tuned the SMI-TED chemical foundation model on 13,666 literature measurements of ionic conductivity and screened 100,000 generated electrolyte formulations. Sixteen selected formulations were measured experimentally, with seven exceeding 10 mS/cm. A model-generated LiFSI formulation measured 19.14 mS/cm at room temperature, 82% above the best literature value in the paper’s comparison; the model had predicted 10.52 mS/cm. Conductivity is only one electrolyte property and does not establish electrochemical stability, safety or battery performance. The article was published 2025-08-30; a later correction changed an author name only.",
    "field": "materials",
    "ai_technology": "SMI-TED chemical foundation model fine-tuned for electrolyte mixture ionic-conductivity prediction and generative screening",
    "organization": "IBM Research",
    "source_url": "https://www.nature.com/articles/s41524-025-01774-4",
    "source_type": "paper",
    "publication_date": "2025-08-30",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41524-025-01774-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Sixteen selected formulations were measured by triplicate EIS at 20 °C; seven exceeded 10 mS/cm. A generated LiFSI electrolyte measured 19.14 mS/cm, compared with a model prediction of 10.52 mS/cm and 82% higher conductivity than the best literature value used in the paper comparison.",
      "unknown": "High ionic conductivity alone does not establish electrochemical stability, compatibility with particular electrodes, safety, cycle life or commercial battery performance.",
      "ai_role": "SMI-TED was fine-tuned on literature ionic-conductivity data and screened 100,000 generated formulations; selected candidates were prepared and measured experimentally.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41524-025-01774-4",
        "https://www.nature.com/articles/s41524-025-01804-1"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher full text including methods, measured outcomes, affiliations and publication date; primary publisher correction."
    }
  },
  {
    "id": "2025-09-active-learning-anode-free-electrolytes",
    "title": "Active learning identifies high-performing electrolyte solvents for anode-free lithium metal cells",
    "summary": "Researchers used sequential Bayesian experimental design with Bayesian model averaging to select electrolyte solvents for testing in Cu||LiFePO4 cells, feeding cycle-capacity results back into the model over seven campaigns. The study reports seven top-performing solvent electrolytes and says four distinct solvents rival state-of-the-art performance; three were previously unreported as electrolytes for anode-free cells. Five of nine compounds tested in the final batch exceeded 100 mAh/g discharge capacity at cycle 20. The result is limited to the tested cell configuration and cycling conditions and does not establish commercial battery performance.",
    "field": "materials",
    "ai_technology": "Active learning with sequential Bayesian experimental design, Bayesian model averaging and Gaussian-process surrogate models",
    "organization": "University of Chicago",
    "source_url": "https://www.nature.com/articles/s41467-025-63303-7",
    "source_type": "paper",
    "publication_date": "2025-09-25",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-63303-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Electrolytes were tested iteratively in Cu||LiFePO4 cells. The study reports seven top-performing solvents; three had not previously been reported as anode-free battery electrolytes, and four distinct solvents rivaled state-of-the-art performance. Five of nine batch-7 electrolytes exceeded 100 mAh/g discharge capacity at cycle 20.",
      "unknown": "Performance is limited to the reported cell chemistry, 1 M LiFSA configuration and cycling protocol; it does not establish full-cell energy density, lifetime in commercial formats or manufacturing readiness.",
      "ai_role": "A sequential active-learning framework selected each batch of candidate solvents using Bayesian experimental design and Bayesian model averaging, then incorporated measured electrochemical outcomes to update its next selections.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-63303-7",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12462480/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Communications full text as indexed; primary PMC text with methods/results/contributions; institutional full-paper manuscript mirror."
    }
  },
  {
    "id": "2023-09-photocat-guides-photocatalytic-reaction-discovery",
    "title": "PhotoCat guides experimental discovery of photocatalytic reactions",
    "summary": "The PhotoCat Transformer was pretrained on general reaction data and fine-tuned on the authors’ photocatalysis database to predict products and reaction conditions. In the 2023 ChemRxiv version, the authors report using it to guide experimental validation of previously unreported photocatalytic reactions. The later 2026 journal version describes four novel validated reactions, whereas the preprint reports five; this entry makes no exact-count claim and treats the journal article as a revised version of the same study.",
    "field": "chemistry",
    "ai_technology": "Transformer-based deep-learning reaction predictor pretrained on USPTO and fine-tuned on PhotoCatDB, with reaction-condition inputs",
    "organization": "Zhejiang University of Technology",
    "source_url": "https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/65004fd999918fe537d200ab/original/photo-cat-an-artificial-intelligence-driven-synthesis-planning-platform-for-photocatalysis.pdf",
    "source_type": "paper",
    "publication_date": "2023-09-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42004-026-01894-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The PhotoCat Transformer was pretrained on general reaction data and fine-tuned on a photocatalysis database to predict products and conditions. The 2023 ChemRxiv version reports wet-lab validation of five predicted reactions; the 2026 journal abstract reports four novel validated reactions.",
      "unknown": "The preprint and journal version differ in their reported count of validated novel reactions. This record states the shared model-guided experimental discovery claim and makes no exact-count claim.",
      "ai_role": "A Transformer reaction predictor was pretrained on USPTO and fine-tuned on PhotoCatDB to rank photocatalytic reaction outcomes and conditions, guiding experimental validation.",
      "evidence_urls": [
        "https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/65004fd999918fe537d200ab/original/photo-cat-an-artificial-intelligence-driven-synthesis-planning-platform-for-photocatalysis.pdf",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12913628/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ChemRxiv v2 full-text PDF and version-history metadata; primary PMC full-text journal version and publication metadata. The preprint and journal versions report different counts (five versus four); the summary preserves this discrepancy."
    }
  },
  {
    "id": "2025-02-ml-screened-molecular-co2-photocatalysis",
    "title": "Machine learning screening identifies an efficient molecular photocatalytic CO2-reduction system",
    "summary": "A machine-learning-accelerated descriptor workflow screened 3,444 molecular photosensitizer–catalyst systems, including 180,000 conformations, and selected six candidates for experimental validation. The primary article reports that the optimal tested system reached a turnover number of 4,390 and uses time-resolved spectroscopy and first-principles calculations to examine dipole coupling. The reported outcome is bounded to the tested system and the study’s screening scope.",
    "field": "chemistry",
    "ai_technology": "Machine-learning-accelerated high-throughput screening using descriptors for photosensitization, electron transfer, and catalysis",
    "organization": "University of Science and Technology of China",
    "source_url": "https://doi.org/10.1038/s41929-025-01291-z",
    "source_type": "paper",
    "publication_date": "2025-02-07",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article reports experimental validation of the machine-learning-screened molecular photocatalytic systems; its optimal system achieved a turnover number of 4,390. The study screened 3,444 systems and selected six promising candidates.",
      "unknown": "The summary does not identify a catalyst or photosensitizer structure because the accessible primary indexed passage did not unambiguously tie a named complex to the reported maximum. Performance is specific to the reported experimental conditions; deployment and independent replication are not established.",
      "ai_role": "Machine learning accelerates high-throughput screening with descriptors spanning photosensitization, electron transfer and catalysis, reducing the 3,444-system search to six candidates selected for experimental validation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41929-025-01291-z",
        "https://chem.ahnu.edu.cn/info/1028/28946.htm"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Catalysis indexed article text, abstract, extended-data descriptions, metadata and affiliation block; institutional research announcement checked as corroboration."
    }
  },
  {
    "id": "2024-01-ml-guided-synthesis-2d-agbi-iodide-perovskites",
    "title": "Machine learning-guided synthesis of two-dimensional AgBi iodide perovskites",
    "summary": "The authors combined an 80-amine high-throughput synthesis dataset, chemistry-informed molecular descriptors, subgroup discovery and a support-vector classifier to rank organic spacers for 2D AgBi iodide perovskites. Of 13 model-selected candidates taken forward, eight were successfully synthesized and characterized by single-crystal X-ray diffraction and powder X-ray diffraction. The study reports a 61.5% success rate in this selected validation set; it does not establish universal synthesis performance across other perovskite families.",
    "field": "materials",
    "ai_technology": "Subgroup discovery and support-vector classification using physicochemical and molecular-topology descriptors",
    "organization": "Southeast University",
    "source_url": "https://www.nature.com/articles/s41467-023-44236-5.pdf",
    "source_type": "paper",
    "publication_date": "2024-01-02",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-44236-5",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary article reports successful synthesis of 8 of 13 machine-learning-selected 2D AgBi iodide perovskites (61.5%), with crystal structures determined by single-crystal X-ray diffraction and bulk-phase purity confirmed by PXRD.",
      "unknown": "The validation concerns a selected set under common laboratory synthesis conditions and one AgBi iodide family; generalization to other material families is not demonstrated.",
      "ai_role": "Subgroup discovery identifies an applicable data region and a support-vector classifier estimates synthesis feasibility from molecular descriptors; the model ranks candidate organic spacers for experimental synthesis.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-023-44236-5.pdf",
        "https://kjc.seu.edu.cn/2024/0111/c29280a478760/page.htm"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Communications publisher PDF indexed full text including Results, Methods, author contributions and article metadata; Southeast University institutional announcement checked for corroboration."
    }
  },
  {
    "id": "2026-06-apex-prionin-antimicrobial-peptides",
    "title": "Deep learning identifies antimicrobial peptides hidden in prion-related proteins",
    "summary": "APEX 1.1 screened 19.3 million peptide fragments from 2,897 prion-related proteins and identified 1,179 candidate antimicrobial peptides, termed prionins. Among 75 synthesized candidates, 59 inhibited at least one tested bacterial strain; two selected peptides reduced Acinetobacter baumannii burden in a mouse skin-abscess model. The study shows that prion-related sequences can yield experimentally active antimicrobial fragments, but does not show that these peptides are naturally released or function as host-defense agents in their source organisms.",
    "field": "biology",
    "ai_technology": "APEX 1.1 deep-learning antimicrobial-peptide prediction and prioritization",
    "organization": "University of Pennsylvania",
    "source_url": "https://www.nature.com/articles/s41564-026-02408-1",
    "source_type": "paper",
    "publication_date": "2026-06-19",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41564-026-02408-1",
    "relevance": {
      "impact": 4,
      "validation": 3,
      "novelty": 4
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Of 75 synthesized APEX-selected prionins, 59 inhibited at least one bacterial strain in vitro; two selected peptides reduced A. baumannii burden in a mouse skin-abscess model.",
      "unknown": "No evidence shows these fragments are naturally released from their source proteins or physiologically function as host-defense peptides in native organisms. Mouse results do not establish human efficacy or safety.",
      "ai_role": "APEX 1.1 scored peptide fragments for predicted antimicrobial activity and prioritized candidates from a search across prion-related proteins for synthesis and experimental screening.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41564-026-02408-1"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Microbiology full text as indexed, including methods, results, author contributions and affiliations."
    }
  },
  {
    "id": "2026-02-ml-screen-identifies-gw3965-against-invasive-gas",
    "title": "Machine learning screening identifies GW3965 activity against invasive group A Streptococcus",
    "summary": "A neural network trained on measured growth responses from 2,560 compounds screened 6,111 FDA-evaluated compounds for activity against multidrug-resistant emm92 group A Streptococcus. Of nine selected compounds tested, GW3965 reduced bacterial survival by 99% at an MIC of 6.25 μM. The authors also report improved wound closure in an infected human skin-equivalent model and reduced lesion area and bacterial burden in mice. This is preclinical evidence and does not establish efficacy or safety in people. The peer-reviewed journal article followed on 2026-05-30 and reports the same core GW3965 result.",
    "field": "biology",
    "ai_technology": "Deep neural network trained on compound SMILES and measured bacterial growth responses; model screened an FDA-evaluated compound library",
    "organization": "West Virginia University",
    "source_url": "https://assets-eu.researchsquare.com/files/rs-8779898/v1/a183f808-a19a-41be-8b56-ad4009293888.pdf",
    "source_type": "paper",
    "publication_date": "2026-02-18",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s44259-026-00225-4",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The model was trained on 2,560 compounds and screened 6,111. Nine selected compounds were validated; GW3965 had MIC 6.25 μM and reduced viable emm92 iGAS by 99%. In a human skin-equivalent model infected tissues treated with GW3965 showed wound closure; in a mouse SSTI model 30 mg/kg reduced lesion size and day-6 wound bacterial burden by about 2 logs on average.",
      "unknown": "Results come from in-vitro assays and animal/tissue models; the preprint reports no synergy with ampicillin. Human clinical efficacy, dosing, safety and mechanism against GAS remain unestablished.",
      "ai_role": "A neural network learned structure–growth-inhibition patterns from measured compound screening data, prioritized GW3965 for follow-up, and the selected candidate was experimentally tested in bacterial, human skin-equivalent, and mouse infection models.",
      "evidence_urls": [
        "https://assets-eu.researchsquare.com/files/rs-8779898/v1/a183f808-a19a-41be-8b56-ad4009293888.pdf"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Research Square v1 full-text PDF, including posted date, affiliations, Methods and experimental outcomes; primary Nature publisher indexed article text and journal-version abstract/results."
    }
  },
  {
    "id": "2026-05-phase-net-reveals-retrospective-dingri-seismicity",
    "title": "AI-assisted analysis reveals retrospective seismicity before the Dingri earthquake",
    "summary": "A PhaseNet-assisted workflow combined phase detection, polarisation analysis and template matching on records from one permanent station to reconstruct seismicity in the two years before the 2025 Mw 7.1 Dingri earthquake. The authors report increased event rate and clustering beginning about 45 days before the mainshock. This retrospective pattern is interpreted as compatible with late-stage nucleation; the study does not establish a prospective earthquake forecast.",
    "field": "physics",
    "ai_technology": "PhaseNet deep neural-network seismic phase picker combined with polarisation analysis and template matching",
    "organization": "Istituto Nazionale di Oceanografia e di Geofisica Sperimentale (OGS)",
    "source_url": "https://www.nature.com/articles/s41598-026-49197-5",
    "source_type": "paper",
    "publication_date": "2026-05-11",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41598-026-49197-5",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "PhaseNet-assisted phase detection, polarisation analysis and template matching reconstructed 289 template events and 1,138 augmented events at one station before the 2025 Dingri earthquake. The authors report increased event rate and clustering beginning about 45 days before the mainshock.",
      "unknown": "This is a retrospective analysis from one station. The authors describe a pattern compatible with late-stage nucleation; it does not establish a prospective forecast or earthquake prediction.",
      "ai_role": "PhaseNet, a deep neural-network phase picker, supported detection and event-catalogue construction; the resulting catalogue enabled retrospective identification of the pre-mainshock activity increase.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41598-026-49197-5"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Scientific Reports indexed article, including abstract, Methods, Results, article dates and author-affiliation block."
    }
  },
  {
    "id": "2026-01-ml-catalogue-reveals-sierra-negra-lp-swarm",
    "title": "Machine-learning seismic catalogue reveals a long-period swarm before Sierra Negra eruption onset",
    "summary": "A machine-learning phase-picking and template-matching workflow expanded the catalogue for the 2018 Sierra Negra eruption onset, identifying a northwestern-caldera swarm of long-period microearthquakes missing from earlier manually picked records. The authors report the swarm began about two hours after a Mw 5.4 intra-caldera earthquake and preceded edifice failure and magma intrusion. The proposed fluid-triggering sequence is an interpretation of retrospective seismic and geodetic observations, not a prospective eruption forecast.",
    "field": "physics",
    "ai_technology": "Volpick EQTransformer-based seismic phase picker combined with template matching and geodetic analysis",
    "organization": "The Chinese University of Hong Kong",
    "source_url": "https://www.nature.com/articles/s41467-026-68645-4",
    "source_type": "paper",
    "publication_date": "2026-01-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-68645-4",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "An EQTransformer-based phase picker and template matching expanded the seismic catalogue for the 2018 Sierra Negra eruption onset. The authors report 1,224 volcano-tectonic and 711 long-period events and identify a long-period swarm missing from the earlier manually picked catalogue, beginning about two hours after the Mw 5.4 earthquake.",
      "unknown": "The proposed fluid-triggering sequence is an interpretation of retrospective seismic and geodetic observations; this study does not establish prospective eruption prediction.",
      "ai_role": "The learned EQTransformer-based phase picker supported a denser event catalogue; combined with template matching, this exposed a previously unreported long-period swarm in the archived seismic record.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-68645-4"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Communications indexed article text and PDF-indexed article content, including abstract, Results, Methods, metadata and author-affiliation section."
    }
  },
  {
    "id": "2025-05-rlxf-aligns-esm2-for-brighter-creilov-variants",
    "title": "Experimental feedback guides ESM-2 to brighter CreiLOV variants",
    "summary": "Reinforcement Learning from eXperimental Feedback (RLXF) used a learned fluorescence reward model to align ESM-2 with measured CreiLOV sequence-function data. In the peer-reviewed study, the authors expressed 60 generated variants in E. coli and measured cellular fluorescence; six of ten aligned ESM-2 five-mutant designs exceeded CreiLOV fluorescence, and three aligned designs exceeded CreiLOV-T7S. The first bioRxiv version (posted 2025-05-08) already described experimentally enhanced fluorescence from RLXF-aligned designs; detailed measurements here are from the 2026 journal version. The earlier ICLR workshop comparison was computational and did not report new wet-lab fluorescence assays.",
    "field": "biology",
    "ai_technology": "650-million-parameter ESM-2 protein language model aligned by reinforcement learning from experimental feedback using a learned sequence-fluorescence reward model",
    "organization": "University of Wisconsin-Madison",
    "source_url": "https://www.biorxiv.org/content/10.1101/2025.05.02.651993v1",
    "source_type": "paper",
    "publication_date": "2025-05-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-77557-2",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The journal article reports 60 CreiLOV designs expressed in E. coli BL21(DE3) and measured for total cellular fluorescence. Six of ten aligned ESM-2 five-mutant designs exceeded CreiLOV fluorescence, and three aligned designs exceeded the prior CreiLOV-T7S control; the brightest reported design was 1.2-fold above T7S.",
      "unknown": "The measurements use total cellular fluorescence and are specific to tested designs/conditions; broader cellular contexts, protein function beyond fluorescence, and independent reproduction are not established. The qualitative enhanced-fluorescence result is present in the 2025 preprint abstract; detailed design-set metrics are verified from the later journal version.",
      "ai_role": "A learned sequence-fluorescence reward model trained on measured CreiLOV variant data guides supervised fine-tuning and PPO alignment of ESM-2. The aligned model generates the candidate sequences that were expressed and assayed.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.1101/2025.05.02.651993v1",
        "https://www.nature.com/articles/s41467-026-77557-2",
        "https://openreview.net/pdf?id=yuiQMh52Bm"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary bioRxiv v1 abstract, posting-date and affiliation metadata; primary GEM@ICLR workshop paper full text for methods and experimental outcomes; primary Nature Communications indexed text for later journal-version comparison."
    }
  },
  {
    "id": "2024-03-metl-gfp-protein-engineering",
    "title": "A biophysics-informed protein language model designs functional GFP variants",
    "summary": "Mutational Effect Transfer Learning (METL) combines transformer pretraining on simulated protein biophysics with fine-tuning on experimental sequence–function data. Using only 64 experimental GFP examples, the authors designed 20 previously untested variants and measured their fluorescence; 16 showed measurable fluorescence, although none was brighter than wild-type GFP. The result demonstrates functional design under a small-data setup, not improved brightness over wild type. The result was present in the bioRxiv version posted 2024-03-17; the GEM@ICLR 2024 workshop paper provides the detailed 20-variant assay report, and the later journal version is the same study.",
    "field": "biology",
    "ai_technology": "Mutational Effect Transfer Learning (METL), a transformer protein language model pretrained on simulated biophysical features and fine-tuned on experimental data",
    "organization": "University of Wisconsin-Madison",
    "source_url": "https://www.biorxiv.org/content/10.1101/2024.03.15.585128v1",
    "source_type": "paper",
    "publication_date": "2024-03-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-025-02776-2",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Twenty METL-designed GFP variants were synthesized and experimentally measured; 16 showed measurable fluorescence, but none exceeded wild-type brightness.",
      "unknown": "The panel contains 20 designed variants in a specific GFP task; the result does not show improvement over wild-type brightness or establish general performance across protein families.",
      "ai_role": "METL-Local was pretrained on simulated GFP sequence/biophysical data, fine-tuned using 64 experimental examples, and used with sequence optimization to select variants for wet-lab testing.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.1101/2024.03.15.585128v1",
        "https://openreview.net/pdf?id=hWISqnHBr7"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary bioRxiv v1 indexed abstract, posting date and affiliation metadata; primary GEM@ICLR workshop paper full text with methods and experimental outcomes."
    }
  },
  {
    "id": "2024-03-mutcomputex-engineered-nb4omt",
    "title": "Machine learning guides engineering of an alkaloid-biosynthesis enzyme",
    "summary": "Researchers developed MutComputeX, a structure-based residual neural network for assessing amino-acid substitutions in protein–ligand environments, and used it with a 4′-O-methylnorbelladine biosensor to engineer the plant methyltransferase Nb4OMT in Escherichia coli. The paper reports engineered variants with 60% higher product titer, twofold higher catalytic activity, threefold lower off-product regioisomer formation and 5 °C higher thermostability; ligand-aware predictions helped identify the beneficial A53M substitution. These enzyme and product results do not demonstrate complete microbial production of galantamine.",
    "field": "biology",
    "ai_technology": "MutComputeX structure-based residual neural network for protein–ligand mutation design",
    "organization": "University of Texas at Austin",
    "source_url": "https://www.nature.com/articles/s41467-024-46356-y",
    "source_type": "paper",
    "publication_date": "2024-03-07",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-46356-y",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "MutComputeX-guided, manually curated Nb4OMT variants were assayed using a 4NB biosensor and biochemical methods. The article reports 60% higher product titer, twofold higher catalytic activity, threefold lower off-product formation and 5 °C higher thermostability for engineered variants.",
      "unknown": "The result concerns a pathway enzyme and branchpoint intermediate in the reported microbial/biochemical systems; full pathway production of galantamine and industrial manufacturing were not demonstrated.",
      "ai_role": "MutComputeX predicted sequence substitutions in a modeled protein–ligand microenvironment and guided selection of Nb4OMT variants for biosensor screening and enzyme assays; ligand-aware predictions supported A53M.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-46356-y"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Communications full text as indexed, including methods/results, author contributions and affiliation block."
    }
  },
  {
    "id": "2024-11-similarity-informed-machine-learning-high-entropy-alloy-oxygen-reduction-catalyst",
    "title": "Similarity-informed machine learning guides experimental high-entropy alloy catalyst discovery",
    "summary": "A graph neural network augmented with quantum-inspired adsorption-site similarity reduced the DFT training data needed for a given prediction accuracy from 1,600 to 800 calculations. The model identified a five-metal Fe–Co–Ni–Ir–Ru high-entropy alloy candidate for oxygen reduction; the author group reports experimental validation and performance above a Pt/C comparison. The publisher abstract gives a predicted overpotential of 0.24 V, which is not presented here as the measured value. This is a laboratory catalyst result, not a fuel-cell deployment claim.",
    "field": "materials",
    "ai_technology": "Similarity-augmented graph neural network for adsorption-site and alloy-composition screening",
    "organization": "Computational Laboratory for Energy and Nanoscience",
    "source_url": "https://clean.energyscience.ca/publications/078.html",
    "source_type": "other",
    "publication_date": "2024-11-06",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.matt.2024.10.001",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ScienceDirect indexed abstract and metadata; primary author-group publication page. Direct publisher full-text open was blocked, so no full Methods or complete article affiliation review is claimed."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The author group reports experimental validation identifying a five-metal high-entropy alloy oxygen-reduction catalyst that outperformed its Pt/C comparison. The publisher describes experimental study of four additional HEAs; the 0.24 V overpotential is a model prediction, not stated here as an experimental measurement.",
      "unknown": "The publisher full Methods and supplementary results were not reviewed in this pass. The abstract-level material does not provide the measured ORR performance values, test uncertainty, durability, or full experimental protocol. No independent replication or fuel-cell device performance is established.",
      "ai_role": "The graph neural network incorporated quantum-inspired similarity between adsorption sites, reducing the required DFT training-set size from 1,600 to 800 while maintaining prediction accuracy. Its predictions nominated HEA compositions for experimental synthesis and ORR evaluation.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S2590238524005289",
        "https://clean.energyscience.ca/publications/078.html"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-08-hydramp-tryptophan-rich-antimicrobial-peptides",
    "title": "HydrAMP-guided design of short tryptophan-rich antimicrobial peptides",
    "summary": "HydrAMP generated analogues from a short Arg/Trp prototype, with AMPlify scoring and empirical filters used to select candidates. The authors synthesized three peptides and measured antimicrobial activity across bacterial strains; AMP3 showed the broadest activity among those tested, including MIC50 values of 15 μM against E. coli and 125 μM against P. aeruginosa. This is a small, model-assisted peptide panel built on a known sequence motif, not evidence of therapeutic efficacy.",
    "field": "biology",
    "ai_technology": "HydrAMP deep-learning peptide sequence generator and AMPlify deep-learning antimicrobial-activity scorer",
    "organization": "University of Napoli Federico II",
    "source_url": "https://chemistry-europe.onlinelibrary.wiley.com/doi/pdfdirect/10.1002/chem.202501918",
    "source_type": "paper",
    "publication_date": "2025-08-13",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/chem.202501918",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Three peptides were synthesized and tested in vitro. The lead AMP3 had reported MIC50 values of 15 μM against E. coli and 125 μM against P. aeruginosa, with broader spectrum than AMP1/AMP2 in the tested panel.",
      "unknown": "The design starts from a known short Arg/Trp peptide motif and tests a very small selected set; clinical efficacy, safety, and broad-spectrum performance beyond the strains tested are unknown. The accessible source comparison establishes it is not the Pexiganan/Temporin-A wet-lab result reported in the earlier HydrAMP paper, but does not assert exhaustive sequence-prior-art completeness.",
      "ai_role": "HydrAMP generated candidate analogues from the AMP1 prototype; the AMPlify learned model scored antimicrobial propensity, followed by empirical candidate filtering and wet-lab testing.",
      "evidence_urls": [
        "https://chemistry-europe.onlinelibrary.wiley.com/doi/pdfdirect/10.1002/chem.202501918",
        "https://pubmed.ncbi.nlm.nih.gov/40801152/",
        "https://www.nature.com/articles/s41467-023-36994-z"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Wiley publisher full-text indexed article including sequences, Methods, MIC results and publication date; PubMed metadata/abstract/affiliation; separate primary Nature Communications article reviewed to compare the prior HydrAMP wet-lab prototypes and results."
    }
  },
  {
    "id": "2026-01-unimol-screening-beta-sheet-peptide-assembly-codes",
    "title": "Uni-Mol screening guides experimental validation of β-sheet tetrapeptide assemblies",
    "summary": "A fine-tuned Uni-Mol model used simulation-derived data to screen 160,000 tetrapeptides. Researchers experimentally tested 26 selected tetrapeptide candidates: 15 formed β-sheet assemblies, including seven hydrogelators, as assessed with structural and assembly assays. The study reports experimental validation across 46 candidates overall; the 15/26 result is specific to the screened tetrapeptide set. This is a materials-science preprint, not evidence of medical function or clinical use. A related 2023 study explored tetrapeptide hydrogelation; this record makes no claim that every sequence is unprecedented and is limited to the present paper’s β-sheet structural validation.",
    "field": "materials",
    "ai_technology": "Fine-tuned Uni-Mol machine-learning model for Amyloid-Like Tendency prediction, guided by hybrid-resolution molecular-dynamics simulations",
    "organization": "Peking University Shenzhen Graduate School",
    "source_url": "https://www.cambridge.org/engage/chemrxiv/article-details/694d4b3be3cb457e13d19e29",
    "source_type": "paper",
    "publication_date": "2026-01-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.26434/chemrxiv-2026-4c5h0",
    "source_review": {
      "date": "2026-11-17",
      "scope": "ChemRxiv v1 page, full manuscript and Supplementary Information; official bioRxiv API metadata/abstract for cited 2025 computational preprint; primary 2023 hydrogel paper and linked supplementary-data index. Full-text source limits and the bounded result are stated."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "In the Uni-Mol-screened tetrapeptide set, 26 candidates were experimentally characterized; 15 formed β-sheet assemblies and seven were reported as hydrogelators. The authors also report validation of 46 candidates across the study overall.",
      "unknown": "The 46-candidate set combines tripeptide and tetrapeptide validations; the 15/26 result is specific to the tetrapeptides. The work is a preprint and establishes laboratory assembly outcomes only, not biomedical function, clinical utility, or independent replication. No claim is made that every sequence is unprecedented.",
      "ai_role": "Uni-Mol was fine-tuned on simulation-derived data and used to screen 160,000 tetrapeptides for Amyloid-Like Tendency; predictions guided candidate selection for further simulation and experimental structural characterization.",
      "evidence_urls": [
        "https://www.cambridge.org/engage/chemrxiv/article-details/694d4b3be3cb457e13d19e29",
        "https://www.cambridge.org/engage/api-gateway/coe/assets/orp/resource/item/694d4b3be3cb457e13d19e29/original/discovery-of-sheet-peptide-assembly-codes-via-an-experimentally-validated-predictive-computational-platform.pdf",
        "https://s3.eu-west-1.amazonaws.com/assets.prod.orp.cambridge.org/02/54654caa1745248e0618e2e4ca692a.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-10-gpr-discovers-selective-aurein-derived-peptides",
    "title": "Iterative Gaussian process screening discovers selective aurein-derived antifungal peptides",
    "summary": "An uncertainty-aware Gaussian process regression workflow iteratively selected 23 synthetic α/β-peptide analogues for synthesis and measurements of fungal inhibition and human-red-blood-cell hemolysis. One lead, peptide 6-4, was a measured variant of an existing aurein-derived template and achieved a selectivity index of 57.1 (60.9 using the study’s finer MIC interval) versus 1.2 for aurein 1.2. Its high selectivity was driven by low hemolysis; the result is a specific peptide-design and assay contribution, not a new peptide scaffold or evidence of therapeutic efficacy. The ChemRxiv v1 result was posted on 2024-10-04; the peer-reviewed journal article followed in 2025.",
    "field": "biology",
    "ai_technology": "Iterative uncertainty-aware Gaussian process regression using molecular descriptors and measured antifungal/hemolysis labels",
    "organization": "University of Wisconsin-Madison",
    "source_url": "https://doi.org/10.26434/chemrxiv-2024-gnpvz",
    "source_type": "paper",
    "publication_date": "2024-10-04",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1039/D4SC06689H",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Across six iterative rounds, uncertainty-aware Gaussian process regression selected 23 peptide variants for synthesis and antifungal/hemolysis assays. Lead 6-4 had a selectivity index of 57.1 (60.9 using the study’s finer MIC interval), versus 1.2 for aurein 1.2; the authors attribute the selectivity to reduced hemolysis.",
      "unknown": "Peptide 6-4 modifies a known aurein-derived template. The result is a measured sequence and assay outcome, not evidence of a new scaffold or therapeutic efficacy. The version-specific first public result is the 2024 ChemRxiv preprint; the journal article followed in 2025.",
      "ai_role": "An uncertainty-aware Gaussian process regression model selected peptide variants for synthesis and assays, and measured outcomes were fed back into subsequent selection rounds.",
      "evidence_urls": [
        "https://chemrxiv.org/engage/chemrxiv/article-details/67a7e4386dde43c9081bdfc1",
        "https://pubs.rsc.org/en/content/articlepdf/2025/sc/d4sc06689h",
        "https://pubmed.ncbi.nlm.nih.gov/40028619/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ChemRxiv version-history record and v2 manuscript; primary RSC indexed full article PDF, including sequence tables, Results, Methods and publication metadata; PubMed primary abstract/affiliation metadata."
    }
  },
  {
    "id": "2026-06-family-specific-machine-learning-enzyme-function",
    "title": "Family-specific machine learning prioritizes enzyme variants for prospective testing",
    "summary": "Researchers screened 1,513 natural SGNH-hydrolase homologues and trained task-specific sequence CNNs on the resulting activity, thermostability, and substrate-preference data. The models helped select candidates for a prospective 230-sequence assay panel, where CNN-selected candidates were enriched for target traits relative to generalist pretrained and physics-based baselines; the panel also included other selection strategies and random controls. Assays used clarified E. coli lysates and measured relative maximum velocity, integrating expression and catalysis. The first result-bearing version was posted to bioRxiv on 2026-06-03; a Research Square version followed on 2026-06-05.",
    "field": "biology",
    "ai_technology": "Task-specific convolutional neural networks trained on amino-acid sequences; CNN-based HEV and UCB selection strategies",
    "organization": "CSIRO Environment",
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.06.02.729712v1.full",
    "source_type": "paper",
    "publication_date": "2026-06-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.06.02.729712",
    "relevance": {
      "impact": 3,
      "validation": 4,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "A prospective panel of 230 previously untested sequences was assayed; CNN selection strategies were enriched for the predicted target traits compared with generalist pretrained and physics-based approaches. The panel pooled sequences selected by multiple strategies and included random controls.",
      "unknown": "Clarified lysate MaxV integrates enzyme expression and catalytic activity and is not a purified-enzyme turnover measurement. The result is scoped to this enzyme family and the reported assay setup; prospective panel composition was mixed across selection strategies.",
      "ai_role": "Supervised family-specific CNNs trained on sequence and assay data ranked untested homologues for activity, thermostability, and substrate preference, and guided candidate selection for prospective experimental testing.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.64898/2026.06.02.729712v1.full"
      ],
      "test_type": "prospective or independent check"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary bioRxiv v1 full text as indexed, including title, posting date, affiliations, methods, prospective assay results and limitations; primary Research Square v1 identity/date record."
    }
  },
  {
    "id": "2022-11-ml-screened-ionic-liquids-enable-ipe-membranes",
    "title": "Machine-learning screening guides ionic-liquid polymer electrolytes for lithium-metal batteries",
    "summary": "A machine-learning and quantum-chemistry workflow screened ionic liquids for conductivity and electrochemical-window properties, then guided selection for rigid-rod polyelectrolyte membranes. The 2022 preprint reports ~50 μm membranes and lithium-cell testing, including symmetric-cell cycling up to 800 hours at room temperature and reversible operation at 6 mA cm−2 at 80°C. Its preprint abstract reports 146 mAh g−1 at 5C/80°C; the later 2023 journal abstract lists 3C, so those version-specific figures are not conflated.",
    "field": "materials",
    "ai_technology": "Graph-convolutional neural network and ensemble SVM/random-forest/XGBoost screening with Psi4 quantum-chemistry descriptors",
    "organization": "Fudan University",
    "source_url": "https://assets-eu.researchsquare.com/files/rs-2220925/v1/0341f570-c3ce-430f-8b00-a72d3fa48184.pdf?c=1684221171",
    "source_type": "paper",
    "publication_date": "2022-11-09",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-38493-7",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The 2022 preprint reports a machine-learning/quantum-chemistry workflow that screened 2,356 ion pairs and recommended 40 for testing in rigid-rod polyelectrolyte membranes. It reports symmetric-cell cycling up to 800 hours at room temperature and reversible operation at 6 mA cm−2 at 80°C; its abstract reports 146 mAh g−1 at 5C/80°C.",
      "unknown": "The 2023 journal abstract reports the same study but gives 146 mAh g−1 at 3C rather than 5C. The record keeps the version-specific preprint metric separate and does not transfer it to the journal version. A publisher correction changes plot labels, not the core study claim.",
      "ai_role": "Machine-learning models and quantum-chemistry calculations screened ionic-liquid properties and guided selection for experimental polymer-electrolyte membranes and cell testing.",
      "evidence_urls": [
        "https://assets-eu.researchsquare.com/files/rs-2220925/v1/0341f570-c3ce-430f-8b00-a72d3fa48184.pdf?c=1684221171",
        "https://www.nature.com/articles/s41467-023-38493-7",
        "https://www.nature.com/articles/s41467-023-39621-z"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Research Square v1 full-text PDF, including version date, affiliations, Methods and Results; primary Nature Communications journal article indexed text and metadata; primary publisher correction notice."
    }
  },
  {
    "id": "2026-06-ml-screen-finds-selective-membrane-active-antibacterial-hits",
    "title": "Machine-learning screening identifies bacterial-selective membrane-active compounds",
    "summary": "The June 2026 bioRxiv version describes an initial machine-learning-guided screen of ZINC20 compounds balancing predicted antibacterial activity, mammalian-cell toxicity, and mechanism-of-action novelty. Among 61 purchased candidates, six showed partial growth inhibition; 49 were non-toxic to HepG2 cells at 50 μM. Flow cytometry indicated membrane disruption for four of the six initial active hits and membrane depolarization for two. A September version reports additional screening and revised models; its expanded results are not attributed to the June version here. The study does not report a novel protein target or clinical antibiotic.",
    "field": "biology",
    "ai_technology": "Graph-convolutional neural networks for antibacterial activity/toxicity with XGBoost mechanism-of-action classification and novelty filtering",
    "organization": "Princeton University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC13278111/",
    "source_type": "paper",
    "publication_date": "2026-06-11",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.06.08.730938",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The June 2026 v1 reports that 61 purchased candidates were tested, six showed partial growth inhibition, and 49/61 were non-toxic to HepG2 cells at 50 μM. Flow cytometry indicated membrane integrity disruption for four initial active hits and membrane depolarization for two.",
      "unknown": "The initial work does not report a novel protein target and is limited to the stated bacterial assays and HepG2 toxicity concentration. The September version adds expanded screening and revised analyses; those later results are not attributed to the June v1.",
      "ai_role": "Learned activity and toxicity models, together with a mechanism-novelty classifier, prioritized compounds for purchase and experimental testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13278111/",
        "https://pubmed.ncbi.nlm.nih.gov/42327090/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PMC-indexed full text for the June bioRxiv v1, primary PubMed preprint abstract/metadata, and primary PMC/PubMed metadata for the later September version to distinguish version-specific results."
    }
  },
  {
    "id": "2024-12-machine-learning-screening-repurposes-two-drugs-against-tuberculosis",
    "title": "Machine-learning screening identifies two drug-repurposing candidates against tuberculosis",
    "summary": "A virtual-screening workflow combined XGBoost and graph-neural-network models to screen 11,576 DrugBank compounds. Researchers selected 15 candidates for evaluation; four met the study’s activity cutoff against Mycobacterium tuberculosis H37Rv. Aldoxorubicin and quarfloxin were further tested against multidrug-resistant isolates, intracellular bacteria and recombinant DNA gyrase. These are laboratory repurposing results, not evidence of patient benefit or clinical safety. A same-team patent-family publication predates the journal article; this record makes no priority claim over patent disclosures. Concentration values are omitted because of inconsistent units in the rendered tables.",
    "field": "medicine",
    "ai_technology": "Ensemble virtual screening with XGBoost and graph neural networks",
    "organization": "Institute for Artificial Intelligence & Department of Computer Science and Technology, Tsinghua University",
    "source_url": "https://academic.oup.com/bib/article/26/1/bbae696/7934648",
    "source_type": "paper",
    "publication_date": "2024-12-30",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/bib/bbae696",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article reports follow-up testing of 15 selected compounds, four meeting the authors’ activity cutoff against H37Rv, and additional assays of aldoxorubicin and quarfloxin against multidrug-resistant isolates and recombinant DNA gyrase.",
      "unknown": "All reported results are in vitro. The study does not establish patient efficacy, clinical safety or the compounds’ cellular mechanism. Precise concentration values are omitted because table units are inconsistent; an earlier patent-family publication is noted without a priority claim.",
      "ai_role": "Multiple learned models, including XGBoost and graph neural networks, predicted antitubercular activity in a DrugBank screen and contributed to selection of candidates for experimental evaluation.",
      "evidence_urls": [
        "https://academic.oup.com/bib/article/26/1/bbae696/7934648"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Oxford Academic full indexed article, including abstract, Methods, results, article history and author affiliations; patent-family chronology checked from the patent family page. A-version patent text was not reviewed."
    }
  },
  {
    "id": "2021-03-ai-guided-antimalarial-labmol-167",
    "title": "AI-assisted virtual screening identifies a dual-stage antimalarial candidate",
    "summary": "An AI-assisted virtual-screening cascade using shape-based and machine-learning models prioritized potential Plasmodium kinase inhibitors. Eight virtual hits were experimentally evaluated; LabMol-167 inhibited P. berghei ookinete conversion and P. falciparum blood-stage parasites at nanomolar concentrations and showed low cytotoxicity in mammalian cells. A possible additional PK5 target was proposed from computational target fishing and docking, not demonstrated experimentally. This is a preclinical hit result and does not establish therapeutic efficacy. The article appeared online on 2021-03-16.",
    "field": "medicine",
    "ai_technology": "AI-assisted virtual screening combining shape-based and machine-learning models",
    "organization": "Federal University of Goiás",
    "source_url": "https://pubmed.ncbi.nlm.nih.gov/33247522/",
    "source_type": "paper",
    "publication_date": "2021-03-16",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/cmdc.202000685",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Eight virtual hits were evaluated; LabMol-167 showed nanomolar activity in P. berghei ookinete conversion and P. falciparum blood-stage assays, with low mammalian-cell cytotoxicity.",
      "unknown": "The PK5 dual-target mechanism is proposed from computational target fishing and docking; in-vivo efficacy and clinical benefit are not established.",
      "ai_role": "Shape-based and machine-learning virtual screening prioritized candidates for experimental evaluation; the abstract explicitly links the integrated AI-assisted screen to this candidate-identification workflow.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/33247522/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PubMed abstract, publication metadata, and author-affiliation block; claim limited to the findings explicitly stated in the abstract."
    }
  },
  {
    "id": "2023-08-physics-informed-machine-learning-curie-temperatures-ce-zr-fe-alloys",
    "title": "Physics-informed machine learning guides experimental Curie-temperature validation in Ce–Zr–Fe alloys",
    "summary": "A machine-learning model trained on measured Curie temperatures selected two compositions in the (Zr1−xCex)Fe2 system, (Zr0.16Ce0.84)Fe2 and (Zr0.94Ce0.06)Fe2, for experimental validation. The study also reports a relationship between electronic states near the Fermi level and Curie temperature, including a breakdown of de Gennes scaling below a rare-earth-content threshold. This is a materials discovery study, not evidence of a finished permanent magnet or deployed technology.",
    "field": "materials",
    "ai_technology": "Physics-informed machine-learning model trained on experimentally measured Curie temperatures",
    "organization": "Ames National Laboratory, U.S. Department of Energy; Iowa State University",
    "source_url": "https://pubs.acs.org/doi/10.1021/acs.chemmater.3c00892",
    "source_type": "paper",
    "publication_date": "2023-08-02",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.chemmater.3c00892",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors selected (Zr0.16Ce0.84)Fe2 and (Zr0.94Ce0.06)Fe2 for experimental Curie-temperature validation of the model-guided predictions.",
      "unknown": "The exact Curie-temperature measurements and uncertainty are not included here. This study does not establish a finished permanent magnet, device-level performance, or zero-carbon deployment.",
      "ai_role": "A learned model trained on experimentally measured Curie temperatures was used to predict multicomponent magnetic-material properties and select the two Ce–Zr–Fe compositions for experimental validation. The paper also reports that the model captured electronic-structure trends associated with Curie temperature.",
      "evidence_urls": [
        "https://pubs.acs.org/doi/10.1021/acs.chemmater.3c00892"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ACS publisher abstract, article metadata and supporting-information description. Full article numerical Curie-temperature results and uncertainty were not reviewed."
    }
  },
  {
    "id": "2023-10-bayesian-optimization-ptpd-cezrox-methane-combustion-catalyst",
    "title": "Bayesian optimization identifies a methane-combustion catalyst formula",
    "summary": "Bayesian optimization guided the experimental search over Pt/Pd and Ce/Zr ratios in PtPd@CeZrOx core–shell nanospheres. The authors report that two optimization rounds comprising 18 experiments found a formula with methane-conversion T50 approaching 330 °C, lower than prior reports cited by the study. The full assay conditions and comparison basis were not reviewed here; this is a bounded laboratory result, not a claim of general superiority, durability, or deployment.",
    "field": "materials",
    "ai_technology": "Bayesian optimization for experimental catalyst-formula selection",
    "organization": "Beihang University",
    "source_url": "https://onlinelibrary.wiley.com/doi/10.1002/anie.202313068",
    "source_type": "paper",
    "publication_date": "2023-10-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/anie.202313068",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The publisher abstract reports that BO guided two experimental optimization rounds (18 experiments total) and that the selected PtPd@CeZrOx formula reached a methane-conversion T50 approaching 330 °C.",
      "unknown": "Full test conditions, exact comparator basis, replicate variability and durability were not assessed in this abstract-limited review. The reported T50 does not establish field emissions performance or general superiority.",
      "ai_role": "Bayesian optimization selected multi-component Pt/Pd and Ce/Zr catalyst-formula ratios across iterative experiments; the reported T50 is the measured outcome associated with the selected formula.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/10.1002/anie.202313068",
        "https://pubmed.ncbi.nlm.nih.gov/37823845/"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Wiley publisher abstract/indexed article text and PubMed primary metadata/abstract read. Full article Methods/Results and supporting information were not accessible."
    }
  },
  {
    "id": "2024-09-deepbiosphere-maps-plant-distributions-and-fire-change",
    "title": "Deep learning maps fine-scale plant distributions and fire-associated change",
    "summary": "DeepBiosphere combines a residual CNN trained on citizen-science observations with aerial imagery to map California distributions for 2,221 plant species. The authors report median AUC 0.95 versus 0.88 for climate-only models and use repeat imagery to map plant-community change associated with the 2013 Rim Fire. These are California-specific ecological maps and observational associations; the study does not establish causal effects of fire or a new species discovery.",
    "field": "biology",
    "ai_technology": "DeepBiosphere residual convolutional neural network for species-distribution mapping",
    "organization": "Carnegie Science",
    "source_url": "https://www.pnas.org/doi/10.1073/pnas.2318296121",
    "source_type": "paper",
    "publication_date": "2024-09-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2318296121",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary article reports held-out mapping median AUC 0.95 versus 0.88 for climate-only models and uses repeat imagery to map community change following the 2013 Rim Fire.",
      "unknown": "The fire-severity relationship is observational and does not establish causation. Mapping performance is California-specific and weaker for rare species.",
      "ai_role": "A residual CNN trained with citizen-science observations and aerial imagery produced fine-scale species-distribution maps that were then used to estimate rapid plant-community change.",
      "evidence_urls": [
        "https://www.pnas.org/doi/10.1073/pnas.2318296121",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11406280/"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PNAS indexed article and PMC full-text indexed content, including abstract, Methods, results and plant/fire case studies; PubMed primary metadata/affiliations and first-online date."
    }
  },
  {
    "id": "2022-01-machine-learning-screening-of-icm104-energetic-material",
    "title": "Machine learning screening selects an energetic material for laboratory characterization",
    "summary": "A high-throughput workflow combined learned molecular-property predictions and crystal-packing classification to screen 25,112 generated structures and select ICM-104. The authors synthesized and characterized the selected material, reporting that its measured properties agreed with the model-assisted predictions. This is one laboratory-scale, study-specific material result; the broad candidate space remains computational, and the study does not establish scale-up, deployment or independent replication.",
    "field": "materials",
    "ai_technology": "Machine-learning property regressors and a convolutional neural network for crystal-packing classification",
    "organization": "Institute of Chemical Materials, China Academy of Engineering Physics",
    "source_url": "https://www.engineering.org.cn/engi/EN/1159869469231210541",
    "source_type": "paper",
    "publication_date": "2022-01-24",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.eng.2022.01.008",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study screened 25,112 generated candidates, selected one target material, then synthesized and characterized it; the publisher reports measured properties in agreement with predictions.",
      "unknown": "Only the selected target was synthesized and characterized. Scale-up, independent replication and operational performance are not established. Some energetic-performance values in the paper are calculated rather than directly measured.",
      "ai_role": "Learned molecular-property models and a CNN packing classifier helped rank and select the target for synthesis and characterization.",
      "test_type": "direct test",
      "evidence_urls": [
        "https://www.engineering.org.cn/engi/EN/1159869469231210541",
        "https://www.engineering.org.cn/engi/CN/1159869469231210541"
      ]
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Official publisher English article text/indexed conclusions and Chinese primary metadata/abstract reviewed; a saved 2026-11-16 review notes an author-hosted primary PDF was also checked."
    }
  },
  {
    "id": "2024-02-slide-links-t-cell-clone-expansion-to-islet-immune-programs",
    "title": "SLIDE links islet T-cell clone expansion to immune gene programs",
    "summary": "SLIDE, an interpretable latent-factor interaction method, was applied to paired single-cell RNA-seq and T-cell receptor data from islet-derived CD4 T cells in NOD mice. The analysis identified latent components associated with clonal expansion and grouped inhibitory-receptor and intracellular signaling/apoptosis-related genes among the associated programs. The result is an inference from existing multiomic data, not a causal perturbation; it is a distinct T1D application in the same article as the separately catalogued systemic-sclerosis analysis.",
    "field": "biology",
    "ai_technology": "SLIDE interpretable latent-factor interaction discovery and exploration",
    "organization": "University of Pittsburgh",
    "source_url": "https://www.nature.com/articles/s41592-024-02175-z",
    "source_type": "paper",
    "publication_date": "2024-02-19",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41592-024-02175-z",
    "relevance": {
      "impact": 3,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The method was applied to paired single-cell RNA-seq/TCR-seq from islet-derived CD4 T cells in NOD mice; the analysis reports latent components associated with clone expansion and compares the findings against an independent published dataset and alternative methods.",
      "unknown": "The inferred gene programs are associations from existing data; the T1D analysis does not prospectively perturb the proposed pathways or establish causality.",
      "ai_role": "SLIDE uses interpretable machine learning to identify interactions among latent factors in high-dimensional omics data and was applied to connect T-cell gene programs with clonal expansion.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41592-024-02175-z",
        "https://pubmed.ncbi.nlm.nih.gov/38374265/"
      ],
      "test_type": "retrospective or benchmark"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Methods indexed full article, including T1D application, Methods, results, author affiliations and publication/version dates; PubMed primary abstract and affiliation metadata."
    },
    "relationships": [
      {
        "type": "shared evidence",
        "target_id": "2024-02-slide-uncovers-a-keratinocyte-centered-signal-in-systemic-sclerosis",
        "reason": "Both records cite the same SLIDE article. This record describes its distinct T1D islet T-cell analysis; the linked record describes the systemic-sclerosis keratinocyte analysis. The relationship identifies a shared paper without asserting the results are identical or independent papers.",
        "source_urls": [
          "https://www.nature.com/articles/s41592-024-02175-z"
        ]
      }
    ]
  },
  {
    "id": "2020-07-transformer-generated-signal-peptides",
    "title": "An attention-based neural network generates functional secretion signal peptides",
    "summary": "An attention-based Transformer trained on signal-peptide data from Swiss-Prot generated candidate secretion sequences. Experimental testing in an industrial Bacillus subtilis strain found that model-generated signal peptides directed secreted enzyme activity competitive with industrially used signal peptides. The result establishes functional candidates in the reported host and assay; it does not establish broad performance across proteins or production hosts. The primary institutional record gives the article publication date as 2020-07-10.",
    "field": "biology",
    "ai_technology": "Attention-based neural network (Transformer) trained on Swiss-Prot signal peptide sequences",
    "organization": "California Institute of Technology",
    "source_url": "https://authors.library.caltech.edu/records/5dqa2-44d08",
    "source_type": "paper",
    "publication_date": "2020-07-10",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acssynbio.0c00219",
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary abstract reports experimental testing in an industrial Bacillus subtilis strain and secreted enzyme activity competitive with industrially used signal peptides.",
      "unknown": "The bounded abstract-level claim does not quantify all peptide/host combinations or demonstrate generalization across unrelated proteins and production conditions.",
      "ai_role": "An attention-based Transformer trained on Swiss-Prot generated signal-peptide sequences, which were then tested experimentally for protein secretion.",
      "evidence_urls": [
        "https://authors.library.caltech.edu/records/5dqa2-44d08"
      ],
      "test_type": "direct test"
    },
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Caltech institutional record with the article abstract, author contribution statement, publication date, and affiliation; the linked PDF download was unavailable in this session."
    }
  },
  {
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_type": "paper",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "id": "2026-09-ml-force-field-superionic-conductors",
    "title": "ML force-field screening identifies 12 superionic conductor candidates",
    "summary": "A symmetry-based search paired with molecular-dynamics calculations using a universal machine-learning force field identified five oxygen-, four lithium- and three sodium-based superionic conductor candidates. The authors report simulated diffusion properties comparable to leading materials and connected diffusion pathways. These are computational predictions; the study abstract does not establish experimental synthesis or measured ionic conductivity.",
    "field": "materials",
    "ai_technology": "Universal machine-learning force field for molecular-dynamics screening",
    "organization": "Zhejiang University; Westlake University",
    "source_url": "https://pubs.acs.org/doi/10.1021/acs.chemmater.6c01341",
    "publication_date": "2026-09-23",
    "paper_url": "https://doi.org/10.1021/acs.chemmater.6c01341",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher indexed abstract, article history, affiliation block and supporting-information listing; full article body not accessible in direct page fetch."
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary abstract reports molecular-dynamics screening with a universal machine-learning force field and identifies five oxygen-, four lithium- and three sodium-based candidates with simulated diffusion comparable to leading materials.",
      "unknown": "No experimental synthesis or measured conductivity is established by the reviewed abstract. The symmetry-based search idea is human-derived; the learned force field supports screening and calculated diffusion.",
      "ai_role": "A universal machine-learning force field was used in molecular-dynamics simulations that screened and evaluated the reported conductor candidates.",
      "evidence_urls": [
        "https://pubs.acs.org/doi/10.1021/acs.chemmater.6c01341"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "source_type": "paper",
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "id": "2026-09-gpt4o-literature-guided-limoobr4-conductor",
    "title": "GPT-4o-assisted literature screening nominates low-lithium LiMoOBr₄ conductor",
    "summary": "A literature-mining workflow using GPT-4o extracted values from 465 papers; a graph-neural-network-based pipeline screened 69,920 oxyhalide structures. Ab initio molecular-dynamics simulations for LiMoOBr₄ report a lithium content of 2.28 mol kg⁻¹, a 300 K extrapolated ionic conductivity of 13.53 mS cm⁻¹ and a 0.22 eV activation energy. The reported material properties are computational predictions, not experimental measurements.",
    "field": "materials",
    "ai_technology": "GPT-4o literature extraction; diffusion-model and graph-neural-network screening",
    "organization": "Tsinghua University; Peking University",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S3051383926001210",
    "publication_date": "2026-09-21",
    "paper_url": "https://doi.org/10.1016/j.jallcom.2026.186738",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher indexed abstract/highlights and saved publisher affiliation metadata; direct publisher page returned 403. University announcement checked only as corroboration, not as primary evidence."
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports computational screening of 69,920 oxyhalides and AIMD estimates for LiMoOBr₄: 2.28 mol kg⁻¹ lithium, 13.53 mS cm⁻¹ ionic conductivity extrapolated to 300 K and 0.22 eV activation energy.",
      "unknown": "The checked abstract/index does not establish experimental synthesis or measured conductivity; predictions should not be described as experimentally confirmed. Full methods were not accessible in the current check.",
      "ai_role": "GPT-4o was used to extract literature data from 465 papers; a learned graph-neural-network screening pipeline contributed to nomination of LiMoOBr₄ for computational validation.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S3051383926001210"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2022-10-bayesian-language-model-antibody-library",
    "title": "Bayesian language-model design improved measured scFv binding",
    "summary": "An end-to-end Bayesian language-model workflow designed diverse single-chain antibody fragment libraries that were measured for binding. In the 2023 journal version, the best generated scFv bound 28.7-fold better than the best directed-evolution comparator, and 99% of designs in the most successful library improved on the initial candidate. The 2022 bioRxiv v1 abstract already reported this result chain, giving 28.8-fold for the best generated scFv; the small difference is version-specific. These are in-vitro binding results, not clinical efficacy.",
    "field": "biology",
    "ai_technology": "Bayesian optimization with a language-model-based antibody sequence representation",
    "organization": "Massachusetts Institute of Technology Lincoln Laboratory; A-Alpha Bio, Inc.",
    "source_url": "https://www.biorxiv.org/content/10.1101/2022.10.07.502662v1",
    "source_type": "paper",
    "publication_date": "2022-10-07",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-39022-2",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Official bioRxiv API v1 date/abstract and primary Nature journal PDF for full result and affiliations."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "Designed scFvs were empirically measured. The publisher abstract reports the best generated scFv had 28.7-fold greater binding than the best directed-evolution scFv and 99% of the designed scFvs in the most successful library improved on the initial candidate.",
      "unknown": "The journal version reports the best generated scFv at 28.7-fold better binding while the v1 abstract reports 28.8-fold; this minor version discrepancy is retained. The measured binding result does not establish clinical efficacy.",
      "ai_role": "A Bayesian, language-model-based workflow designed candidate scFv libraries, whose binding was experimentally measured.",
      "evidence_urls": [
        "https://api.biorxiv.org/details/biorxiv/10.1101/2022.10.07.502662",
        "https://www.nature.com/articles/s41467-023-39022-2.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-deep-learning-minimal-cross-reactive-anti-pdl1-antibodies",
    "title": "Deep learning identified minimal mutations for cross-reactive anti-PD-L1 antibodies",
    "summary": "A deep-learning model trained on deep-mutational-scanning sequence–binding data selected minimal mutation combinations for a human anti-PD-L1 antibody with mouse cross-reactivity. Variants with four to five substitutions showed in-vitro and cellular binding comparable to conventionally recombined variants with 13–15 substitutions; the paper also reports blockade assays and therapeutic activity in a mouse tumor model. These are preclinical antibody-engineering results, not evidence of human efficacy. The preprint reports patent applications related to the subject matter.",
    "field": "biology",
    "ai_technology": "Deep-learning model trained on deep-mutational-scanning sequence–binding data",
    "organization": "Université Paris-Saclay; CEA; INRAE; CNRS; Inserm; Institut Agro Rennes",
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.07.15.738714v1",
    "source_type": "paper",
    "publication_date": "2026-07-16",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.07.15.738714",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary bioRxiv v1 indexed full-text/abstract, author-affiliation metadata and official API metadata; direct page open returned 403."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The preprint reports in-vitro and cellular binding properties of selected variants, PD-1/PD-L1 competition/blockade assays, epitope mapping and mouse tumor-model assessment. The 4–5 substitution variants are compared with highly mutated variants requiring 13–15 substitutions.",
      "unknown": "Mouse-model activity is preclinical and does not establish human therapeutic efficacy. The preprint reports patent applications relevant to the work; no freedom-to-operate or patent-priority conclusion is made.",
      "ai_role": "A deep-learning model trained on DMS-derived sequence-binding data selected minimal mutation combinations for experimental antibody characterization.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.64898/2026.07.15.738714v1.full",
        "https://www.biorxiv.org/content/10.64898/2026.07.15.738714v1",
        "https://api.biorxiv.org/details/biorxiv/10.64898/2026.07.15.738714"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-01-ml-pipeline-antimicrobial-hexapeptides",
    "title": "Machine-learning pipeline identifies antimicrobial hexapeptides",
    "summary": "A sequential machine-learning pipeline screened virtual libraries of 6–9-residue peptides and identified three leading antimicrobial hexapeptides. The paper reports activity against multidrug-resistant clinical isolates and therapeutic effects in a mouse bacterial-pneumonia model. These are laboratory and animal results, not clinical efficacy.",
    "field": "biology",
    "ai_technology": "Sequential machine-learning pipeline for peptide selection, classification, ranking and activity regression",
    "organization": "Zhejiang University",
    "source_url": "https://www.nature.com/articles/s41551-022-00991-2",
    "source_type": "paper",
    "publication_date": "2023-01-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41551-022-00991-2",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature indexed article text and PubMed abstract/metadata/affiliation record for the ML hexapeptide screening study, its January 2023 publication date and reported mouse pneumonia assays."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors report a sequential machine-learning pipeline over virtual 6–9 amino-acid sequence space; the leading three identified hexapeptides were tested against clinical multidrug-resistant isolates and in a mouse bacterial-pneumonia model.",
      "unknown": "The mouse model and reported comparison with penicillin do not establish human efficacy. The PubMed primary metadata gives journal online date 2023-01-12; no earlier pre-2020 version is identified in the reviewed primary material.",
      "ai_role": "Sequential learned modules selected, classified, ranked and regressed peptide candidates from the virtual sequence library; the reported three lead hexapeptides are outputs of that workflow.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41551-022-00991-2",
        "https://pubmed.ncbi.nlm.nih.gov/36635418/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-09-ecnet-guided-tem1-variants",
    "title": "ECNet-guided selection of higher-order TEM-1 variants",
    "summary": "ECNet integrated evolutionary context with learned sequence representations to prioritize previously untested higher-order TEM-1 variants. The authors constructed and measured 37 selected variants under ampicillin selection, using relative resistance as an enzyme-fitness proxy. The assays do not imply clinical treatment outcomes, and the reported group of 37 should not be read as showing that every variant improved.",
    "field": "biology",
    "ai_technology": "ECNet evolutionary-context-integrated deep-learning framework",
    "organization": "University of Illinois Urbana-Champaign",
    "source_url": "https://www.nature.com/articles/s41467-021-25976-8",
    "source_type": "paper",
    "publication_date": "2021-09-30",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-021-25976-8",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature indexed article and author-hosted full text; official bioRxiv API checked and identified the previously associated preprint DOI as a different paper."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Thirty-seven previously untested higher-order TEM-1 variants selected with ECNet were constructed and experimentally measured under ampicillin selection against wild-type and study controls. The study uses relative resistance as an enzyme-fitness proxy.",
      "unknown": "The bounded entry does not imply that all 37 variants improved equally, does not establish clinical outcomes, and makes no claim of independent sequence novelty beyond the authors’ screened literature.",
      "ai_role": "ECNet integrated evolutionary context and a learned representation to prioritize higher-order TEM-1 variants for experimental construction and ampicillin-selection measurements.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-021-25976-8",
        "https://zhaogroup.web.illinois.edu/publications/HZ370.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-cosann-induced-fit-glucokinase-design",
    "title": "CoSaNN designs and tests an induced-fit carbohydrate kinase",
    "summary": "CoSaNN combined neural structure prediction, learned sequence optimization and solubility ranking to design variants of an induced-fit ROK carbohydrate kinase. The authors measured expression, thermal stability and activity in vitro. The paper reports measurable expression for 189 of 348 designs in one table, while the strategy-level counts do not sum to that total; no success-rate claim is made here.",
    "field": "biology",
    "ai_technology": "CoSaNN: AlphaFold2-based conformation modeling, ProteinMPNN sequence design and SolvIT graph neural network ranking",
    "organization": "Enzymit Ltd.",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC10945820/",
    "source_type": "paper",
    "publication_date": "2024-03-04",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1073/pnas.2313809121",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PMC full text and PNAS DOI record for CoSaNN-guided glucokinase design, Enzymit affiliations, experimental activity results and the discrepancy in reported design totals."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors applied CoSaNN to redesign the Arthrobacter polyphosphate glucokinase ROK kinase and performed expression, thermal-stability and enzyme-activity assays on designed variants. The primary Results table reports 189 designs with detectable soluble expression among 348 listed designs; its strategy counts (119+95+94 plus 40 in another strategy; rows in the prose/table are internally inconsistent) do not reconcile cleanly, so the record does not use a single total as a success-rate denominator.",
      "unknown": "The study tests a company-designed enzyme set in vitro; it does not establish industrial deployment. The primary paper’s count inconsistency remains explicitly disclosed.",
      "ai_role": "CoSaNN uses deep-learning structure prediction (AlphaFold2), sequence optimization with ProteinMPNN and a learned solubility classifier (SolvIT) to rank enzyme designs for experimental characterization.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10945820/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-01-active-learning-automatic-feature-engineering-ocm-catalysts",
    "title": "Active learning selected oxidative methane coupling catalysts",
    "summary": "The authors combined automatic feature engineering with active learning and high-throughput experiments for oxidative coupling of methane. Of 36 selected catalysts, 30 reached combined ethane and ethylene yields of at least 15% in the reported tests. This is a small-data, reaction-specific catalyst screening outcome; the paper also tests feature engineering on other catalysis datasets. A ChemRxiv version predates the journal publication, and this record does not claim journal-version priority.",
    "field": "chemistry",
    "ai_technology": "Automatic feature engineering with active learning",
    "organization": "Japan Advanced Institute of Science and Technology; Hokkaido University",
    "source_url": "https://www.nature.com/articles/s42004-023-01086-y",
    "source_type": "paper",
    "publication_date": "2024-01-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s42004-023-01086-y",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Full published article text in German National Library repository and publisher-indexed primary metadata/abstract; earlier ChemRxiv version noted but not compared."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article reports 36 selected oxidative coupling of methane catalysts experimentally assessed; 30 achieved combined ethane and ethylene yields of at least 15% under the reported test conditions.",
      "unknown": "The paper also appeared as an earlier ChemRxiv version; this check did not establish whether every quantitative detail appeared in that version. The result is limited to the reported catalyst set and reaction conditions.",
      "ai_role": "Automatic feature engineering extracted catalyst descriptors, and active learning selected catalysts for high-throughput experimental testing.",
      "evidence_urls": [
        "https://d-nb.info/1324333405/34",
        "https://www.nature.com/articles/s42004-023-01086-y"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-02-active-learning-magnesium-air-2-3-dihydroxynaphthalene",
    "title": "Adaptive learning selected an additive for a magnesium-air battery",
    "summary": "An iterative machine-learning adaptive-design workflow selected 2,3-dihydroxynaphthalene for a lab-made (Mg-0.2Ca)-air battery. The authors report a cell voltage of 1.82 V and 3.37 kWh kg⁻¹ specific energy calculated from total Mg-anode mass loss, not whole-cell or pack mass. The publisher abstract reports 83% anodic utilization, while the full article reports 85% at a stated 0.5 mA cm⁻² condition; this discrepancy is not resolved here and the utilization figure is omitted. The authors also state that the surrogate model overestimated performance predictions for this additive.",
    "field": "materials",
    "ai_technology": "Iterative adaptive design with SOAP descriptors and a KPCovR surrogate model",
    "organization": "Institute of Surface Science, Helmholtz-Zentrum Hereon; Hebei University of Technology; Kiel University",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S2405829725001205",
    "source_type": "paper",
    "publication_date": "2025-02-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.ensm.2025.104120",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Elsevier indexed abstract and full publisher PDF text hosted by ResearchGate; primary Hereon repository metadata cross-check. Specific energy basis and internal utilization-efficiency discrepancy reviewed."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors report a lab-made (Mg-0.2Ca)-air battery with the model-selected 2,3-dihydroxynaphthalene additive at 1.82 V. Reported specific energy is 3.37 kWh kg−1 calculated from total Mg-anode mass loss. No whole-cell or pack-mass specific energy is claimed.",
      "unknown": "Publisher abstract reports 83% anodic utilization; full article text gives 85% at a stated 0.5 mA cm−2 condition. The difference is unresolved, so no utilization percentage is asserted. The authors report that the surrogate model overestimated predictions for this additive.",
      "ai_role": "A SOAP-descriptor/KPCovR model in iterative adaptive experimental design contributed to selecting 2,3-dihydroxynaphthalene for electrochemical validation.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S2405829725001205",
        "https://www.researchgate.net/publication/388941650_Machine_learning-guided_discovery_of_high-efficiency_electrolyte_additives_for_aqueous_magnesium-air_batteries"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-01-ml-predicted-ucd3",
    "title": "Machine learning guides a proposed UCd3 intermetallic phase assignment",
    "summary": "A support-vector classifier using composition-derived elemental features predicted a Cu₃Au-type structure for UCd₃ and guided synthesis. Powder X-ray diffraction from the second of two samples showed peaks the authors fit to a UCd₃ phase with a = 4.648(2) Å; the sample also contained uranium, uranium oxide and cadmium, while the first sample differed. This is the authors’ proposed phase assignment from a multiphase powder, not phase-pure material or independent structural confirmation.",
    "field": "materials",
    "ai_technology": "Support-vector machine discriminant analysis; Gaussian-process regression for feature completion",
    "organization": "The Cooper Union for the Advancement of Science and Art; Manhattan College",
    "source_url": "https://chemrxiv.org/engage/chemrxiv/article-details/63b23bbd16e9a864303c5bf0",
    "source_type": "paper",
    "publication_date": "2023-01-03",
    "added_date": "2026-11-16",
    "verification_status": "verified",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ChemRxiv v1 full text, affiliation/date metadata, methods and diffraction discussion, plus primary UCd11 literature review for prior-compound context."
    },
    "paper_url": "https://doi.org/10.26434/chemrxiv-2023-0nlzl",
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors report synthesis and powder X-ray diffraction; the proposed phase was present in the second of two samples.",
      "unknown": "Only one synthesis batch/sample (the second of two samples) showed the additional peaks assigned by the authors to UCd₃; the powder was multiphase. The checked 2022 primary review says UCd₁₁ was the only previously reported U-Cd compound, but this is not an independent confirmation of the proposed UCd₃ structure.",
      "ai_role": "SVM classification of composition-derived descriptors predicted a Cu₃Au-type structure and motivated experimental synthesis/testing.",
      "evidence_urls": [
        "https://www.cambridge.org/engage/api-gateway/coe/assets/orp/resource/item/63b23bbd16e9a864303c5bf0/original/machine-learning-descriptors-in-materials-chemistry-prediction-and-experimental-validation-synthesis-of-novel-intermetallic-u-cd3.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-02-ml-guided-xylm-15fold-biosensor-productivity",
    "title": "Machine learning guided XylM variant improved biosensor-indicated productivity",
    "summary": "Two machine-learning-assisted directed-evolution cycles produced XylM-D140E-V144K-F243L-N244S, reported to have 15-fold higher 3-methylsalicylic acid productivity than wild-type XylM. Productivity was estimated using XylS-dependent fluorescence in a sensor strain; this is a biosensor-based productivity readout, not a purified-enzyme kinetic constant.",
    "field": "biology",
    "ai_technology": "Machine-learning-assisted directed evolution using biosensor-derived activity data",
    "organization": "The University of Tokyo; National Institute of Advanced Industrial Science and Technology; AIST-Waseda University Computational Bio Big-Data Open Innovation Laboratory",
    "source_url": "https://pubs.acs.org/doi/abs/10.1021/acssynbio.2c00577",
    "source_type": "paper",
    "publication_date": "2023-02-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acssynbio.2c00577",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ACS publisher indexed abstract and affiliation/article-history metadata, cross-checked against primary PubMed abstract and metadata."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The publisher reports XylM-D140E-V144K-F243L-N244S with 15-fold greater 3-methylsalicylic-acid productivity than wild type after two ML-assisted directed-evolution cycles, using XylS-dependent fluorescence from a sensor strain as the measured indicator.",
      "unknown": "The signal is an indirect biosensor-based productivity proxy; the reported value is not a purified-enzyme kinetic constant.",
      "ai_role": "Machine learning used sensor-derived data to guide two rounds of directed evolution and prioritize XylM variants.",
      "evidence_urls": [
        "https://pubs.acs.org/doi/abs/10.1021/acssynbio.2c00577",
        "https://pubmed.ncbi.nlm.nih.gov/36734676/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-04-ml-guided-cu-co2-reduction-additive-discovery",
    "title": "Machine learning guided Cu catalyst additive discovery for CO₂ reduction",
    "summary": "Across three iterative cycles of experimental testing, machine-learning analysis, prediction and redesign, the authors obtained Cu catalysts selective for CO, formic acid and C₂+ products in electrochemical CO₂ reduction. This summary follows the primary abstract and does not assign disputed sample identifiers or report cell-specific percentages.",
    "field": "chemistry",
    "ai_technology": "Machine-learning-guided iterative additive screening and redesign",
    "organization": "Xiamen University; Northwestern Polytechnical University",
    "source_url": "https://pubs.acs.org/doi/abs/10.1021/jacs.1c00339",
    "source_type": "paper",
    "publication_date": "2021-04-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/jacs.1c00339",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ACS abstract, author-affiliation and article-history metadata; full text and supporting information not reviewed."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The primary abstract reports three iterative cycles of experimental testing and model-guided redesign, producing Cu catalysts selective for CO, formic acid and C₂+ products. No disputed sample-level metrics are included.",
      "unknown": "Full text and supporting information were not reviewed. The exact sample-to-performance mapping and the third-party rendering’s sample-number discrepancy remain unverified.",
      "ai_role": "Machine-learning analysis of experimental additive/catalyst data guided predictions and redesign over three iterative experimental cycles.",
      "evidence_urls": [
        "https://pubs.acs.org/doi/abs/10.1021/jacs.1c00339"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2021-11-mutcompute-bst-polymerase-variants",
    "title": "MutCompute guides thermostable Bst DNA polymerase variants",
    "summary": "MutCompute predicted substitutions in Bst DNA polymerase; the authors constructed and tested variants in loop-mediated isothermal amplification assays. The Mut235 triple variant showed the highest activity among four tested triple-mutant variants and remained active at 73 °C; its reported melting temperature was 78.1 °C. The evidence is in vitro and does not establish clinical diagnostic performance. The journal article cites a bioRxiv v1 posted 2021-04-15, but that version’s full text was inaccessible in this review; this entry uses the verified journal result and does not claim when the Mut235 result first appeared.",
    "field": "biology",
    "ai_technology": "MutCompute machine-learning amino-acid environment scoring",
    "organization": "The University of Texas at Austin",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC9514386/",
    "source_type": "paper",
    "publication_date": "2021-11-11",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.biochem.1c00451",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PMC full journal text for MutCompute, LAMP assays, University of Texas affiliation and measured variants; primary bioRxiv v1 endpoint was attempted but blocked, so earlier-version result details remain bounded to what the journal and cited metadata support."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors assayed ML-predicted Bst polymerase substitutions in LAMP/LAMP-OSD reactions. Mut235 showed the highest activity among four triple-mutant variants tested, supported amplification at 73 °C, and had a measured melting temperature of 78.1 °C versus 75.5 °C for parental Bst-LF in the reported thermal-shift assay.",
      "unknown": "The article cites an earlier bioRxiv preprint dated 2021-04-15, but this check did not retrieve its full v1 text; no claim is made that every final-version assay/result was present in v1. Results are in-vitro amplification assays, not clinical diagnostic validation.",
      "ai_role": "MutCompute machine learning predicted amino-acid substitutions in Bst DNA polymerase; the authors constructed and assayed variants including Mut235.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC9514386/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-11-alphazero-extremal-graphs",
    "title": "AlphaZero finds improved finite extremal-graph bounds",
    "summary": "A neural network-guided AlphaZero search with a curriculum generated triangle- and 4-cycle-free graphs that improve known lower bounds for the extremal function ex(n,{C3,C4}) over finite sizes; the proceedings report AlphaZero improvements for n=64–136 and release the graph constructions. The paper also finds improvements using a separate tabu-search method, which often matches or outperforms AlphaZero. This does not solve the asymptotic Erdős conjecture. The result was present in the arXiv v1 posted 2023-11-06; the 2024 IJCAI proceedings paper is a later version.",
    "field": "mathematics",
    "ai_technology": "AlphaZero reinforcement learning with neural Pairformer representation and curriculum",
    "organization": "Google DeepMind; Imperial College London; Yonsei University; University of Bonn; Worcester Polytechnic Institute",
    "source_url": "https://arxiv.org/html/2311.03583v1",
    "source_type": "paper",
    "publication_date": "2023-11-06",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://ijcai.org/proceedings/2024/0772",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 full text and metadata plus IJCAI 2024 proceedings page/PDF for AlphaZero versus tabu-search graph construction results, theorem limits and author affiliations."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The paper reports released graphs with improved lower bounds for ex(n,{C3,C4}) for node counts 64–134. Incremental AlphaZero (neural network-guided Monte Carlo tree search) is explicitly reported to improve bounds for sizes 64–136 in the proceedings version; incremental tabu search also improves bounds and the authors report it can match or exceed AlphaZero. This is a finite-graph construction result, not a proof of the asymptotic Erdős conjecture.",
      "unknown": "The result consists of explicit finite constructions/lower bounds; the asymptotic conjecture remains open. The work does not establish that learned search outperforms strong local search.",
      "ai_role": "AlphaZero with a neural Pairformer representation and a curriculum generated extremal graphs improving the established lower bounds for n=64–136; tabu search is a separate, non-neural method and its results are not attributed to AI.",
      "evidence_urls": [
        "https://arxiv.org/html/2311.03583v1",
        "https://ijcai.org/proceedings/2024/0772",
        "https://ijcai.org/proceedings/2024/0772.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-evolvepro-guided-hpabc-l-dopa",
    "title": "Machine learning and directed evolution improved an L-Dopa-producing monooxygenase",
    "summary": "Zero-shot protein models prioritized HpaBC substitutions, which were experimentally assayed with a L-Dopa-responsive GFP biosensor. EVOLVEpro then combined experimentally validated substitutions, including directed-evolution-derived variants, and selected higher-order designs. The best reported combined variants produced up to 57–58 times the wild-type fluorescence signal under the stated assay conditions. This is a biosensor proxy for L-Dopa production, not a purified-enzyme kinetic measurement; the work also builds on prior directed-evolution mutations and does not establish clinical benefit.",
    "field": "biology",
    "ai_technology": "MutCompute, ProteinMPNN, ESM protein language models and EVOLVEpro combinatorial design",
    "organization": "The University of Texas at Austin",
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.07.31.742090v1",
    "source_type": "paper",
    "publication_date": "2026-08-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.07.31.742090",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary bioRxiv v1 API metadata and Crossref posted-date metadata checked; primary v1 full-text indexed extract reviewed for methods, biosensor-normalized outcomes, model role and affiliation. Direct full-text page open was blocked."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "329 model-predicted variants were assayed; 19 showed at least 1.25-fold biosensor signal. EVOLVEpro-selected combined variants yielded up to 57–58-fold normalized fluorescence relative to wild type under the reported assay conditions.",
      "unknown": "Fluorescence from a DopA/GFP biosensor is an indirect L-Dopa-production proxy, not a purified-enzyme catalytic constant. The work includes directed-evolution-derived mutations; G295T had prior directed-evolution support. No human therapeutic efficacy is established.",
      "ai_role": "MutCompute, ProteinMPNN and ESM prioritized single substitutions for experimental testing; EVOLVEpro combined tested mutations and iteratively selected higher-order variants, including designs incorporating directed-evolution results.",
      "evidence_urls": [
        "https://api.biorxiv.org/details/biorxiv/10.64898/2026.07.31.742090",
        "https://www.biorxiv.org/content/10.64898/2026.07.31.742090v1.full"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-07-progressive-learning-single-atom-oxide-oer-catalysts",
    "title": "Progressive learning screened single-atom oxide catalysts for acidic oxygen evolution",
    "summary": "A progressive-learning workflow combining density-functional theory and machine learning screened 261 single-atom-doped metal oxides and identified nine candidates with predicted OER overpotential below 0.5 V. The publisher abstract reports experimental evaluation confirming predicted activity and durability for Mn single atoms on RuO₂ under acidic conditions. Candidate-wide values are theoretical predictions; only the specified material’s validation is summarized as experimental.",
    "field": "chemistry",
    "ai_technology": "Progressive learning with DFT-derived adsorption-energy features and machine learning",
    "organization": "University of Puerto Rico, Río Piedras; Zhejiang University; Shandong University of Technology; Chung-Ang University",
    "source_url": "https://onlinelibrary.wiley.com/doi/10.1002/anie.202510965",
    "source_type": "paper",
    "publication_date": "2025-07-12",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/anie.202510965",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher abstract, first-publication date, and author-affiliation metadata; full article and supporting information not reviewed."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The abstract reports experimental evaluation confirming predicted activity and durability for MnSA-RuO₂ under acidic conditions. The nine candidates and their <0.5 V overpotentials are theoretical screening outcomes, not nine experimental validations.",
      "unknown": "The abstract does not provide detailed experimental conditions or an independent comparison; those are not inferred.",
      "ai_role": "Progressive learning combined DFT with machine learning, using adsorption-energy predictions as auxiliary features to screen single-atom-doped oxide catalysts.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/10.1002/anie.202510965"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-ml-enriched-mthupo-beta-damascone-hydroxylation",
    "title": "Machine learning enriched MthUPO variants for beta-damascone hydroxylation",
    "summary": "Researchers combined experimental activity measurements with machine learning to enrich a second MthUPO peroxygenase library for substrate-specific activity. Across 506 distinct variants screened and sequenced, 97 of 108 variants in the ML-enriched library outperformed wild type in the screening assay, compared with 14% in the initial htFuncLib library. The authors report an 11.8-fold turnover number for the best variant producing 4-hydroxy-beta-damascone, with greater than 99% regioisomeric excess. Separate 5–12-fold enrichment and recovery figures in the paper are simulation benchmarks and are not treated as experimental outcomes.",
    "field": "biology",
    "ai_technology": "Machine-learning enrichment of experimental enzyme-activity data following htFuncLib design",
    "organization": "Martin Luther University Halle-Wittenberg",
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.07.20.739108v1.full",
    "source_type": "paper",
    "publication_date": "2026-07-21",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.07.20.739108",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Official bioRxiv API v1 metadata/abstract and primary indexed v1 full-text excerpts; simulation and experimental results kept separate."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The preprint reports screening and sequencing 506 distinct MthUPO variants. In the ML-enriched library, 97 of 108 variants outperformed wild type in the screening assay, compared with 14% in the initial htFuncLib library. The best reported variant had an 11.8-fold turnover number for 4-hydroxy-beta-damascone with greater than 99% regioisomeric excess.",
      "unknown": "The preprint is not peer reviewed. The paper separately reports simulation-only MLEE/htFuncLib comparisons, which are excluded from the experimental result summary. Assay outcomes do not establish process-scale productivity or commercial suitability.",
      "ai_role": "Machine learning used experimental enzyme-activity measurements to enrich a library of MthUPO variants for testing, following an earlier htFuncLib design stage.",
      "evidence_urls": [
        "https://api.biorxiv.org/details/biorxiv/10.64898/2026.07.20.739108",
        "https://www.biorxiv.org/content/10.64898/2026.07.20.739108v1.full"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-11-neural-network-screened-oxide-catalysts-acetylene-semihydrogenation",
    "title": "Machine learning screened metal oxides for acetylene semihydrogenation",
    "summary": "A neural-network model was used to predict metal-oxide catalyst performance for acetylene semi-hydrogenation. The authors report experimental validation of predictions for CuO, ZnO and V2O5, and report 99.6% conversion with 90% ethylene selectivity for CuO under the abstract-described conditions. The entry is limited to the publisher abstract; detailed model validation, candidate selection and full assay context were not reviewed.",
    "field": "chemistry",
    "ai_technology": "Neural-network prediction of metal-oxide catalyst performance",
    "organization": "Zhejiang University of Technology",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S0926860X24004149",
    "source_type": "paper",
    "publication_date": "2024-11-25",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.apcata.2024.119969",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher abstract for identity, displayed date, model and reported experiment; official Zhejiang University of Technology faculty profile/publication list for named contributor affiliation. Full paper methods and supplement not reviewed."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The publisher abstract says CuO, ZnO and V2O5 predictions were experimentally validated; it reports 99.6% conversion and 90% ethylene selectivity for CuO at 50–100°C.",
      "unknown": "Detailed selection chronology, model split/performance, full test conditions and results for ZnO and V2O5 were not established from the reviewed abstract. The date is the publisher-displayed date, not a claim of first public appearance.",
      "ai_role": "A neural-network model predicted catalyst performance and informed the metal-oxide candidates reported as experimentally validated.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S0926860X24004149",
        "https://homepage.zjut.edu.cn/ljh_en/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-ml-optimized-photocatalytic-co2-micelles",
    "title": "Machine learning guided multi-metric optimization of photocatalytic CO2 reduction",
    "summary": "Bayesian optimization guided experiments on a five-component self-assembled photocatalytic micelle system for CO₂-to-CO reduction. The authors report that optimized conditions increased measured CO yield from 1.48 ± 0.24 to 5.72 ± 0.12 μmol h⁻¹, with improved turnover and quantum-yield metrics while maintaining CO selectivity. The article’s narrative fold-change (4.2×) is inconsistent with the displayed means (about 3.9×), so this record gives the means and omits the fold claim. The findings apply to the tested aqueous micelle system and conditions.",
    "field": "chemistry",
    "ai_technology": "Bayesian optimization for experimental selection; random-forest models and SHAP for later analysis",
    "organization": "University of Cambridge; Politecnico di Torino; Uppsala University",
    "source_url": "https://doi.org/10.1021/jacs.4c01305",
    "source_type": "paper",
    "publication_date": "2024-05-20",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ACS publisher metadata and University of Cambridge repository copy of the article full text checked for article identity, affiliations, date, optimization method, results and source discrepancy."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article reports CO yield means of 5.72±0.12 and 1.48±0.24 μmol h−1 for optimized vs initial conditions, along with turnover and quantum-yield metrics and maintained selectivity. The narrative states a 4.2-fold yield improvement, inconsistent with these displayed means; the record does not repeat the fold value.",
      "unknown": "This demonstrates a study-specific photocatalytic system optimization, not industrially scaled carbon conversion or independent replication. The source’s stated yield fold change conflicts with its displayed means.",
      "ai_role": "Bayesian optimization guided the experimental parameter search and selected test conditions in a five-component micelle system; random-forest models and SHAP then analyzed parameter effects.",
      "evidence_urls": [
        "https://api.repository.cam.ac.uk/server/api/core/bitstreams/9bb1fb68-8d73-4b07-b47b-376146e130f8/content",
        "https://pubs.acs.org/doi/pdf/10.1021/jacs.4c01305?download=true"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-12-active-learning-knn-piezoceramic-composition",
    "title": "Active learning identified a KNN ceramic composition with high piezoelectric response",
    "summary": "An active-learning loop combined machine-learning prediction, surrogate optimization and experimental testing to explore lead-free potassium-sodium niobate (KNN) ceramics. After three experiment and optimization iterations, the authors report identifying a composition with a measured piezoelectric coefficient d33 of approximately 353 pC/N from thousands of possible compositions. This is a laboratory material-property measurement, not validation of a finished sensor or device.",
    "field": "materials",
    "ai_technology": "Support vector regression with a radial basis function kernel in an active-learning experimental-design loop",
    "organization": "Nanjing University of Aeronautics and Astronautics",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S0272884224048338",
    "source_type": "paper",
    "publication_date": "2024-12-15",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.ceramint.2024.10.312",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher indexed full text/abstract and primary institutional repository metadata/abstract for article date, author affiliation, model workflow and experimental result."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "After three active-learning-guided experiment iterations, the authors report a synthesized KNN composition with measured d33 of approximately 353 pC/N.",
      "unknown": "The reported coefficient is a laboratory material-property measurement, not device-level qualification or commercial performance.",
      "ai_role": "An SVR model predicted KNN piezoelectric response and guided composition selection across three experiment-optimization iterations.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S0272884224048338",
        "https://research.mpu.edu.mo/en/publications/multi-iteration-active-learning-for-the-composition-design-of-pot/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-ml-screened-pore-forming-antimicrobial-peptides",
    "title": "Machine-learning screening identified experimentally active pore-forming peptides",
    "summary": "Machine-learning models and membrane-related filters were used to prioritize natural peptide sequences for testing. The study reports seven experimentally active peptides; three showed broad-spectrum activity and pore-forming evidence in the reported membrane assays. Peptide 615 had already been disclosed in a Peking University patent published in March 2025, before the related preprint and journal article. This entry represents one study campaign and makes no first-disclosure or sequence-novelty claim. Assay results do not establish clinical safety or efficacy.",
    "field": "biology",
    "ai_technology": "ProtBERT-BFD sequence embeddings with a support-vector classifier and membrane-interaction, secondary-structure and contact-map predictions",
    "organization": "Peking University; Yantai University; Institute of Physics, Chinese Academy of Sciences; University of Chinese Academy of Sciences; École Polytechnique Fédérale de Lausanne; University of Geneva; Songshan Lake Materials Laboratory",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12904036/",
    "source_type": "paper",
    "publication_date": "2025-12-14",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/advs.202516470",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PMC full text, metadata and affiliation block; published patent text for earlier public disclosure; primary preprint record identified in the frozen source review."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports seven experimentally active selected peptides and pore-forming measurements for three broad-spectrum candidates.",
      "unknown": "One peptide sequence, 615, was disclosed in an earlier university patent. The article does not establish clinical safety or efficacy, and no universal mechanism is claimed.",
      "ai_role": "Machine-learning sequence and membrane-related models prioritized peptide candidates for experimental testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12904036/",
        "https://patents.google.com/patent/CN119591677A/zh",
        "https://doi.org/10.1101/2025.05.17.654650"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-06-broadamp-gpt-amp-s13",
    "title": "BroadAMP-GPT generated and experimentally tested broad-spectrum AMP_S13",
    "summary": "BroadAMP-GPT combined transformer-based peptide generation, computational screening and experimental testing. The paper reports that 57% of AI-generated candidates showed potent activity against ESKAPE pathogens. AMP_S13 showed broad-spectrum antimicrobial activity, stability under tested conditions, low cytotoxicity and hemolysis, and efficacy in Galleria mellonella and a murine MRSA wound model. These preclinical findings do not establish human safety or efficacy; the 57% abstract metric is kept distinct from the separately reported assays of 14 synthesized candidates.",
    "field": "biology",
    "ai_technology": "Transformer-based peptide generation with deep-learning-guided screening",
    "organization": "Nanjing Agricultural University; Hainan University; China Pharmaceutical University",
    "source_url": "https://www.tandfonline.com/doi/full/10.1080/19490976.2025.2523811",
    "source_type": "paper",
    "publication_date": "2025-06-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1080/19490976.2025.2523811",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher full article/PDF, publisher publication history and NCBI/PubMed metadata/abstract; first-public preprint and sequence priority are not claimed."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports a 57% potent-activity rate among AI-generated candidates and describes AMP_S13 broad-spectrum activity, stability, low cytotoxicity/hemolysis, Galleria mellonella infection testing and a murine MRSA wound-infection model. A separate cohort statement reports six of 14 synthesized candidates inhibiting both E. coli and S. aureus at MIC ≤16 μg/mL; these are distinct denominators/outcomes.",
      "unknown": "The results are laboratory and animal preclinical findings, not human efficacy or safety. The entry makes no claim that the peptide sequence was first disclosed here or is globally novel.",
      "ai_role": "BroadAMP-GPT generated peptide candidates with a transformer and applied deep-learning-guided screening before experimental validation; AMP_S13 is the named lead candidate in the reported assays.",
      "evidence_urls": [
        "https://www.tandfonline.com/doi/full/10.1080/19490976.2025.2523811",
        "https://www.tandfonline.com/doi/pdf/10.1080/19490976.2025.2523811"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-04-oxigraphx-predicted-perovskite-oxygen-vacancies",
    "title": "OxiGraphX predicted vacancy formation energies in high-entropy perovskites",
    "summary": "OxiGraphX, a graph neural network, predicted oxygen-vacancy formation energies for three high-entropy perovskite oxides that the authors synthesized. Experimental measurements reported 29% oxygen vacancies in BSL4B, 27% in SL4B and no detected vacancies in L4B, broadly tracking the predicted and DFT-calculated formation energies. The article contains inconsistent text assigning the relative oxygen-evolution overpotentials; this entry does not make that comparison.",
    "field": "materials",
    "ai_technology": "OxiGraphX graph neural network for oxygen-vacancy formation-energy prediction",
    "organization": "University of North Carolina at Greensboro; University of North Georgia; Kennesaw State University; University of Texas at Austin; Brookhaven National Laboratory",
    "source_url": "https://onlinelibrary.wiley.com/doi/10.1002/smll.202501946",
    "source_type": "paper",
    "publication_date": "2025-04-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/smll.202501946",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher indexed full text, PMC full text and NCBI/PubMed metadata/affiliations. Disputed OER overpotential comparison omitted."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Three high-entropy perovskite oxides were synthesized; oxygen-vacancy measurements reported 29% for BSL4B, 27% for SL4B and undetected for L4B. The article compares these measurements with OxiGraphX predictions and DFT calculations.",
      "unknown": "The source text conflicts on relative oxygen-evolution overpotentials, which are not included here. These tests do not establish device-level or industrial performance.",
      "ai_role": "OxiGraphX predicted oxygen-vacancy formation energies that informed selection of perovskite compositions for synthesis and comparison with measured vacancy content.",
      "evidence_urls": [
        "https://onlinelibrary.wiley.com/doi/10.1002/smll.202501946",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12105445/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-08-knowledge-guided-graph-network-ni-superalloy",
    "title": "Knowledge-guided graph learning designed a printable nickel superalloy",
    "summary": "A knowledge-informed graph attention model with uncertainty quantification selected the KG-AMS nickel superalloy composition and processing conditions for laser directed-energy deposition. The study reports experimental validation with defect area fractions of 0.037%–0.133% and hardness of 391–417 HV across tested parameter sets; it also reports room-temperature tensile strength of 1307 MPa and elongation of 22.2%. These measurements apply to the tested alloy and processing window, and larger builds remain untested. This entry covers KG-AMS only; the paper’s separate aluminum alloy KG-AMAA had already been reported by the same group in 2025.",
    "field": "materials",
    "ai_technology": "Knowledge-informed graph attention network with uncertainty quantification (UQ-KGAT)",
    "organization": "Northeastern University",
    "source_url": "https://www.nature.com/articles/s41467-026-77119-6",
    "source_type": "paper",
    "publication_date": "2026-08-26",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-026-77119-6",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Communications article, linked supplementary affiliation block, and indexed PMC full text checked for date, alloy/model link and tested properties; record is limited to KG-AMS to avoid duplication of KG-AMAA."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports fabricated KG-AMS nickel alloy specimens with defect area fractions of 0.037%–0.133% and hardness of 391–417 HV over the reported processing parameters; it also reports room-temperature tensile strength 1307 MPa and elongation 22.2%.",
      "unknown": "Results are bounded to the specimens and processing conditions in the study; larger builds and independent replication are not established. The separate KG-AMAA aluminum alloy is not counted here because the same group reported it earlier.",
      "ai_role": "UQ-KGAT, a knowledge-informed graph attention network with uncertainty quantification, identified the KG-AMS alloy and its processing region for experimental validation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-026-77119-6",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13612592/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-05-neural-network-ranked-unions-lens-survey",
    "title": "Neural-network-ranked survey found candidate lenses and spectroscopically followed seven",
    "summary": "Two residual neural networks ranked multiband UNIONS galaxy images for strong-lens candidates; researchers visually graded the candidates. The paper lists 1,346 candidates and reports independent Keck/Gemini spectroscopy for seven selected high-quality systems, with background-source redshifts above 2.1. The authors report overlaps with other lens surveys. This entry represents one AI-assisted survey and its selected spectroscopic follow-up, not 1,346 confirmed lenses or seven uniquely first-discovered objects; object-level priority and complete cross-survey overlap remain unresolved.",
    "field": "physics",
    "ai_technology": "Two residual neural networks trained on real lens images and applied to multiband UNIONS imaging",
    "organization": "University of Hawaiʻi at Mānoa; Lawrence Berkeley National Laboratory; University of San Francisco; University of California, Berkeley; Université Paris-Saclay; Laboratoire d’Astrophysique de Marseille; National Research Council Herzberg Astronomy and Astrophysics; University of Waterloo; Perimeter Institute; University of British Columbia",
    "source_url": "https://arxiv.org/abs/2505.05032v1",
    "source_type": "paper",
    "publication_date": "2025-05-08",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.3847/1538-4357/ae381c",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 full paper and history; authors’ primary candidate-catalog deposit and later dataset record for coordinates, overlaps and affiliations."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The paper reports 1,346 human-reviewed lens candidates and spectroscopy of seven selected high-quality systems with background-source redshifts above 2.1.",
      "unknown": "Candidates overlap other surveys; not every candidate is spectroscopically confirmed, and full object-level priority/cross-survey matching is not claimed.",
      "ai_role": "Residual neural networks ranked imaging candidates for human inspection and selection for spectroscopic follow-up.",
      "evidence_urls": [
        "https://arxiv.org/pdf/2505.05032v1",
        "https://zenodo.org/records/15320520",
        "https://zenodo.org/records/20601415"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-10-reac-discovery-catalytic-reactor",
    "title": "AI-guided reactor design improves a specific continuous-flow cycloaddition",
    "summary": "Reac-Discovery combines learned geometry screening with automated experimental optimization of continuous-flow catalytic reactors. The peer-reviewed article reports a Schoen_G structured reactor with space–time yield of 803 g L−1 h−1 for a specific CO₂–epichlorohydrin cycloaddition, compared with 412 in the study’s initial data set. The comparison is study-specific and restricted to the authors’ reactor conditions; it does not establish industrial scale or net carbon removal. The article was published online on 13 October 2025; an earlier preprint described the platform and reaction, so this record attributes the 803 result to the journal version.",
    "field": "chemistry",
    "ai_technology": "Neural geometry screening and machine-learning-guided experimental optimization",
    "organization": "Universitat Jaume I",
    "source_url": "https://www.nature.com/articles/s41467-025-64127-1",
    "source_type": "paper",
    "publication_date": "2025-10-13",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-025-64127-1",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature Communications article metadata and publisher PDF front matter/result checked; result/date/affiliation source checked."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports a measured Schoen_G reactor space-time yield of 803 g L−1 h−1 for CO₂–epichlorohydrin cycloaddition, compared with 412 in the study’s initial data set. The comparison is limited to the article’s experimental conditions.",
      "unknown": "This study-specific result does not establish industrial scalability, net carbon removal or independent replication. The earlier preprint describes the platform and reaction; the 803 result is assigned to the journal article version.",
      "ai_role": "Machine-learning optimization selected process parameters and topological descriptors for the reactor; the article reports the measured cycloaddition result for the optimized reactor.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-025-64127-1.pdf",
        "https://www.nature.com/articles/s41467-025-64127-1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-03-topology-aware-generative-antimicrobial-qacs",
    "title": "Generative design and screening produced experimentally tested antimicrobial QACs",
    "summary": "A topology-aware generative workflow and activity predictor prioritized quaternary ammonium compounds (QACs) for synthesis and testing. Across two workflows the authors report 29 selected compounds characterized and 11 QACs designated as novel, with measured minimum inhibitory concentrations of 1–32 μM against bacterial pathogens. Earlier same-team conference disclosures had already described AI-generated QAC design and synthesis; this record covers the later paper’s bounded tested results and does not claim the first AI-generated QACs or universal chemical novelty. Activity and cytotoxicity remain assay-specific and do not establish clinical use.",
    "field": "chemistry",
    "ai_technology": "Topology-aware QAC generative model with structure/activity prediction and filtering (Q-HERO)",
    "organization": "George Mason University; Villanova University; Emory University; Saint Joseph’s University",
    "source_url": "https://pubs.acs.org/doi/10.1021/acs.jcim.6c00390",
    "source_type": "paper",
    "publication_date": "2026-03-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1021/acs.jcim.6c00390",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ACS indexed full article, publication history, methods/results, named affiliations and indexed supporting-information description; earlier team conference disclosures as identified in the frozen source review."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The article reports synthesis and antimicrobial testing across two workflows, with 29 selected compounds characterized and 11 reported novel QACs with MIC values of 1–32 μM.",
      "unknown": "Prior same-team AI-QAC design and synthesis were described in 2022 and 2024 institutional conference abstracts. The measured activity is limited to the reported laboratory assays and does not establish clinical efficacy.",
      "ai_role": "A topology-aware generative model proposed QAC structures, and learned activity filtering helped prioritize candidates for synthesis and antimicrobial assays.",
      "evidence_urls": [
        "https://pubs.acs.org/doi/10.1021/acs.jcim.6c00390",
        "https://pubs.acs.org/jcisd8/article/66/6/3117/5080435/Topology-Aware-Generation-and-Activity-Based"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-03-ml-selected-nucleoside-hydrogels",
    "title": "Machine learning selected nucleoside derivatives for hydrogel testing",
    "summary": "A machine-learning model trained on 71 previously reported nucleoside derivatives was applied to an external set, from which the authors selected 24 candidates for hydrogel testing. They report gel formation for 20 of the 24 tested molecules and characterize cation-independent hydrogel behavior for 8-aminoguanosine and 8-hydroxyguanosine. A 2023 Sichuan University patent by the same team had already disclosed the model, 12 selected candidates and 10 gelation positives, including these two derivatives. This entry represents one study/campaign and does not claim that the 2024 article was the first public disclosure or that the compounds were novel.",
    "field": "chemistry",
    "ai_technology": "Machine-learning classifier using molecular descriptors to predict nucleoside derivative hydrogel formation",
    "organization": "Sichuan University",
    "source_url": "https://www.nature.com/articles/s41467-024-46866-9",
    "source_type": "paper",
    "publication_date": "2024-03-23",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-024-46866-9",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature article page and PubMed Central indexed full text for model selection and experimental outcomes; PubMed record for date and affiliations; published patent application text for same-team prior disclosure and overlapping candidates."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports that 20 of 24 externally selected nucleoside derivatives formed hydrogels in experimental validation. It also reports cation-independent hydrogel behavior for 8-aminoguanosine and 8-hydroxyguanosine.",
      "unknown": "A 2023 patent application from the same university/team already disclosed the same model, 12 selected derivatives, 10 gelation positives, and included 8-AG and 8-OHG. Do not treat the overlapping paper and patent as independent discoveries or claim first disclosure/compound novelty. Hydrogel behavior does not establish clinical benefit.",
      "ai_role": "The ML model predicted gel-forming ability and selected nucleoside derivatives for external experimental validation.",
      "evidence_urls": [
        "https://www.nature.com/articles/s41467-024-46866-9",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10960799/",
        "https://pubmed.ncbi.nlm.nih.gov/38521777/",
        "https://patents.google.com/patent/CN116741306A/en"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-active-learning-discovers-beta-sheet-pentapeptides",
    "title": "Active learning found unconventional self-assembling pentapeptides",
    "summary": "An experiment-driven active-learning workflow selected pentapeptide sequences for testing of beta-sheet self-assembly. Among 268 synthesized and tested peptides, 96 formed beta-sheet assemblies, including sequences such as ILFSM, LMISI, MITIY, MISIW and WKIYI that conventional propensity rules did not predict. The result concerns measured assembly under the study conditions; it does not establish a functional material application or therapeutic effect. The outcome was already reported in the September 2024 preprint and is dated accordingly.",
    "field": "materials",
    "ai_technology": "AI-driven active learning with machine-learning models that selected peptide sequences for iterative synthesis and testing",
    "organization": "Indian Institute of Technology Madras; Argonne National Laboratory; University of Illinois Chicago",
    "source_url": "https://assets-eu.researchsquare.com/files/rs-5100605/v1_covered_2002f639-5284-4380-a401-67cb063dd68a.pdf",
    "source_type": "paper",
    "publication_date": "2024-09-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1126/sciadv.adt9466",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PMC full article for active-learning workflow and measured assemblies; primary Research Square v1 manuscript for September 2024 version date and named affiliations; primary AAAS article for the peer-reviewed version and reported results."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The v1 reports synthesis/testing of 268 pentapeptides and 96 beta-sheet-forming assemblies, including several unconventional sequences.",
      "unknown": "Assembly findings are bounded to the reported experimental conditions; no application-level performance, therapeutic efficacy or independent replication is established. The specific result was present in the preprint before journal publication.",
      "ai_role": "AI-driven active learning prioritized sequences for iterative synthesis and experimental characterization, focusing in part on cases where model predictions differed from conventional propensity tables.",
      "evidence_urls": [
        "https://assets-eu.researchsquare.com/files/rs-5100605/v1_covered_2002f639-5284-4380-a401-67cb063dd68a.pdf",
        "https://www.nature.com/articles/sciadv.adt9466"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-07-machine-learning-selected-fractionally-doped-perovskites",
    "title": "Machine learning selected fractionally doped perovskites for synthesis",
    "summary": "A function-confined gradient-boosting model trained on 632 records was used to rank fractionally doped perovskite oxides for a solar thermochemical hydrogen-production application. The authors synthesized 11 selected compositions, seven of which the model classified as relevant to that application. This is a composition-discovery and synthesis result; it does not show that all seven were experimentally validated for hydrogen-production performance.",
    "field": "materials",
    "ai_technology": "Function-confined gradient-boosting classifier",
    "organization": "Clemson University",
    "source_url": "https://www.nature.com/articles/s43246-022-00269-9",
    "source_type": "paper",
    "publication_date": "2022-07-06",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s43246-022-00269-9",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary Nature publisher record and full article author manuscript; composition synthesis separated from predicted application relevance."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors report synthesizing 11 model-selected fractionally doped perovskite oxide compositions; seven were classified as relevant to solar thermochemical hydrogen production. The model was evaluated on 632 training records and achieved reported 95.4% accuracy; this classification metric is not a performance result for the 11 synthesized materials.",
      "unknown": "The evidence reviewed does not establish experimentally measured hydrogen-production performance for all seven materials. The entry does not claim each composition is novel outside the authors’ checked data or that model accuracy generalizes beyond the reported dataset.",
      "ai_role": "A function-confined gradient-boosting classifier ranked compositions for synthesis based on target application criteria.",
      "evidence_urls": [
        "https://www.nature.com/articles/s43246-022-00269-9",
        "https://par.nsf.gov/servlets/purl/10356865"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-active-learning-antimicrobial-polymer-p7",
    "title": "Active learning selected antimicrobial polymer P7",
    "summary": "An automated Design-Build-Test-Learn campaign applied machine learning and high-throughput synthesis/characterization to a library of 13,728 polymer combinations. Across four active-learning rounds, the authors selected seven candidates for detailed testing. The leading polymer P7 had a reported MIC of 2 μg/mL and IC20 of 256 μg/mL, and was tested in in-vitro and animal models. These laboratory and preclinical results do not establish clinical efficacy or safety.",
    "field": "materials",
    "ai_technology": "Graph Transformer model with active-learning selection",
    "organization": "Zhejiang University; Institute of Intelligent Chemical Manufacturing and iChemFoundry Platform, Zhejiang University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC12769852/",
    "source_type": "paper",
    "publication_date": "2025-12-17",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.bioactmat.2025.12.020",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PMC indexed full text and NCBI/PubMed metadata/abstract for publication date, affiliations, model, assay outcomes and P7 result."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The study reports seven top candidates from four active-learning rounds. P7 had MIC 2 μg/mL and IC20 256 μg/mL, with in-vitro characterization and animal-model testing reported by the authors.",
      "unknown": "These are laboratory and animal results, not human efficacy or safety. The entry reports the authors’ selection and assay result without a claim of global chemical novelty or clinical readiness.",
      "ai_role": "A graph-transformer prediction model and active-learning loop ranked polymer candidates for automated synthesis and antimicrobial/biocompatibility testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC12769852/",
        "https://pubmed.ncbi.nlm.nih.gov/41502979/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-ml-guided-pet-binding-module-tcbm13-1",
    "title": "Machine learning guided a PET-binding module into a reusable whole-cell catalyst",
    "summary": "An HMM and supervised PET hydrolase model screened sequence candidates and prioritized the PET-binding module tCBM13-1. Experiments found greater PET adsorption and retention than the CBM2 comparator; when displayed with FAST-PETase on E. coli, the system increased PET-film depolymerization by about 43% and retained over 64% of initial activity after ten cycles. These are laboratory assays, not evidence of industrial-scale recycling.",
    "field": "biology",
    "ai_technology": "Profile-HMM homology search combined with supervised PET hydrolase machine-learning model PETML",
    "organization": "Nanjing Tech University; State Key Laboratory of Materials-Oriented Chemical Engineering, Nanjing Tech University",
    "source_url": "https://www.sciencedirect.com/science/article/pii/S0960852426013611",
    "source_type": "paper",
    "publication_date": "2026-07-01",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1016/j.biortech.2026.135279",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher-indexed full text and primary NCBI/PubMed metadata/abstract, including official article date and affiliations."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors experimentally measured tCBM13-1 adsorption and retention on PET and tested whole-cell co-display with FAST-PETase. The publisher abstract reports about 43% greater PET-film depolymerization (2793 μg/(d·cm²)) and over 64% residual activity after ten cycles.",
      "unknown": "Results are laboratory PET assays; they do not establish industrial-scale recycling, environmental deployment, or lifecycle benefit.",
      "ai_role": "A supervised PET hydrolase machine-learning model within an HMM-based sequence-mining workflow prioritized tCBM13-1 from candidate modules for experimental adsorption and whole-cell depolymerization tests.",
      "evidence_urls": [
        "https://www.sciencedirect.com/science/article/pii/S0960852426013611",
        "https://pubmed.ncbi.nlm.nih.gov/42385911/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-12-random-forest-selected-trapum-pulsars",
    "title": "Random Forest selection guided discovery of nine TRAPUM millisecond pulsars",
    "summary": "A Random Forest classifier identified Fermi-LAT sources as pulsar candidates; the team observed 79 selected sources with MeerKAT and reported nine new millisecond pulsars, eight in binaries. The paper reports that all nine candidate signals were subsequently confirmed by additional detections. The authors separately note an independent redetection of one pulsar that had already been discovered elsewhere; it is not counted among the nine new MSPs. This is the 2022 preprint version of a 2023 MNRAS paper.",
    "field": "physics",
    "ai_technology": "Random Forest classification of unassociated Fermi-LAT sources",
    "organization": "Max Planck Institute for Gravitational Physics (Albert Einstein Institute); Leibniz Universität Hannover; Jodrell Bank Centre for Astrophysics, The University of Manchester; Max-Planck-Institut für Radioastronomie; South African Radio Astronomy Observatory",
    "source_url": "https://arxiv.org/html/2212.08528v1",
    "source_type": "paper",
    "publication_date": "2022-12-16",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stac3742",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 full text and submission metadata; journal reference confirmed on primary arXiv record."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The source reports that a Random Forest classified 4FGL unassociated sources, 79 selected sources were observed with MeerKAT, and nine new MSPs were detected and subsequently confirmed by additional detections. Eight were in binary systems.",
      "unknown": "The paper describes nine discoveries as new and distinguishes a separate previously discovered pulsar redetected independently. This review did not conduct a separate external catalog audit of every pulsar designation.",
      "ai_role": "A Random Forest classified unassociated 4FGL sources to choose pulsar candidates for the targeted MeerKAT survey in which the nine new MSPs were discovered.",
      "evidence_urls": [
        "https://arxiv.org/html/2212.08528v1",
        "https://arxiv.org/abs/2212.08528"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-metasynbgc-ai-prioritized-metagenomic-compounds",
    "title": "Deep-learning prioritization of metagenomic products yielded cytotoxic compounds",
    "summary": "MetasynBGC used gene-cluster-family-guided metagenomic mining and a Chemprop graph neural network to prioritize predicted natural-product structures for synthesis. Six selected NRPS-derived compounds showed in-vitro cytotoxic activity across seven cancer cell lines, with reported IC50 values as low as 38.04 μM; compounds D and E showed cell-line-selective activity. These are in-vitro results and do not establish in-vivo efficacy or therapeutic benefit.",
    "field": "biology",
    "ai_technology": "Chemprop graph neural network used to prioritize predicted biosynthetic products by antitumor activity",
    "organization": "Shandong University; Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences; Jining Medical University",
    "source_url": "https://journals.asm.org/doi/10.1128/spectrum.01098-26",
    "source_type": "paper",
    "publication_date": "2026-10-01",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1128/spectrum.01098-26",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ASM article abstract, methods and author affiliation block checked for online date, model role, synthesized compounds and measured cytotoxicity."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Six synthesized NRPS-derived compounds were tested in vitro across seven cancer cell lines; the paper reports IC50 values as low as 38.04 μM and selective activity for compounds D and E.",
      "unknown": "No in-vivo efficacy or clinical therapeutic benefit is established. The reported potency is limited to the tested in-vitro cancer cell-line assays.",
      "ai_role": "A Chemprop graph neural network scored PRISM-predicted biosynthetic product structures for antitumor potential; the authors jointly used predicted scores and structural feasibility to prioritize compounds for synthesis and cytotoxicity testing.",
      "evidence_urls": [
        "https://journals.asm.org/doi/10.1128/spectrum.01098-26"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2023-10-machine-learning-detects-terminal-singularities",
    "title": "Machine learning exposed a structure in terminal-singularity classification",
    "summary": "A neural-network classifier trained on eight-dimensional, Picard-rank-two toric varieties achieved 95% accuracy on a five-million-example test set and enabled rapid generation of a much larger predicted landscape of Q-Fano varieties. The authors report that the model’s classification pattern led them to formulate a new combinatorial criterion for terminal singularities, which they prove mathematically. The learned model suggested the route to the theorem; it did not supply the proof.",
    "field": "mathematics",
    "ai_technology": "Feed-forward neural-network classifier trained on toric-variety weight matrices",
    "organization": "Imperial College London; University of Nottingham",
    "source_url": "https://arxiv.org/abs/2310.20458",
    "source_type": "paper",
    "publication_date": "2023-10-31",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.48550/arXiv.2310.20458",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 metadata and full text for posting date, author affiliations, learned-classifier results and connection to the stated mathematical criterion; linked NeurIPS proceedings identity checked."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors report a 95% classification accuracy on a balanced test split of five million eight-dimensional toric-variety examples and use the model to explore a larger Q-Fano landscape. They formulate and prove a new combinatorial criterion for terminal singularities, motivated by the ML analysis.",
      "unknown": "The 95% is internal held-out classification performance, not independent external validation; the authors state the neural network does not itself perform the theorem’s divisibility-based test or proof.",
      "ai_role": "The trained neural-network classifier revealed a tractable pattern in the classification landscape and directly motivated the authors’ new combinatorial criterion; the authors separately establish its correctness by proof.",
      "evidence_urls": [
        "https://arxiv.org/abs/2310.20458",
        "https://arxiv.org/html/2310.20458v1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-09-neural-operators-microfluidic-waves",
    "title": "Neural operators reveal nonreciprocal interactions behind microfluidic nonlinear waves",
    "summary": "Interpreting a microfluidic droplet experiment movie with a Fourier neural operator and sparse regression yielded a KdV–Burgers-type continuum model containing a dispersive term absent from the prior Burgers description. The authors’ simulations attribute the nonlinear-wave propagation and stabilization to nonreciprocal hydrodynamic interactions. The work analyzes existing experimental data and does not report a new experiment or a general proof for all nonequilibrium crystals.",
    "field": "physics",
    "ai_technology": "Fourier neural operator, interpretable neural-network pipeline and sparse regression",
    "organization": "University of Chicago; Old Dominion University; Université Claude Bernard Lyon 1; ENS de Lyon; CNRS",
    "source_url": "https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.133.107301",
    "source_type": "paper",
    "publication_date": "2024-09-03",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1103/PhysRevLett.133.107301",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary APS abstract and accepted manuscript for the neural-operator and sparse-regression analysis of microfluidic wave data, affiliation block, source data and publication chronology."
    },
    "relevance": {
      "impact": 2,
      "validation": 2,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The authors trained/interpreted a neural-operator network on a microfluidic droplet experiment movie and used sparse regression to derive a KdV–Burgers-type model. The new dispersive term improved the shock-front description relative to Burgers’ equation; a microscopic model attributes nonlinear-wave stabilization/propagation to nonreciprocal hydrodynamic interactions. The source says the model was trained on an existing experimental movie.",
      "unknown": "This is a model-derived interpretation of a specific microfluidic experiment and supporting simulations, not a new experiment or a general proof for all nonequilibrium crystals.",
      "ai_role": "A Fourier neural operator learned latent experimental dynamics from the microfluidic movie; sparse regression extracted the governing equation, and the learned term guided the interpretation of nonreciprocal hydrodynamic interactions.",
      "evidence_urls": [
        "https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.133.107301",
        "https://link.aps.org/accepted/10.1103/PhysRevLett.133.107301"
      ],
      "test_type": "retrospective or benchmark"
    }
  },
  {
    "id": "2023-02-ai-generated-rop-catalyst-surrogates-tested",
    "title": "AI-generated catalyst design led to experimentally tested polymerization catalysts",
    "summary": "A regression-transformer model fine-tuned on historical ring-opening-polymerization data generated candidate catalysts. Researchers selected synthetically accessible, modified surrogates of one model-generated catalyst for laboratory testing. One surrogate, 5c, reached 60% conversion in 32 minutes for monomer 2a and 100% conversion in 45 minutes for monomer 2d, with reported dispersities of 1.06 and 1.34, respectively. The first ChemRxiv v1 (2022-06-27) described computational generation and recovery of a literature-known catalyst. The experimental results appeared in ChemRxiv v2, posted 2023-02-13, and the 2023 peer-reviewed journal article followed on 2023-06-21. The measured tests concern modified surrogates, not the unmodified generated structure.",
    "field": "chemistry",
    "ai_technology": "Regression Transformer generative model fine-tuned on historical polymerization data represented with Chemical Markdown Language",
    "organization": "IBM Research–Almaden; IBM Research–Zurich; ETH Zurich; Arctoris",
    "source_url": "https://chemrxiv.org/engage/chemrxiv/article-details/63e668409da0bc6b33b1870c",
    "source_type": "paper",
    "publication_date": "2023-02-13",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1038/s41467-023-39396-3",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary ChemRxiv v1/v2 records and linked full v1 PDF for version chronology; primary PubMed Central journal full text for model-to-catalyst relationship and measured outcomes; PubMed metadata for journal publication date and affiliations."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The journal reports model-derived catalyst 3g and synthetically accessible modified surrogates 5c/5d. Surrogate 5c produced 60% conversion of monomer 2a in 32 minutes (dispersity 1.06) and 100% conversion of monomer 2d in 45 minutes (dispersity 1.34).",
      "unknown": "The tested catalysts were modified surrogates rather than the unmodified model-generated 3g structure. These are laboratory ring-opening-polymerization outcomes, not process-scale or commercial performance. The earlier 2022 v1 reported model generation and literature-known catalyst recovery; the experimentally validated result is reported in v2 on 2023-02-13 and in the later journal article.",
      "ai_role": "A regression transformer generated catalyst 3g from historical polymerization data; researchers tested synthetically accessible modified surrogates 5c and 5d that preserved its endocyclic guanidine motif.",
      "evidence_urls": [
        "https://chemrxiv.org/engage/chemrxiv/article-details/62b60865e84dd185e60214af",
        "https://chemrxiv.org/engage/chemrxiv/article-details/63e668409da0bc6b33b1870c",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC10284867/",
        "https://pubmed.ncbi.nlm.nih.gov/37344485/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-10-d2ml-guided-ros1-selective-compound-optimization",
    "title": "D2ML guided prospective ROS1 potency and selectivity optimization",
    "summary": "Assay-aware models were used to select 84 new compounds for synthesis and prospective testing in a ROS1 kinase program, with TRKB as an anti-target. The most selective prospective compound had a measured ROS1 IC50 of 14.7 nM and 233-fold selectivity over TRKB, compared with 20.3 nM and 79-fold for the most selective training-set compound. The evidence is biochemical and does not establish therapeutic efficacy or safety.",
    "field": "biology",
    "ai_technology": "D2ML multi-stage deep neural networks trained across direct-to-biology assay data",
    "organization": "Kimia Therapeutics",
    "source_url": "https://arxiv.org/html/2610.41070v1",
    "source_type": "paper",
    "publication_date": "2026-10-29",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://arxiv.org/abs/2610.41070",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 abstract, submission history and full text for date, affiliation, model methods and prospective assay result."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "The authors prospectively synthesized and assayed 84 model-selected compounds. The most selective new compound had ROS1 IC50 14.7 nM and 233-fold selectivity over TRKB; the most selective training compound had 20.3 nM ROS1 IC50 and 79-fold selectivity.",
      "unknown": "This is an in-vitro biochemical discovery result. It does not establish cellular activity, therapeutic efficacy, safety or clinical benefit. The paper compares a historical analog but makes no claim of global chemical novelty.",
      "ai_role": "Assay-aware models using data from multiple direct-to-biology screening stages selected compounds for prospective synthesis and ROS1/TRKB testing.",
      "evidence_urls": [
        "https://arxiv.org/abs/2610.41070",
        "https://arxiv.org/html/2610.41070v1"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2022-08-rnn-guided-nonhemolytic-anticancer-peptides",
    "title": "Machine-learning-guided peptide design yielded non-hemolytic anticancer candidates",
    "summary": "A recurrent neural network generated peptide sequences and classified predicted activity and hemolysis; a genetic-algorithm route also generated analogues. After sequence and physicochemical filters, researchers synthesized and tested 33 peptides. The paper reports 11 active anticancer peptides in its cell assays, four non-hemolytic under its red-cell assay. These in-vitro results do not establish cancer-cell specificity, clinical safety or therapeutic efficacy. A related 2020 modeling paper is separate from this experimental study; no claim of absolute sequence novelty is made.",
    "field": "medicine",
    "ai_technology": "Generative recurrent neural network, genetic algorithm, and recurrent-neural-network activity and hemolysis classifiers",
    "organization": "University of Bern",
    "source_url": "https://chemistry-europe.onlinelibrary.wiley.com/doi/10.1002/cmdc.202200291",
    "source_type": "paper",
    "publication_date": "2022-08-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/cmdc.202200291",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary publisher metadata and primary PubMed abstract/affiliation block; full PMC endpoint returned a browser verification challenge in this review."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The abstract reports that 33 peptides were experimentally evaluated, 11 were active in anticancer cell assays, and four of the active peptides were non-hemolytic in the reported assays.",
      "unknown": "In-vitro activity and red-cell hemolysis do not establish cancer-cell specificity, clinical safety or treatment efficacy. Detailed potency measurements and absolute sequence novelty were not reassessed in this bounded review.",
      "ai_role": "RNN and genetic-algorithm generation with RNN activity and hemolysis classifiers guided peptide selection for synthesis and testing.",
      "evidence_urls": [
        "https://pubmed.ncbi.nlm.nih.gov/35880810/",
        "https://chemistry-europe.onlinelibrary.wiley.com/doi/10.1002/cmdc.202200291"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-08-machine-learning-screened-ulk1-inhibitors",
    "title": "Machine-learning screening identified three compounds with measured ULK1 inhibition",
    "summary": "A Gaussian naive-Bayes classifier using ECFP6 fingerprints ranked commercial compounds from a 13-million-compound library, followed by docking, property filtering and manual selection. Researchers purchased and assayed 25 candidates; three showed micromolar ULK1 inhibition, with reported IC50 values of 37.11 ± 4.41, 7.67 ± 2.82 and 3.15 ± 0.77 μM. The molecules were purchased rather than newly synthesized, and this entry does not claim absolute chemical novelty. The study’s comparator MRT68921 was more potent; the results do not establish cellular activity or clinical utility.",
    "field": "medicine",
    "ai_technology": "Gaussian naive-Bayes classifier on ECFP6 molecular fingerprints (NB-ECFP6)",
    "organization": "The First Affiliated Hospital of Wenzhou Medical University; Oujiang Laboratory; Wenzhou Medical University; Macao Polytechnic University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11485869/",
    "source_type": "paper",
    "publication_date": "2024-08-15",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1080/17568919.2024.2385288",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PMC full article and linked publisher-version PDF, including methods, kinase assay results, date and author affiliation block."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports 25 purchased candidates tested in a biochemical ULK1 kinase assay; three had measured micromolar inhibition with IC50 values 37.11 ± 4.41, 7.67 ± 2.82 and 3.15 ± 0.77 μM.",
      "unknown": "The candidates were purchased, not synthesized. The reported assays do not establish cell activity or clinical effect; the reference MRT68921 was more potent in the paper’s comparison. Absolute structure novelty is not claimed here.",
      "ai_role": "The NB-ECFP6 model screened a large commercial compound set and prioritized candidates for downstream docking and biochemical testing.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11485869/",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11485869/pdf/IFMC_16_2385288.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-07-active-learning-whole-cell-borrelia-antibiotic-screening",
    "title": "Active learning improved whole-cell antibiotic screening and validation",
    "summary": "A closed-loop active-learning campaign for Borrelia burgdorferi whole-cell screening increased the experimental hit rate five-fold, from 0.2% in investigator-selected plates to 1.0% in the AL-guided campaign. The same study separately reports prospective in-silico selection of diverse compounds across bacterial species, with 11.0% experimental validation of predicted hits and 53-fold enrichment over investigator-directed screening; those validated compounds showed the intended narrow-spectrum B. burgdorferi activity. These are assay and screening results, not clinical efficacy.",
    "field": "biology",
    "ai_technology": "Active learning using a directed message-passing neural network and evidential Dirichlet head",
    "organization": "Harvard Medical School; Tufts University; Tufts Lyme Disease Initiative; Texas A&M University; Broad Institute of MIT and Harvard; Tufts University School of Medicine; Stuart B. Levy Center for Integrated Management of Antimicrobial Resistance; Tufts University School of Engineering; Mass General Hospital",
    "source_url": "https://www.biorxiv.org/content/10.64898/2026.07.04.736489v1",
    "source_type": "paper",
    "publication_date": "2026-07-05",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.64898/2026.07.04.736489",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary bioRxiv indexed text, PubMed metadata/abstract and PMC indexed full-text excerpt; direct bioRxiv and PMC page loads were blocked."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Preprint",
      "tested": "In a closed-loop Borrelia burgdorferi campaign, active-learning-guided selection produced a 1.0% experimental screening hit rate versus 0.2% in investigator-selected plates. A separate prospective in-silico selection across bacterial species achieved 11.0% experimental validation of predicted hits (53-fold enrichment over investigator-directed screening); the paper reports intended narrow-spectrum activity for B. burgdorferi.",
      "unknown": "Preprint evidence is not independent replication or clinical testing. The cross-species hit-validation experiment is distinct from the closed-loop Borrelia hit-rate comparison. Direct full page retrieval was blocked; checked material was the primary indexed preprint and archived primary article text.",
      "ai_role": "A directed message-passing neural network with an evidential Dirichlet head supported active-learning compound selection for whole-cell inhibition testing; selected compounds were then experimentally screened.",
      "evidence_urls": [
        "https://www.biorxiv.org/content/10.64898/2026.07.04.736489v1",
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC13344992/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-ensemble-screening-identifies-cdk8-inhibitors",
    "title": "Ensemble machine learning identified CDK8 inhibitor hits",
    "summary": "A six-model ensemble classified 1.6 million commercial molecules and reduced them to a 1,672-compound CDK8-focused screening library. Docking and visual inspection selected 25 candidates for enzymatic assays; six showed CDK8 inhibition. The most potent hit, T479-0984, had a reported IC50 of 99.9 nM. The paper reports low structural similarity (Tanimoto <0.300) to a comparison set of 30 diverse known CDK8 inhibitors. These are assay hits and starting points, not validated drugs or evidence of therapeutic efficacy.",
    "field": "biology",
    "ai_technology": "Six-model ensemble classifier (logistic regression, k-nearest neighbors, support vector classification, random forest, XGBoost and multilayer perceptron)",
    "organization": "Taipei Medical University; The Chinese University of Hong Kong (Shenzhen); Academia Sinica; National Institutes of Health; National Taiwan University",
    "source_url": "https://pmc.ncbi.nlm.nih.gov/articles/PMC11081523/",
    "source_type": "paper",
    "publication_date": "2024-05-09",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1002/pro.5007",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary PMC full text, article metadata, methods/results, author affiliations and compound-level structural comparison reviewed."
    },
    "relevance": {
      "impact": 3,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "Twenty-five candidates from the focused library underwent CDK8 enzymatic testing; six showed inhibition. T479-0984 was reported with IC50 99.9 nM. The paper compared hits against 30 diverse known CDK8 inhibitors and reports Tanimoto similarity below 0.300.",
      "unknown": "The assay establishes in-vitro enzyme inhibition only, not cellular target engagement, animal efficacy, clinical utility, or exhaustive prior patent/publication status for the compounds.",
      "ai_role": "The six-model ensemble filtered a 1.6-million-molecule commercial library into a 1,672-molecule CDK8-focused library; docking and visual inspection selected candidates that were then tested enzymatically.",
      "evidence_urls": [
        "https://pmc.ncbi.nlm.nih.gov/articles/PMC11081523/"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2026-02-random-forest-targeted-trapum-msp-discoveries",
    "title": "Random-forest target selection supported nine millisecond pulsar discoveries",
    "summary": "A random-forest classifier ranked unidentified Fermi-LAT sources, from which the team selected 79 targets for MeerKAT UHF observations. The survey reported nine new millisecond pulsars, all detected in coherent tied-array beams; seven were also confirmed as Fermi-LAT counterparts through gamma-ray pulsations. This entry covers only the nine MSP discoveries. Six slower pulsars were also reported, but five were found in a wider incoherent beam and are not included in the AI-linked result. The authors note earlier optical redback-candidate identification for PSR J1910−5320 and single-pulse reports in the J0657−4657 beam; no claim is made that these were the first observations at all wavelengths.",
    "field": "physics",
    "ai_technology": "Random Forest classifier used to rank unidentified Fermi-LAT sources as pulsar-like candidates",
    "organization": "The University of Manchester; National Astronomical Research Institute of Thailand",
    "source_url": "https://academic.oup.com/mnras/article/547/4/stag406/8501212",
    "source_type": "paper",
    "publication_date": "2026-02-27",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stag406",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 full text for model-based target selection, coherent-beam results, affiliations and prior-source caveats; primary OUP indexed journal record and author-deposited peer-reviewed version for publication date and claim confirmation."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The survey observed 79 random-forest-selected Fermi-LAT sources and reports nine new MSPs, all detected in coherent tied-array beams; seven were confirmed as Fermi-LAT counterparts through gamma-ray pulsations.",
      "unknown": "The nine-MSP record is limited to the coherent-beam result. The paper also reports six slow pulsars, but five were detected in the wider incoherent beam and are not included in this AI-linked claim. The authors note an earlier optical redback-candidate identification for J1910−5320 and single-pulse reports from the J0657−4657 beam; the entry does not claim priority across wavelengths or treat those reports as prior radio-pulsar discoveries.",
      "ai_role": "A random-forest model ranked unassociated Fermi-LAT sources for target selection; radio searches of 79 selected targets yielded the nine coherent-beam MSP discoveries.",
      "evidence_urls": [
        "https://arxiv.org/html/2602.20811v1",
        "https://academic.oup.com/mnras/article/547/4/stag406/8501212",
        "https://pure.mpg.de/rest/items/item_3695468_3/component/file_3702533/content"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-05-pics-assisted-smc-pulsar-survey",
    "title": "PICS-assisted survey reported seven new Small Magellanic Cloud pulsars",
    "summary": "A MeerKAT survey of the Small Magellanic Cloud reported seven new radio pulsars. Its pipeline partially classified folded candidates with PICS, a machine-learning system, and applied score and signal-to-noise thresholds before volunteer visual review and follow-up. The paper does not identify each pulsar’s PICS score, so this is an AI-assisted survey-level result rather than a claim that the model independently discovered all seven sources. The paper also reports pulsar–nebula associations; the later timing paper follows this cohort and is not counted again.",
    "field": "physics",
    "ai_technology": "Pulsar Image-based Classification System (PICS), a machine-learning classifier for folded pulsar candidates",
    "organization": "The University of Manchester; Max Planck Institute for Radio Astronomy; South African Radio Astronomy Observatory; INAF-Osservatorio Astronomico di Cagliari; University of Cape Town; Max Planck Institute for Gravitational Physics; Leibniz University Hannover; Max Planck Institute for Extraterrestrial Physics; Western Sydney University",
    "source_url": "https://arxiv.org/abs/2405.12029",
    "source_type": "paper",
    "publication_date": "2024-05-21",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stae1310",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 methods, results and author front matter; peer-reviewed author-deposited journal PDF for publication date, author affiliations and survey result."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 2
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The survey reports seven new SMC radio pulsars; candidate triage included PICS filtering followed by human review and observational follow-up.",
      "unknown": "The paper does not report per-pulsar PICS scores; it does not establish model-only detection or that PICS was essential for each individual pulsar.",
      "ai_role": "PICS partially classified folded candidates and filtered the candidate set before human visual classification.",
      "evidence_urls": [
        "https://arxiv.org/html/2405.12029v1",
        "https://pure.mpg.de/pubman/item/item_3597707_4/component/file_3597709/stae1310.pdf"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2024-08-pics-assisted-lmc-pulsar-survey",
    "title": "PICS-assisted survey reported seven new Large Magellanic Cloud pulsars",
    "summary": "A MeerKAT survey of its first four Large Magellanic Cloud pointings reported seven new radio pulsars. The search pipeline applied the Pulsar Image-based Classification System (PICS), a deep-neural image classifier, to folded candidates and retained candidates above a PICS score threshold before human visual classification. The AI narrowed the candidate set; volunteers performed the final candidate classification and follow-up observations supplied additional checks. Six pulsars were detected again in follow-up; the incoherent-beam pulsar was not. A later paper reports 12 discoveries from additional pointings and timing follow-up of some Paper I pulsars, which are not additional discoveries in this entry.",
    "field": "physics",
    "ai_technology": "Pulsar Image-based Classification System (PICS), a deep-neural image classifier for folded candidate plots",
    "organization": "University of Cape Town",
    "source_url": "https://academic.oup.com/mnras/article/533/3/2570/7731149",
    "source_type": "paper",
    "publication_date": "2024-08-09",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/stae1917",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary OUP full-text indexed article for candidate selection workflow, score threshold, seven-pulsar result, follow-up status, publication date and first-author affiliation; primary arXiv v1 and OUP Paper II for version and overlap context."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The survey reports seven new LMC radio pulsars. Folded candidates were screened using PICS scores and then visually classified; six of the seven were detected again in follow-up, while the incoherent-beam pulsar was not confirmed in follow-up.",
      "unknown": "PICS filtered candidates but humans classified the candidates as new pulsars. The one incoherent-beam pulsar lacked a follow-up detection in this paper. Later timing work on Paper I objects is not a separate discovery result.",
      "ai_role": "PICS applied a pulsar-like score to folded candidate plots and filtered the survey candidate set before human visual classification; the paper does not attribute final discovery confirmation to the model alone.",
      "evidence_urls": [
        "https://academic.oup.com/mnras/article/533/3/2570/7731149",
        "https://arxiv.org/html/2405.12029v1",
        "https://academic.oup.com/mnras/article/545/4/staf2193/8376644"
      ],
      "test_type": "direct test"
    }
  },
  {
    "id": "2025-12-pics-assisted-trapum-lmc-pulsar-discoveries",
    "title": "PICS-assisted survey reported 12 new Large Magellanic Cloud pulsars",
    "summary": "A MeerKAT search of 18 additional Large Magellanic Cloud pointings reported 12 new radio pulsars. Folded candidates were classified using PICS, a deep-learning image classifier; high-ranking candidates were inspected, and the reported pulsar discoveries were identified from that candidate set. The entry represents the 12 new search detections, not the separate timing solutions for seven pulsars reported in the same paper, which include earlier Paper I discoveries. A journal correction changed the reported folded signal-to-noise for PSR J0537−6957 from 26.1 to 21.7; the publisher states the conclusions were unaffected.",
    "field": "physics",
    "ai_technology": "Pulsar Image-based Classification System (PICS), a deep-learning model for image-based candidate classification",
    "organization": "University of Cape Town",
    "source_url": "https://academic.oup.com/mnras/article/545/4/staf2193/8376644",
    "source_type": "paper",
    "publication_date": "2025-12-10",
    "added_date": "2026-11-17",
    "verification_status": "verified",
    "paper_url": "https://doi.org/10.1093/mnras/staf2193",
    "source_review": {
      "date": "2026-11-17",
      "scope": "Primary arXiv v1 full text for candidate ranking and 12 discoveries; primary journal author-deposited version for publication date; primary OUP correction and Paper I full text for correction and cohort overlap."
    },
    "relevance": {
      "impact": 2,
      "validation": 3,
      "novelty": 3
    },
    "evidence": {
      "publication_status": "Peer-reviewed paper",
      "tested": "The paper reports 12 new LMC radio pulsars in 18 additional pointings. Its pipeline used PICS to classify folded candidates, then inspected high-ranking candidates and identified new pulsars from them.",
      "unknown": "The model ranks candidate plots; researchers perform follow-up and final identification. Seven timing solutions in this paper include earlier Paper I discoveries and are not independent new pulsar discoveries. The publisher correction lowers PSR J0537−6957 folded S/N from 26.1 to 21.7 without changing conclusions.",
      "ai_role": "PICS deep-learning classification was used to rank folded candidates; high-ranking candidates were manually inspected, from which the 12 new pulsars were identified.",
      "evidence_urls": [
        "https://arxiv.org/html/2512.12765v1",
        "https://pure.mpg.de/rest/items/item_3687092_6/component/file_3687093/staf2193.pdf",
        "https://academic.oup.com/mnras/article/547/4/stag518/8540231",
        "https://academic.oup.com/mnras/article/533/3/2570/7731149"
      ],
      "test_type": "direct test"
    }
  }
]
