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.
Technology: PROBY two-stage message-passing neural network pipeline (PROBY-L1 fluorescence classification and PROBY-L2 photophysical-property prediction)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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).
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/41441857/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep learning for nuclear-emulsion analysis
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.riken.jp/en/news_pubs/research_news/pr/2025/20251218_1/index.html
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Random forest and support-vector-machine feature ranking with XGBoost classification
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12809278/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Random forest and XGBoost models with SHAP feature attribution
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Meta-transcriptomic comparison of heat-tolerant and susceptible rice genotypes yielded 409 meta-differentially expressed genes.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/41424172/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 88 of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CycleRFdiffusion backbone design, ProteinMPNN sequence design and HighFold structural prediction/screening
Family RFdiffusion: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The lead TP4 showed weak but reproducible TREM2 binding in spectral-shift, microscale-thermophoresis and surface-plasmon-resonance assays.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/41435973/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Diffusion model guided by random-forest nitrogen-adsorption predictor
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Authors report Co3Mo1 catalyst performance under a defined gas-flow, temperature, support, loading and pretreatment protocol; Co75Mo25 is the corresponding generated candidate.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12814173/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary PMC full text opened; article metadata and candidate/experimental mapping reconciled using continuation audit dated 2026-11-16..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: EGrf random-forest enhancer–gene interaction model, trained on the AstroREG CRISPRi/single-cell RNA-sequencing screen
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/41413662/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Adaptive AI decision interface and AI advisor integrated with autonomous electronic-material fabrication and testing
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s44286-025-00318-3
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AIGFS, AIGEFS, and HGEFS (GraphCast-derived)
Family GraphCast: profile · models page
Publication: Announcement only · Test context: Test context unspecified in annotation. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: not reported
- 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.
- Still 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 sources: https://www.noaa.gov/news-release/noaa-deploys-new-generation-of-ai-driven-global-weather-models
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 5/5 · Validation 1/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Graph Transformer model with active-learning selection
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12769852/ · https://pubmed.ncbi.nlm.nih.gov/41502979/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary PMC indexed full text and NCBI/PubMed metadata/abstract for publication date, affiliations, model, assay outcomes and P7 result..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: GPT-5
Family GPT-5: profile · models page
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- 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.
- Still 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 sources: https://openai.com/index/accelerating-biological-research-in-the-wet-lab/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CellPaint-POSH with CP-DINO-1640 self-supervised embeddings and a vision transformer
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Self-supervised CP-DINO analysis of pooled Cell Painting CRISPR profiles nominated AURKAIP1 among genes with mitochondrial and mTORC1-related phenotypes.
- Still 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 sources: https://www.nature.com/articles/s41467-025-66778-6
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Graph neural networks for crystal defects
Publication: Publication status unspecified in annotation · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Not reported
- Test type: direct test
- 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.
- Still 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 sources: https://energy.sandia.gov/news/sandia-team-uses-ai-ml-models-to-predict-useful-water-splitting-and-hydrogen-generation-materials/
The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 60 of works of the same type, year and subfield (OpenAlex work metrics, 3 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: ProtBERT-BFD sequence embeddings with a support-vector classifier and membrane-interaction, secondary-structure and contact-map predictions
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The paper reports seven experimentally active selected peptides and pore-forming measurements for three broad-spectrum candidates.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12904036/ · https://patents.google.com/patent/CN119591677A/zh · https://doi.org/10.1101/2025.05.17.654650
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review 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..
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Automated Recommendation Tool ensemble of machine-learning regressors
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-025-66304-8_reference.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary publisher article/PDF and institutional preprint metadata reviewed for identity, bounded outcomes, AI role, affiliations and chronology..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: GPT-5.2 Pro
Family GPT-5: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- 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.
- Still 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 sources: https://arxiv.org/abs/2512.10220
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: tFold System; tFold-Ab; tFold-Ag; AlphaFold-Multimer
Family AlphaFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Binding and competition assays on generated antibodies; RBD-antibody neutralization.
- Still 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 sources: https://www.nature.com/articles/s41467-025-67361-9?error=cookies_not_supported · https://www.biorxiv.org/content/10.1101/2024.02.05.578892v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Pulsar Image-based Classification System (PICS), a deep-learning model for image-based candidate classification
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review 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..
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Variational autoencoder, active-learning experimental planner and neural-network equation learner
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-025-66916-0
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CACLENS (Cross-Attention & Contrastive Learning-enabled Enzyme Selection), a multimodal multitask deep-learning enzyme classifier
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The multitask deep-learning enzyme-screening framework CACLENS selected ten candidate enzymes for zearalenone (ZEN) degradation, which were expressed and tested.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12904076/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AlphaFold
Family AlphaFold: profile · models page
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- 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.
- Still 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 sources: https://deepmind.google/blog/engineering-more-resilient-crops-for-a-warming-climate/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: 19 open and closed LLMs (including GPT-4.5, GPT-4o, Grok-3)
Family GPT-4: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.aea3884&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: RFdiffusion2
Family RFdiffusion: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Designed enzymes were expressed and screened experimentally for retroaldolase, cysteine hydrolase and zinc hydrolase activity, with kinetic constants measured for the best variants.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12791007/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: RFdiffusion2 with quantum-chemistry active-site models
Family RFdiffusion: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://doi.org/10.1038/s41586-025-09746-w
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: other. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Riff-Diff (RFdiffusion with rotamer-based motif placement)
Family RFdiffusion: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Designed-enzyme activity and six experimental high-resolution structures.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AI-guided enzyme mutation prediction combining binding-free-energy analysis of active-site substitutions with an experimental FRET-based rapid product-detection platform
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.sciencedirect.com/science/article/pii/S1385894725106992
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 47 of works of the same type, year and subfield (OpenAlex work metrics, 2 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: SMFIF, a stacked ensemble of five selected classifiers built from multi-algorithm feature selection and SHAP-ranked routine laboratory features
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12660861/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: TARSA reinforcement learning; MCMC posterior sampling; learned cytotoxicity models
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: 105 selected peptides were experimentally evaluated; 15 showed activity in the reported breast-cancer-cell assays.
- Still 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 sources: https://www.nature.com/articles/s41467-025-66748-y
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Publisher full article including publication date, model description, methods and experimental results..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AlphaFold
Family AlphaFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 48 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AlphaFold, AlphaFold 3
Family AlphaFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 12 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep neural networks for spatiotemporal seismicity rate forecasting, benchmarked against ETAS
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://link.springer.com/article/10.1186/s40623-025-02241-6
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Algorithmic Iterative Reticular Synthesis (AIRES)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s44160-025-00939-9
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: popEVE (variant-effect prediction model)
Publication: Peer-reviewed paper · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: prospective or independent check
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 4/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: GPT-5
Family GPT-5: profile · models page
Publication: Preprint · Test context: Test context unspecified in annotation. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: not reported
- Tested: Not reported
- Still 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 sources: https://openai.com/index/gpt-5-mathematical-discovery/
The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 1/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Ensemble molecular-descriptor linear classifiers
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Primary article reports eight in-silico hits and selected-stage experimental confirmation; tables qualify cytotoxicity caveats.
- Still 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 sources: https://doi.org/10.3390/pharmaceutics17121513
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: publisher primary full text and tables. Material checked for the specific claims stated in the record..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: RFD2-MI; ProteinMPNN; AlphaFold 3
Family ProteinMPNN: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: The source reports the study-specific result summarized here.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12668006/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary-source abstract and publication metadata reviewed; full text scope is specified in source_checks..
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: EqGPT, a generative transformer using equation tokenization and a generation–evaluation–optimization loop
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: A knowledge-guided generative model trained on mathematical handbook equations generated candidate PDEs, then evaluated and optimized them against data.
- Still 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 sources: https://www.nature.com/articles/s41467-025-65114-2
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: GPT-5 and GPT-5 Pro
Family GPT-5: profile · models page
Publication: Publication status unspecified in annotation · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Not reported
- Test type: direct test
- 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.
- Still 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 sources: https://openai.com/index/accelerating-science-gpt-5/
The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AdaBoost; LightGBM
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The primary source reports the study-specific measured result summarized here.
- Still 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 sources: https://www.nature.com/articles/s41598-025-24713-1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source indexed full text or abstract and publication metadata reviewed; scope stated in source_checks..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CalcAMP machine-learning activity ranking
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The primary article reports in-vitro antimicrobial testing, biofilm assays and a 3D human epidermal infection model for selected parent/retro-inverso peptides.
- Still 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 sources: https://www.mdpi.com/2079-6382/14/11/1172
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary MDPI full article, abstract/results and author-affiliation block reviewed..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Evo 1.5, a long-context generative genomic language model trained on prokaryotic DNA sequences
Family Evo: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41586-025-09749-7.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Bayesian optimization for composition-spread experiments
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01828-7
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Elements-To-Ionics LightGBM
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Publisher full-text indexed article identifies LightGBM screening and specific synthesized conductor conductivity results.
- Still 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 sources: https://doi.org/10.1002/smll.202509918
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for identity, date, model role and bounded reported result..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Seven machine-learning classifiers (including random forest and XGBoost) with LASSO feature selection
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arthritis-research.biomedcentral.com/articles/10.1186/s13075-025-03676-8
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: LASSO feature selection and logistic regression, random forest, support vector machine and XGBoost classifiers applied to untargeted and targeted metabolomics
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://link.springer.com/article/10.1186/s12967-025-07265-w
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 88 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Kosmos (agents with a structured world model)
Family Kosmos: profile · models page
Publication: Preprint · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: prospective or independent check
- 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.
- Still 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 sources: https://advances.edisonscientific.com/research/announcing-kosmos/ · https://arxiv.org/abs/2511.02824
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 4/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep language model semantic-feature decoder with text generation search
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: Video viewing and instructed recall in six participants; 2,180 unique videos in training and comparisons with analyses excluding the canonical language network.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 15 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 5/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AlphaEvolve
Family AlphaEvolve: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arxiv.org/abs/2511.02864
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Polymer informatics models predicting material and recyclability properties
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Primary paper reports the stated assay results.
- Still unknown: Independent replication and exhaustive prior-art review not established.
- AI role: Polymer informatics models predicting material and recyclability properties
Evidence sources: https://arxiv.org/abs/2511.04704
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary publisher full text and bibliographic metadata reviewed for identity, AI role, dates, affiliations and bounded results..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Random Forest machine-learning model; deep-learning comparison model
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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).
- Still 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 sources: https://www.nature.com/articles/s41467-025-64542-4
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 10 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: MAGIC adaptive imaging platform with an XGBoost micronuclei classifier
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41586-025-09632-5
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 28 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: LASSO regression score feature selection with neural-network, support-vector-machine, ridge-regression, logistic-regression and naive-Bayes classifiers
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://link.springer.com/article/10.1038/s44321-025-00325-0
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 86 of works of the same type, year and subfield (OpenAlex work metrics, 9 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: GNEprop deep-learning antibacterial activity predictor
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41587-025-02814-6
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 38 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Explainable neural-network classifiers with LIME feature attribution applied to single-nucleus RNA-sequencing data
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/41120302/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 88 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: DeepSomatic (convolutional neural network)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 59 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.’
Technology: C2S-Scale 27B (Gemma)
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- 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.
- Still 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 sources: https://blog.google/technology/ai/google-gemma-ai-cancer-therapy-discovery/ · https://www.biorxiv.org/content/10.1101/2025.04.14.648850
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep-learning activity prediction and SHAP-guided feature-fragment generation
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The primary source reports the described assays.
- Still unknown: Independent replication and exhaustive sequence-priority review not established.
- AI role: Deep-learning activity prediction and SHAP-guided feature-fragment generation
Evidence sources: https://www.nature.com/articles/s41467-025-64378-y.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary publisher article/indexed full text and bibliographic metadata reviewed for identity, AI role, dates, affiliations, and bounded result..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Neural geometry screening and machine-learning-guided experimental optimization
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-025-64127-1.pdf · https://www.nature.com/articles/s41467-025-64127-1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary Nature Communications article metadata and publisher PDF front matter/result checked; result/date/affiliation source checked..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Light Gradient Boosting Machine and other machine-learning classifiers applied to mass-spectrometry metabolomics
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: Separate models reported AUC 0.97 for influenza A and 0.99 for RSV.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/41065362/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 0 of works of the same type, year and subfield (OpenAlex work metrics, 0 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Gaussian-process regression with Bayesian active learning and a physics-informed average thermoelectric quality-factor descriptor
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://doi.org/10.1002/adma.202515054
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 90 of works of the same type, year and subfield (OpenAlex work metrics, 14 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: OncoSeek AI-empowered blood test combining seven protein tumor markers with individual clinical data
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.nature.com/articles/s41698-025-01105-2.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 78 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Neural-network-inversion satellite soil-moisture product (NNSm), analyzed with Liang–Kleeman information-flow causal inference
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-025-63999-7
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 42 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: OpenEvolve (LLM-driven evolutionary code search, ADRS framework)
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arxiv.org/abs/2510.06189
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CellTransformer
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: The study evaluated the data-driven domains against the Allen Mouse Brain Common Coordinate Framework and reported strong alignment alongside the novel subdivisions.
- Still 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 sources: https://www.nature.com/articles/s41467-025-64259-4
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 16 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Protein Transformer Variational AutoEncoder (ProT-VAE), using ProtT5nv transformer representations and a family-specific variational autoencoder
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The transformer–variational-autoencoder ProT-VAE generated synthetic enzyme sequences that were tested experimentally.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12541330/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 62 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Support-vector-machine metabolite selection and the Renal Cell Carcinoma Artificial Intelligence Detector (RCAID)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/41047317/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 83 of works of the same type, year and subfield (OpenAlex work metrics, 8 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning screening and DiffDock
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 28 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: HuDiff-Ab; HuDiff-Nb
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Newly humanized molecules underwent binding assays; journal abstract reports neutralization.
- Still 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 sources: https://www.nature.com/articles/s42256-025-01120-9?error=cookies_not_supported · https://www.biorxiv.org/content/10.1101/2024.10.22.619416v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: TriVerity on the Myrna instrument, using logistic-regression classifiers on expression measurements for 29 host-response mRNAs
Publication: Peer-reviewed paper · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: prospective or independent check
- 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.
- Still 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 sources: https://www.nature.com/articles/s41591-025-03933-y
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 48 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 4/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep-learning histopathology classifier with Shapley-value interpretation
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/41038216/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 10 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Active learning with sequential Bayesian experimental design, Bayesian model averaging and Gaussian-process surrogate models
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-025-63303-7 · https://pmc.ncbi.nlm.nih.gov/articles/PMC12462480/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary Nature Communications full text as indexed; primary PMC text with methods/results/contributions; institutional full-paper manuscript mirror..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CRESt multimodal model; knowledge-assisted Bayesian optimization
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The source reports the study-specific result summarized here.
- Still 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 sources: https://www.nature.com/articles/s41586-025-09640-5
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary-source abstract and publication metadata reviewed; full text scope is specified in source_checks..
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Embedding-based neural ranking; Gaussian-process Bayesian optimization
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.cambridge.org/engage/chemrxiv/article-details/68cc1a7323be8e43d6f4b2fb
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review 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..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: residual neural network (convolutional classifier for strong lens candidates)
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- 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.
- Still 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 sources: https://arxiv.org/abs/2509.18078
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: DiffSyn generative diffusion model
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: The primary source reports the study-specific result summarized here.
- Still 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 sources: https://arxiv.org/abs/2509.17094
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Physics-informed neural networks
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- Tested: Physics-informed neural networks trained to near machine precision found unstable singularities, and the accuracy was assessed numerically against the governing equations.
- Still 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 sources: https://deepmind.google/blog/discovering-new-solutions-to-century-old-problems-in-fluid-dynamics/ · https://arxiv.org/abs/2509.14185
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 5/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Evo (genome language model)
Family Evo: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- 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.
- Still 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 sources: https://arcinstitute.org/news/hie-king-first-synthetic-phage · https://www.biorxiv.org/content/10.1101/2025.09.12.675911
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 5/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Delphi-2M (GPT-style generative transformer)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: Existing-record predictions in UK Biobank and Danish registries.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12589094/ · https://pmc.ncbi.nlm.nih.gov/articles/PMC12657216/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 161 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: KPPepGen knowledge-aware prompt diffusion model
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.202507457
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 8 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Gauss (Lean autoformalization agent)
Publication: Announcement only · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: prospective or independent check
- Tested: A published Lean formalization of the encoded theorem; this review did not compile it.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: other. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 4/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: OpenAI o3-mini-high reasoning model
Family o1 / o3: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://journals.aps.org/prb/abstract/10.1103/y5vc-3t6q
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 84 of works of the same type, year and subfield (OpenAlex work metrics, 8 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Combination of machine-learning techniques for seismicity structure and b-value analysis (paper abstract does not name a single model)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL116118
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 72 of works of the same type, year and subfield (OpenAlex work metrics, 3 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Ensemble machine-learning models for natural-product RORγt activity and counter-screening, combining graph neural networks and molecular-fingerprint random forests
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC13227160/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 82 of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning binding predictors for MMP-1, MMP-3 and MMP-9
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Primary article reports an experimentally tested selected N-TIMP2 variant with differential MMP binding profile.
- Still 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 sources: https://febs.onlinelibrary.wiley.com/doi/10.1111/febs.70249 · https://pubmed.ncbi.nlm.nih.gov/40912922/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for identity, date, model role and bounded reported result..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep Loop Shaping (reinforcement-learning control)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 16 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Feedforward neural-network multilayer perceptron trained on protein-language-model embeddings and substrate fingerprints
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12559476/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 13 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Retrained machine-learning seismic detection and phase-picking workflow developed at Stanford
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://doi.org/10.1126/science.adw9038
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 32 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Three-round iterative machine-learning pipeline (including VAE/rank model, supervised activity predictors, ESM embeddings and property predictors)
Family ESM / ESMFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Researchers iteratively combined sequence mining, machine-learning activity/property predictors and high-throughput assays to screen natural enzyme homologs.
- Still 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 sources: https://doi.org/10.1021/acscatal.5c03460
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 39 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Gaussian process regression using Magpie composition descriptors
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nims.go.jp/eng/press/2025/09/202509040.html
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 58 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning models applied to data-independent acquisition mass-spectrometry proteomics and biomarker-panel discrimination
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://actaneurocomms.biomedcentral.com/articles/10.1186/s40478-025-02109-6
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: SMI-TED chemical foundation model fine-tuned for electrolyte mixture ionic-conductivity prediction and generative screening
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01774-4 · https://www.nature.com/articles/s41524-025-01804-1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary publisher full text including methods, measured outcomes, affiliations and publication date; primary publisher correction..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: BindCraft (AlphaFold2-based design)
Family AlphaFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-09429-6&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 454 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Physics-informed conditional variational autoencoder (CVAE) and artificial neural network (ANN)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01775-3
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 22 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine learning on molecular features of additives
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 70 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
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
Family ESM / ESMFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12375029/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 27 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: GPT-4b micro
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- 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.
- Still 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 sources: https://openai.com/index/accelerating-life-sciences-research-with-retro-biosciences/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: FlowER (flow matching on bond-electron matrices)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: Specific pyrazole-pathway predictions compared with known patent products and supplemented by calculated intermediate free energies.
- Still 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 sources: https://arxiv.org/html/2502.12979v1 · https://www.nature.com/articles/s41586-025-09426-9?error=cookies_not_supported
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 58 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning classification model trained on Olink plasma proteomic features
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.nature.com/articles/s41591-025-03890-6
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 65 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AlphaCD, a machine-learning model trained on sequence, predicted structure and related protein features
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40825882/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 80 of works of the same type, year and subfield (OpenAlex work metrics, 7 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: XGBoost machine-learning models applied to serum proteomics; DeepPRM-Stellar targeted mass spectrometry supported peptide validation and panel optimization
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40692393/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 42 of works of the same type, year and subfield (OpenAlex work metrics, 2 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Generative AI for molecule design
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.cell.com/cell/abstract/S0092-8674(25)00855-4
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review date: 16 Nov 2026. Review 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..
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 127 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
Directed relationships from this entry; they do not merge catalog counts.
- 2020-02-halicin · extends: 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
- 2023-05-abaucina · extends: 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
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.
Technology: Brain-computer interface and machine learning
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 87 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Support vector machine and other machine-learning models, with recursive feature elimination for panel selection
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://academic.oup.com/brain/article/149/4/1254/8234010
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: OBSTransformer deep-learning phase picker, transfer-learned from EQTransformer and used through SeisBench
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GC012253
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: LAISS machine-learning anomaly alert for transient detection
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://iopscience.iop.org/article/10.3847/1538-4357/adea38
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning binary classifiers and consensus QSAR models trained on experimentally measured chloride-transport activity
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12368578/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 56 of works of the same type, year and subfield (OpenAlex work metrics, 2 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: PepMLM, ESM-2 fine-tuned with target-sequence-conditioned masked language modeling
Family ESM / ESMFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41587-025-02761-2
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 103 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: HydrAMP deep-learning peptide sequence generator and AMPlify deep-learning antimicrobial-activity scorer
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review 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..
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: APEX 1.1 deep learning antimicrobial peptide predictor
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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).
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12408343/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 87 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Symbolic regression applied to high-fidelity turbulence simulation data
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.cambridge.org/core/journals/journal-of-fluid-mechanics/article/abs/dataenabled-discovery-of-specific-and-generalisable-turbulence-closures/4077B2AAC80E5386F5A941ED13FBB451
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 20 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Interpretable decision trees and logistic regression
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: The authors report conventional proofs of criteria motivated by model interpretation.
- Still 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 sources: https://arxiv.org/html/2508.06670v1 · https://link.springer.com/article/10.1007/s40687-026-00620-3
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary arXiv v1 and publisher full text, including result, proof scope, dates and affiliations..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: BEAUT, ESM-2 protein-language-model embeddings plus dense neural-network classifier, augmented using active-site-pocket similarity
Family ESM / ESMFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The authors predicted over 600,000 candidate enzymes and experimentally tested selected predictions; assays identified new bile-acid activities including MABH and ADS.
- Still 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 sources: https://cqb.pku.edu.cn/info/1101/3119.htm
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 58 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning predictive models for zeolite–molecule matching and screening
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Reported experimental findings as stated in article.
- Still unknown: Independent replication and broader priority not established.
- AI role: Machine-learning predictive models for zeolite–molecule matching and screening
Evidence sources: https://www.nature.com/articles/s43588-025-00842-5
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary publisher article reviewed for identity, affiliations, dates, AI role and bounded experimental result..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
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
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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%.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40768600/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 47 of works of the same type, year and subfield (OpenAlex work metrics, 2 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Random-forest vacancy-enthalpy model and stability classifier
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Publisher indexed primary text confirms model screens and synthesis/thermochemical experiments on four compositions.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01726-y
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Physics-tailored neural network for inferring force laws from particle trajectories
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1073/pnas.2505725122&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 20 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AlphaEarth Foundations
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://deepmind.google/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/ · https://arxiv.org/abs/2507.22291
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Multiple AI agents (Virtual Lab)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://doi.org/10.1038/s41586-025-09442-9
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 388 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: LHCb boosted decision tree (BDT) classifier
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://journals.aps.org/prl/abstract/10.1103/r3v2-kmmp
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 88 of works of the same type, year and subfield (OpenAlex work metrics, 10 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Predictive model; permutation-invariant mixture model; generative design
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: The primary source reports the study-specific result summarized here.
- Still 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 sources: https://arxiv.org/html/2504.18728v3
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary article/preprint text and metadata as listed in source_checks..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Genetic algorithm with robotics
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1016/j.matt.2025.102336&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 90 of works of the same type, year and subfield (OpenAlex work metrics, 14 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Bayesian optimization with robotics
Publication: Publication status unspecified in annotation · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Not reported
- Test type: direct test
- 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.
- Still 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 sources: https://foundry.lbl.gov/2025/07/25/accelerating-experiment-design-through-ai-driven-robot/
The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 24 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Aeneas
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://doi.org/10.1038/s41586-025-09292-5
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 37 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: SONAR, an XGBoost machine-learning pipeline using thousands of sequence features to predict protein abundance
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Reporter assays found that CT-rich 3′UTR motifs increased protein output while AG-rich motifs were generally repressive.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12285716/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 66 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: EarthLink multi-agent AI research system based on foundation language models and climate data/tools
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arxiv.org/abs/2507.17311
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Gemini Deep Think
Family Gemini: profile · models page
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- 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.
- Still 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 sources: https://deepmind.google/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep convolutional neural network trained on structured electronic-health-record diagnosis text to estimate Noonan syndrome risk
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: The confirmed cases had pathogenic variants in Noonan-associated genes, including PTPN11 and LZTR1.
- Still 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 sources: https://www.nature.com/articles/s41525-025-00512-5
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 80 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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%).
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
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- 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.
- Still 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 sources: https://arxiv.org/abs/2507.16110
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Unsupervised k-means clustering of DFT-derived ligand descriptors
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Synthesis, X-ray characterization, CO₂ insertion assays and catalytic carboxylation tests reported for selected Ni complexes.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12395408/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary JACS full text, abstract, methods/results, chronology and affiliations..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AEbagging ensemble deep-learning acoustic-emission picker
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2024JB029934
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 76 of works of the same type, year and subfield (OpenAlex work metrics, 4 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: MANet deep convolutional segmentation network with attention, transfer learning and two-stage fine-tuning
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- 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.
- Still 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 sources: https://arxiv.org/pdf/2507.13420
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: EchoNext (deep learning model on 12-lead ECG)
Publication: Peer-reviewed paper · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: prospective or independent check
- Tested: Internal/external diagnostic validation, a controlled cardiologist comparison and prospective clinical-trial screening of patients without previous cardiac imaging.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41586-025-09227-0%22&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 163 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 5/5 · Validation 4/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Boosted decision tree event classifier; neural-network particle-identification variables
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41586-025-09119-3
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 94 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Random forest; gradient boosting; neural-network regression
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Primary full text reports synthesis and characterization, magnetic ordering and magnetocaloric measurements for ML-selected compositions.
- Still 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 sources: https://backend.orbit.dtu.dk/ws/portalfiles/portal/408621365/1-s2.0-S1359645425005695-main.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Author-institution repository PDF full text and publisher metadata; composition-level previous-work comparison is bounded, not exhaustive..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: APEX deep-learning model for bacterial strain-specific minimum inhibitory concentration prediction
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-025-60051-6.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 61 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Progressive learning with DFT-derived adsorption-energy features and machine learning
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://onlinelibrary.wiley.com/doi/10.1002/anie.202510965
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary publisher abstract, first-publication date, and author-affiliation metadata; full article and supporting information not reviewed..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
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
Publication: Peer-reviewed paper · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: prospective or independent check
- Tested: Across an international dataset of 8,461 slides it achieved mean AUC 0.847 internally and 0.870 in external cohorts.
- Still 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 sources: https://www.nature.com/articles/s41591-025-03780-x
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 106 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 4/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: PhaseNet deep neural-network seismic phase picker in the LOC-FLOW workflow
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025GL114675
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 87 of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Extremely Randomized Trees cross-spectral response prediction model
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-025-60499-6
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 16 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: VenusMine; ProstT5; ESM2
Family ESM / ESMFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: 34 selected proteins were tested; the source reports PET-hydrolyzing activity for 14 and characterizes KbPETase experimentally.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12228687/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Indexed primary article text including title, methods, results and publication metadata; direct PMC open triggered CAPTCHA..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Daisy image-analysis models and offline reinforcement-learning synthesis planner trained on historical microstructure data
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01701-7
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary publisher full text as indexed, including article date, methods, results, author contributions and affiliations..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Centaur (Llama 3.1 70B fine-tuned on Psych-101)
Family Llama: profile · models page
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-09215-4&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 227 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Random-forest classifiers and regressors; chemical similarity filtering
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Primary full text reports 17 experimentally tested compounds and 11 active; compound-specific IC50/MIC values are stated with different fluconazole comparators.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12320878/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary full-text article on PMC indexed text and methods/results read; online date, title and DOI verified..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine learning and LLM-guided variant design with biofoundry automation
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Enzyme activity and purified-variant kinetic measurements in two campaigns.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12215622/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 140 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Multi-omics integration and machine-learning predictive models for feature selection and classification
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40598791/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 12 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Transformer-based peptide generation with deep-learning-guided screening
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.tandfonline.com/doi/full/10.1080/19490976.2025.2523811 · https://www.tandfonline.com/doi/pdf/10.1080/19490976.2025.2523811
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary publisher full article/PDF, publisher publication history and NCBI/PubMed metadata/abstract; first-public preprint and sequence priority are not claimed..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AlphaGenome
Family AlphaGenome: profile · models page
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://deepmind.google/blog/alphagenome-ai-for-better-understanding-the-genome/ · https://www.biorxiv.org/content/10.1101/2025.06.25.661532
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Bayesian optimization (COMBO) and convolutional neural network coupled to molecular-dynamics simulations
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01678-3
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 17 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: MEGNet graph neural network with five-model ensemble and iterative active learning
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01682-7
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 78 of works of the same type, year and subfield (OpenAlex work metrics, 8 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Skala (neural network for DFT)
Family Skala: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep neural network speech decoder for intracortical recordings
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-09127-3&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 123 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 5/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: FerroAI deep learning model; NLP-extracted phase-transition dataset
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The primary article reports the study-specific material and measurement summarized here.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01778-0 · https://arxiv.org/abs/2506.10970
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary publisher indexed full text/abstract and version metadata reviewed..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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%.
Technology: Transfer-learned ULMFiT chemical language model (FnG) for ligand generation and a 30-model deep-learning ensemble (EnP) for enantioselectivity prediction
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubs.rsc.org/en/content/articlepdf/2025/sc/d5sc01098e
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 65 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: PGCGM generative Wasserstein GAN for ternary crystal structures; ALIGNN graph neural network for decomposition-enthalpy stability screening
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubs.rsc.org/en/content/articlehtml/2025/ra/d5ra00427f
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 65 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Multiscale topological learning; persistent homology; affinity propagation clustering; ab initio molecular dynamics
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12186519/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary PMC full text and publication metadata; article identity, methods, bounded computational outcomes and affiliation block..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Enoch (palaeographic dating model)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: The Enoch dates for 135 manuscripts, following new radiocarbon dating, were compared with palaeographers' judgments, with about 79% agreement.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 54 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Pharma.AI (generative AI for targets and molecules)
Publication: Peer-reviewed paper · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: prospective or independent check
- 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.
- Still 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 sources: https://doi.org/10.1038/s41591-025-03743-2
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 239 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 4/5 · Novelty 5/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning activity map; theoretical descriptor modeling
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12134119/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: DiET convolutional neural network and GrASP graph-based station association, with hybrid envelope-correlation/energy-ratio hypocenter location
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025JB031348
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 17 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
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
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40447567/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 10 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: BTSbot machine-learning transient classifier integrated with the BTSbot-nearby autonomous follow-up program
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://iopscience.iop.org/article/10.3847/1538-4357/adcf1e
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 16 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Active-learning-guided relative-binding-free-energy (AL-RBFE) workflow using molecular-dynamics thermodynamic-integration calculations and an ML affinity-prediction model
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12152950/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 79 of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning biomarker scoring matrix combining random forest and OPLS-DA
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40436140/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 85 of works of the same type, year and subfield (OpenAlex work metrics, 4 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: SISSO symbolic regression with bootstrap and Monte Carlo feature-dropout ensembles in an active-learning loop
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.nature.com/articles/s41524-025-01596-4
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 34 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Samudra UNet-style ocean emulator
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arxiv.org/abs/2412.03795
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 52 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Robin (Crow, Falcon, and Finch agents)
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: Human researchers lab-tested ten candidates proposed by Robin, and a ROCK inhibitor was found to increase phagocytosis by retinal pigment epithelial cells.
- Still 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 sources: https://www.futurehouse.org/research/demonstrating-end-to-end-scientific-discovery-with-robin-a-multi-agent-system · https://arxiv.org/abs/2505.13400
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: LASSO feature selection with logistic regression, using five-fold cross-validation and age/sex covariates
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://pure.amsterdamumc.nl/ws/portalfiles/portal/140778915/Large-scale-plasma-proteomic-profiling-unveils-diagnostic-biomarkers-and-pathways-for-alzheimers-disease.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 63 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Microsoft Discovery (agentic orchestration and high-performance computing)
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- Tested: A prototype PFAS-free coolant was synthesized, and some of its primary properties were tested and matched the AI predictions.
- Still 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 sources: https://azure.microsoft.com/en-us/blog/transforming-rd-with-agentic-ai-introducing-microsoft-discovery/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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%.
Technology: AlphaEvolve (Gemini Flash and Pro)
Family Gemini: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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%.
- Still 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 sources: https://deepmind.google/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/ · https://arxiv.org/abs/2506.13131
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: scDORI; deep learning for enhancer-linked gene-regulatory network inference
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Patient-derived cell perturbations and mouse xenograft experiments tested inferred regulators.
- Still 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 sources: https://www.nature.com/articles/s41586-026-11177-0 · https://www.biorxiv.org/content/10.1101/2025.05.13.653733v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Predictive Biomarker Modeling Framework (PBMF), an ensemble of neural networks using a contrastive treatment/control objective; knowledge-distilled decision tree
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://ellykipkogei.io/assets/pdf/1-s2.0-S1535610825001308-main.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 65 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning mohometry regression model with feature-importance filtering
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024JB030404
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 88 of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning enhancer design model developed by Lars Velten's group
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.crg.eu/en/news/ai-designed-dna-controls-genes-healthy-mammalian-cells-first-time
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 37 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: PheWASNet, a multi-task deep-learning model based on a one-dimensional EfficientNet architecture
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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).
- Still 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 sources: https://academic.oup.com/ehjdh/article/6/4/595/8126967
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: 650-million-parameter ESM-2 protein language model aligned by reinforcement learning from experimental feedback using a learned sequence-fluorescence reward model
Family ESM / ESMFold: profile · models page
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review 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..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Two residual neural networks trained on real lens images and applied to multiband UNIONS imaging
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- 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.
- Still 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 sources: https://arxiv.org/pdf/2505.05032v1 · https://zenodo.org/records/15320520 · https://zenodo.org/records/20601415
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary arXiv v1 full paper and history; authors’ primary candidate-catalog deposit and later dataset record for coordinates, overlaps and affiliations..
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Previously developed machine-learning seismic phase picker, fine-tuned on similar waveforms, with two-step event association
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arxiv.org/abs/2505.03496
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Chemical Space Neural Networks (CSNN), neighborhood-to-prediction model using network homophily and graph neural networks with labels as features
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12048553/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 85 of works of the same type, year and subfield (OpenAlex work metrics, 8 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: LIMITS (Lead Informed Machine Interrogation of Therapeutic Sequences), using random-forest and feed-forward neural-network predictors
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40310992/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 7 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: SKYNET deep-learning P/S phase picker with GENIE graph-neural-network phase association
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://seismica.library.mcgill.ca/article/view/1431
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 15 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Random forests; XGBoost; deep neural networks; graph convolutional networks
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The primary source reports the study-specific result summarized here.
- Still 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 sources: https://www.nature.com/articles/s41467-025-56818-6
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Three-stage CNN characterization pipeline (metal-poor galaxy classifier, XMP classifier, and N2-index predictor)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://academic.oup.com/mnras/article/540/1/128/8121403
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AlphaFold structure prediction, followed by structural comparison of the predicted PHGDH fold
Family AlphaFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12204802/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 20 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: RFdiffusion and PLACER
Family RFdiffusion: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Enzyme activity assays and crystallographic comparison of designed catalysts.
- Still unknown: Not assessed: complete design success rates and supplementary data.
- AI role: RFdiffusion generated backbones; PLACER assessed catalytic-state geometry.
Evidence sources: https://pubmed.ncbi.nlm.nih.gov/39946508/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 275 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: MOFGen (multi-agent LLM system with diffusion model)
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: The abstract says five AI-designed MOFs were synthesized through high-throughput experiments; the validation is described only generally.
- Still 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 sources: https://arxiv.org/abs/2504.14110
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Automated Recommendation Tool for active-learning medium optimization
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s42003-025-08039-2.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary publisher article/PDF and institutional preprint metadata reviewed for identity, bounded outcomes, AI role, affiliations and chronology..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: PCA; K-means clustering; recommendation engine
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The primary source reports the study-specific result summarized here.
- Still 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 sources: https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/67bc9edbfa469535b9bd0d65/original/TbIr3_final.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Aardvark Weather
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 229 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 5/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Urania (continuous optimization with ML-inspired search)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Simulated performance checks of newly designed interferometer topologies under experimental constraints.
- Still unknown: Not assessed: detailed noise and hardware feasibility assumptions in the full article.
- AI role: Machine-learning-inspired continuous optimization searched detector configurations.
Evidence sources: https://journals.aps.org/prx/abstract/10.1103/PhysRevX.15.021012 · https://arxiv.org/abs/2312.04258
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 88 of works of the same type, year and subfield (OpenAlex work metrics, 11 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Generative and predictive molecule models
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Seven generated non-fullerene acceptor molecules were synthesized and tested, with power conversion efficiencies close to the model predictions.
- Still 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 sources: https://research.a-star.edu.sg/articles/highlights/man-machine-team-up-boosts-solar-materials-search/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 67 of works of the same type, year and subfield (OpenAlex work metrics, 9 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AI neuron segmentation
Publication: Peer-reviewed paper · Test context: Test context unspecified in annotation. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: not reported
- Tested: Not reported
- Still 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 sources: 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
The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 367 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 5/5 · Validation 1/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep subdomain adaptation network (DSAN) with low-reflectance artifact detection (LRAD)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://acp.copernicus.org/articles/25/4035/2025/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 88 of works of the same type, year and subfield (OpenAlex work metrics, 9 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AMIE (Articulate Medical Intelligence Explorer), a large language model agent for diagnostic dialogue
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-08866-7&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 824 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: OxiGraphX graph neural network for oxygen-vacancy formation-energy prediction
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://onlinelibrary.wiley.com/doi/10.1002/smll.202501946 · https://pmc.ncbi.nlm.nih.gov/articles/PMC12105445/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary publisher indexed full text, PMC full text and NCBI/PubMed metadata/affiliations. Disputed OER overpotential comparison omitted..
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: ESM-Ezy, using ESM-1b protein-language-model embeddings and semantic-space similarity
Family ESM / ESMFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40188191/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 38 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep-learning peptide activity scoring and sequence-optimization algorithms
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Deep-learning-guided sequence optimization produced P2.2, which showed increased antibacterial activity, membrane effects and reduced hemolysis in assays.
- Still 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 sources: https://homepage.zjut.edu.cn/_upload/article/files/51/c3/6b78dffc40f7932148ad26cbf65b/e1d2421e-c600-4dac-82a5-bc29fa545929.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 12 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Supervised/unsupervised materials models, active learning and domain adaptation
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still unknown: Independent replication and exhaustive prior-art review are not established.
- AI role: Supervised/unsupervised materials models, active learning and domain adaptation
Evidence sources: https://pubmed.ncbi.nlm.nih.gov/40184453/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for study identity, AI operation, dates, affiliations and the bounded result..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Mask R-CNN; Spatial-ID; nnUNet
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Spatial transcriptomics and retrograde tracing directly supported the cell-type/projection-zone relationship within the discovery study.
- Still 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 sources: https://hal.science/hal-05210393v1/file/Lei%20et%20al%202025%20Cell.pdf · https://pubmed.ncbi.nlm.nih.gov/40185102/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: RED-PAN deep-learning seismic phase picker
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://link.springer.com/article/10.1007/s44195-025-00093-x
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 84 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: DeepLabV3+ semantic segmentation
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1371/journal.pone.0320452&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 9 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Ensemble machine-learning models trained on molecular-dynamics data
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: Two nominated vitrimers were synthesized and measured; the paper reports higher Tg than the selected comparator set.
- Still 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 sources: https://arxiv.org/abs/2503.20956
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CataPro
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The team experimentally identified Sphingobium sp.
- Still 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 sources: https://doi.org/10.1038/s41467-025-58038-4
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 140 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Image-classification models
Publication: Preprint · Test context: Test context unspecified in annotation. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: not reported
- 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.
- Still 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 sources: https://www.esa.int/ESA_Multimedia/Images/2025/03/Strong_gravitational_lenses_captured_by_Euclid · https://arxiv.org/abs/2503.15324
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 1/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: MEDUSA Search, an isotope-distribution-driven HRMS search engine with machine-learning ion-threshold regression and false-positive classification
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-025-56905-8.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 28 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: The AI Scientist-v2
Publication: Announcement only · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: prospective or independent check
- 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.
- Still 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 sources: https://sakana.ai/ai-scientist-first-publication/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 2/5 · Validation 4/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
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)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://doi.org/10.1016/j.cels.2025.101236
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 45 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Enzyme-specific ML-hybrid substrate prediction models trained from high-throughput peptide-array assays
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s42004-025-01717-6.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 77 of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Basenji2-long and Basenji2-3K-NAM sequence-to-expression deep-learning models with editing-plasticity analysis
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://link.springer.com/article/10.1186/s13059-025-03516-7
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 23 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: WenHai deep neural network ocean forecasting model
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-025-57389-2&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 87 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AI-driven structural modeling (Kozakov laboratory; specific model not named in institutional announcement)
Publication: Peer-reviewed paper · Test context: Test context unspecified in annotation. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: not reported
- Tested: The map adds experimentally grounded candidate links between bacterial metabolites and proteins involved in core cellular processes.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/39862855/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 14 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 1/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Convolutional neural networks combined with satellite spectral data
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41586-025-08663-2&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 55 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: LogiRx logic-based mechanistic machine learning
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Primary abstract reports experimental pathway validation, cardiomyocyte and mouse outcomes, and patient database associations.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/40035765/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary PubMed metadata/abstract and publisher full-text PDF indexed result; retrospective human association caveat preserved..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: XGBoost regression; crystal-chemical descriptors
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: Ca2SrSc6O12:Ce3+ was synthesized and its excitation/emission measured.
- Still 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 sources: https://arxiv.org/abs/2502.18859
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for identity, publication date, AI workflow and the bounded experimental result stated..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AIFS (machine-learning model)
Family AIFS (ECMWF): profile · models page
Publication: Announcement only · Test context: Test context unspecified in annotation. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: not reported
- Tested: Not reported
- Still 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 sources: https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational
The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 5/5 · Validation 1/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning electrolyte property and synthesizability screening
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Primary paper and PubMed abstract support virtual screening, two synthesized candidates and cell testing.
- Still 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 sources: https://doi.org/10.1002/anie.202424493
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for identity, date, model role and bounded reported result..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CorrDiff (diffusion model)
Family CorrDiff: profile · models page
Publication: Peer-reviewed paper · Test context: Test context unspecified in annotation. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: not reported
- Tested: Not reported
- Still 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 sources: https://blogs.nvidia.com/blog/earth-2-ai-high-resolution-forecasts/
The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 274 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 1/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Mal-ID (MAchine Learning for Immunological Diagnosis), an ensemble using repertoire feature models and ESM-2 protein-language-model embeddings
Family ESM / ESMFold: profile · models page
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: The authors also evaluated classification on independent COVID-19 cohorts.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/39977494/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 132 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Artera multimodal artificial-intelligence (MMAI) prostate biomarker combining deep-learning features from digitized biopsy histopathology with age, PSA and clinical T-stage
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: A previously developed multimodal AI score was retrospectively applied to archived diagnostic biopsies and follow-up from a prospectively collected Swedish prostatectomy cohort.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC11995000/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 25 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Autoencoder and deep embedded clustering
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The study applies clustering to seismic records and follows with source-location and particle-motion analyses.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary PubMed metadata/abstract plus full primary article PDF from institutional repository; methods, conclusion and affiliations..
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.’
Technology: BioEmu-1
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.microsoft.com/en-us/research/blog/exploring-the-structural-changes-driving-protein-function-with-bioemu-1/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: FragFold, using AlphaFold
Family AlphaFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The researchers experimentally confirmed more than half of its binding or inhibition predictions, including cases without prior structural interaction data.
- Still 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 sources: https://news.mit.edu/2025/ai-system-fragfold-predicts-protein-fragments-0220
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 15 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AI co-scientist (Gemini 2.0, multi-agent)
Family AI Co-Scientist: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://doi.org/10.1038/s41586-026-10644-y
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Evo 2 (genome language model)
Family Evo: profile · models page
Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arcinstitute.org/news/evo2
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Editorial assessment: Impact 4/5 · Validation 2/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Variational autoencoder and random-forest species delimitation
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: Molecular analyses, microscopy and multiple delimitation approaches supported the formal description.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC11836703/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary full-text species-delimitation Methods, taxonomic Results, date and affiliations; supplements not reviewed..
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Importance-guided Bayesian optimization with automated synthesis and characterization
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Automated processing experiments on conductive polymer films; recipes yielded reported mean conductivity exceeding 4,500 S/cm.
- Still 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 sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A%2210.1038%2Fs41467-024-55655-3%22&resultType=core&format=json
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 87 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: PC-corr machine-learning-based discriminative network analysis
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC11832173/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 53 of works of the same type, year and subfield (OpenAlex work metrics, 2 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Ensemble of five convolutional neural networks
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: Primary v1 and journal paper report spectroscopically supported lens with redshifts and lens modeling.
- Still 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 sources: https://arxiv.org/abs/2502.09802v1
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary source checked for identity, date, model role and bounded reported result..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Iterative adaptive design with SOAP descriptors and a KPCovR surrogate model
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: 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
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review 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..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: OptPDE coefficient optimization with the CQFinder conserved-quantity search
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: Automatic differentiation optimized partial-differential-equation coefficients to increase the number of conserved quantities.
- Still 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 sources: https://journals.aps.org/pre/abstract/10.1103/PhysRevE.111.025303
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 85 of works of the same type, year and subfield (OpenAlex work metrics, 10 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Radio U-Net, a machine-learning detector for diffuse radio emission
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arxiv.org/abs/2502.04823
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 80 of works of the same type, year and subfield (OpenAlex work metrics, 7 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Machine-learning-accelerated high-throughput screening using descriptors for photosensitization, electron transfer, and catalysis
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41929-025-01291-z · https://chem.ahnu.edu.cn/info/1028/28946.htm
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 17 Nov 2026. Review scope: Primary Nature Catalysis indexed article text, abstract, extended-data descriptions, metadata and affiliation block; institutional research announcement checked as corroboration..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: ProtGPS protein language model and generative design algorithm
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: The paper applies the model to predict disease-mutation-associated localization shifts and reports cellular testing of 20 examples.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC12034300/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 109 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: LOC-FLOW automated earthquake-detection and location workflow using PhaseNet deep-learning phase picking
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.sciencedirect.com/science/article/abs/pii/S0040195124004037
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 66 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Latent diffusion model coupled to molecular dynamics and peptide filtering/ranking
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC11797553/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 103 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Ensembles of LSBG Detection Transformers (LSBG DETR) and LSBG Vision Transformers (LSBG ViT)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://arxiv.org/abs/2502.03142
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 77 of works of the same type, year and subfield (OpenAlex work metrics, 6 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: CHANCE, a supervised gradient-boosted-tree model integrating coding/non-coding variants, protein–protein interaction networks, drug targets and drug-response data
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC11794852/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 79 of works of the same type, year and subfield (OpenAlex work metrics, 5 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Random-forest classifier
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Primary article reports tested CEP-37440/AZD-1480 antiplasmodial activity.
- Still 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 sources: https://doi.org/10.1186/s13065-025-01395-4
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: publisher primary full text. Material checked for the specific claims stated in the record..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Deep graph neural networks for reaction outcome prediction; multi-objective molecular optimization
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: Fourteen selected synthesized compounds were reported with subnanomolar MAGL inhibition and co-crystal structures for three ligands.
- Still 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 sources: https://www.cambridge.org/engage/chemrxiv/article-details/6798ae2881d2151a024590d5
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary ChemRxiv v2 abstract and version metadata reviewed; full manuscript and supplements not inspected..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Chemistry42 and hybrid quantum-classical models
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Fifteen candidates from the hybrid process were tested in the lab, and two strongly targeted mutated KRAS variants in living cells.
- Still 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 sources: https://www.utoronto.ca/news/ai-quantum-computing-used-target-undruggable-cancer-protein
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 125 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: EVOLVEpro (protein language model with active learning)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: Experimental engineering of proteins for RNA production, genome editing and antibody binding.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/39571002/ · https://news.massgeneralbrigham.org/en/mgb-bidmc-unveil-ai-protein-engineer
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 264 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Augmented ridge-regression ML models within a cell-free sequence-function mapping and enzyme-engineering workflow
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- Tested: A cell-free design-build-test-learn platform generated sequence-function data from 1,217 McbA variants across 10,953 reactions.
- Still 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 sources: https://www.nature.com/articles/s41467-024-55399-0.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 153 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: MaSIF (deep learning on molecular surfaces)
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://actu.epfl.ch/news/a-deep-learning-pipeline-for-controlling-protein-3/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 93 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Org-Mol 3D molecular transformer
Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Preprint
- Test type: direct test
- Tested: The primary source reports the study-specific result summarized here.
- Still 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 sources: https://arxiv.org/abs/2501.09896
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review date: 16 Nov 2026. Review scope: Primary-source indexed article or abstract and bibliographic metadata reviewed; complete source scope is given in source_checks..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Explainable deep-learning antimicrobial-peptide potency model with gradient-based virtual sequence evolution
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41564-024-01907-3 · https://pubmed.ncbi.nlm.nih.gov/39825096/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 17 Nov 2026. Review scope: Primary Nature Microbiology indexed article text and primary PubMed abstract/metadata/affiliations; conclusions limited to reported preclinical assays..
Editorial assessment: Impact 4/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: MatterGen (diffusion model)
Family MatterGen: profile · models page
Publication: Announcement only · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Announcement only
- Test type: direct test
- 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%.
- Still 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 sources: https://www.microsoft.com/en-us/research/blog/mattergen-a-new-paradigm-of-materials-design-with-generative-ai/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: blog. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 933 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AI protein design (RFdiffusion)
Family RFdiffusion: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.ipd.uw.edu/2025/01/neutralizing-deadly-snake-toxins/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 196 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 4/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: ALDE (Active Learning-assisted Directed Evolution), using a bootstrapped deep-neural-network ensemble and Thompson sampling
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC11739421/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 186 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Graph neural network (GNN) trained on BESIII electromagnetic-calorimeter shower patterns
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-024-55042-y
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 24 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Gaussian-process models; multi-objective Bayesian active learning
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41524-024-01480-7
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review date: 16 Nov 2026. Review scope: Primary npj Computational Materials indexed full text, publication history, experimental characterization and author affiliations..
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: GET (General Expression Transformer) foundation model for cell-type-specific regulatory grammar; AlphaFold2/AlphaFold Multimer for structural hypothesis assessment
Family AlphaFold: profile · models page
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://www.nature.com/articles/s41586-024-08391-z.pdf
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 156 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Semi-supervised anomaly detection, extreme gradient boosting, and random-forest classifiers applied to Gaia DR3 astrometric-orbit solutions
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: A machine-learning search of Gaia DR3 two-body astrometric orbits flagged G 15-6 as a likely low-mass companion system.
- Still 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 sources: https://arxiv.org/abs/2404.09350
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 84 of works of the same type, year and subfield (OpenAlex work metrics, 9 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: AI support for mammogram reading (Vara)
Publication: Publication status unspecified in annotation · Test context: prospective or independent check. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Not reported
- Test type: prospective or independent check
- 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.
- Still 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 sources: https://www.uni-luebeck.de/aktuelles/pressemitteilung/artikel/ki-verbessert-brustkrebserkennung.html
The legacy label “Not reported” does not distinguish information absent from reviewed material from information not assessed. It is not a scientific finding of absence.
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: press release. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 280 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 4/5 · Validation 4/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Protein language model embeddings coupled with regression in the SeekRank enzyme-prioritization workflow
Publication: Peer-reviewed paper · Test context: direct test. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: direct test
- 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.
- Still 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 sources: https://pubmed.ncbi.nlm.nih.gov/39777462/
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: at percentile 82 of works of the same type, year and subfield (OpenAlex work metrics, 14 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: GlaViTU (Glacier-VisionTransformer-U-Net)
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- Tested: Retrospective satellite-image mapping and regional glacier-area estimates at two periods; model-generalization and expert-outline comparisons.
- Still 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 sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC11695715/ · https://www.ebi.ac.uk/europepmc/webservices/rest/PMC11695715/supplementaryFiles
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: not linked in this catalog; this does not mean no paper exists.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 1% of works of the same type, year and subfield (OpenAlex work metrics, 61 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.
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.
Technology: Previously validated deep-learning models for MRI segmentation and bone marrow fat-fraction measurement
Publication: Peer-reviewed paper · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.
Result, AI contribution and limits
Evidence context (descriptions follow the cited sources; test type is a catalog classification):
- Publication status: Peer-reviewed paper
- Test type: retrospective or benchmark
- 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.
- Still 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 sources: https://www.nature.com/articles/s41467-024-55422-4
Source checked · editorial details
The source page was checked against this entry. Independent confirmation is not assessed by this label.
Source type: paper. Paper: linked separately.
Review scope and date are not recorded; source checked does not establish full-text review.
Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 30 citations; as of 15 Nov 2026). Attention, not a measure of correctness; not part of the score.
Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.
Follow-up and relationships
No follow-up recorded; this is not a claim that no follow-up exists.
Follow-up review scope and date are not recorded.
No source-supported relationships recorded.