AI Science Tracker

AI discoveries in 2022

91 discoveries announced in 2022

Discoveries

Integrated proteomics and metabolomics with LASSO nominate a four-analyte early IgA-nephropathy plasma panel

A study of biopsy-confirmed IgA nephropathy (IgAN) integrated plasma proteomics and untargeted metabolomics and used LASSO logistic regression to select PRKAR2A, IL6ST, SOS1 and palmitoleic acid as candidate discriminators from healthy controls. The panel had AUC 0.994 in the training set and 0.977 in the test split; ELISA assays in a separate 9-case/10-control sample set yielded an AUC of 0.91. Proteomic and metabolomic analyses also indicated complement/immune activation and altered energy and amino-acid metabolism in the studied cases.

Technology: LASSO logistic regression (glmnet), with tenfold cross-validation and additional random-forest/SVM model checks

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 study of biopsy-confirmed IgA nephropathy (IgAN) integrated plasma proteomics and untargeted metabolomics and used LASSO logistic regression to select PRKAR2A, IL6ST, SOS1 and palmitoleic acid as candidate discriminators from healthy controls.
  • Still unknown: This is an early candidate biomarker result, not a validated clinical test or biopsy replacement. The discovery analysis included 33 biopsy-confirmed IgAN cases and 20 healthy controls; the separate ELISA group contained only 9 IgAN cases and 10 controls from the same hospital. The test split is small, and the study did not test specificity against other kidney diseases or demonstrate prospective/clinical utility. The proposed biomarker biology remains speculative.
  • AI role: LASSO feature selection on integrated proteomic and metabolomic measurements nominated the four-analyte panel; the study evaluated its classification in a held-out split and measured the selected analytes by ELISA in an independent sample batch.

Evidence sources: https://link.springer.com/article/10.1186/s12014-022-09387-5

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, 40 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.

Random Forest selection guided discovery of nine TRAPUM millisecond pulsars

A Random Forest classifier identified Fermi-LAT sources as pulsar candidates; the team observed 79 selected sources with MeerKAT and reported nine new millisecond pulsars, eight in binaries. The paper reports that all nine candidate signals were subsequently confirmed by additional detections. The authors separately note an independent redetection of one pulsar that had already been discovered elsewhere; it is not counted among the nine new MSPs. This is the 2022 preprint version of a 2023 MNRAS paper.

Technology: Random Forest classification of unassociated Fermi-LAT sources

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 that a Random Forest classified 4FGL unassociated sources, 79 selected sources were observed with MeerKAT, and nine new MSPs were detected and subsequently confirmed by additional detections. Eight were in binary systems.
  • Still unknown: The paper describes nine discoveries as new and distinguishes a separate previously discovered pulsar redetected independently. This review did not conduct a separate external catalog audit of every pulsar designation.
  • AI role: A Random Forest classified unassociated 4FGL sources to choose pulsar candidates for the targeted MeerKAT survey in which the nine new MSPs were discovered.

Evidence sources: https://arxiv.org/html/2212.08528v1 · https://arxiv.org/abs/2212.08528

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 and submission metadata; journal reference confirmed on primary arXiv 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.

Machine-learning-guided electrochemical screening isolates five new reaction products

A machine-learning-guided, chemist-in-the-loop screen predicted electrochemical reactivity for 38,865 potential partners with 1,4-dicyanobenzene (DCB). Automated tests of 20 prioritized, previously untested partners yielded DCB conversion in 16; five produced an isolable major product judged of synthetic or mechanistic interest. The reported examples include a silyl-enol-ether coupling and electrochemical benzylic arylation of benzyltrimethylsilane with DCB, plus three mechanistically unexpected products.

Technology: Machine-learning-guided reaction-competency classifier using a DFT-informed molecular representation

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 tests of 20 prioritized, previously untested partners yielded DCB conversion in 16; five produced an isolable major product judged of synthetic or mechanistic interest.
  • Still unknown: This is an early reaction-discovery demonstration under one fixed DCB electrolysis system, not a general reaction predictor or optimized synthetic platform. Seven of the 16 DCB-converting selected reactions gave complex mixtures without an appreciable single isolated product; five yielded isolable major products, reported at low yields and without method optimization. DCB conversion alone does not establish a product identity, and the three mechanistically ambiguous outcomes require further study.
  • AI role: The model screened a 38,865-member virtual set for reaction competency and, with chemist selection, prioritized 20 new partners for automated electrolysis. The experiments measured DCB conversion, then product analysis isolated and characterized five major products from the selected set.

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC9756344/

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, 78 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.

Unsupervised seismic clustering reveals distinct drainage signals at Gornergletscher

An unsupervised analysis of five weeks of seismic records from Gornergletscher, Switzerland, separated cryoseismic signals into clusters associated with the 2007 Gornersee lake flood and with afternoon/evening meltwater flow. One high-energy, broadband noise cluster began about two days before the reported onset of subglacial flooding; the authors suggest it captures the earlier initiation of englacial and subglacial drainage and improves the flood chronology.

Technology: SpecUFEx unsupervised feature extraction: non-negative matrix factorization, hidden Markov modeling, principal component analysis, and k-means clustering

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 unsupervised analysis of five weeks of seismic records from Gornergletscher, Switzerland, separated cryoseismic signals into clusters associated with the 2007 Gornersee lake flood and with afternoon/evening meltwater flow.
  • Still unknown: The analysis covers five weeks at one glacier using the 2007 flood record. The inferred drainage start precedes lake-level evidence by about two days, but the source cannot definitively exclude ice-bed slip, till deformation, or small fractures as explanations for some spectral features. The study interprets cluster timing and spectra as hydrologic signals; it does not directly image subglacial drainage pathways.
  • AI role: The unsupervised learning pipeline reduced seismic spectrograms to fingerprints and grouped similar continuous-noise and icequake signals without labels. Comparing the learned clusters with lake level, GPS, temperature, and event timing exposed distinct clusters associated with lake-flood drainage and daily meltwater flow, which supported the revised drainage chronology.

Evidence sources: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JF006909

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 74 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 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.

CNN tracking reveals irradiation-driven nanovoid size fluctuations and shrinkage in copper

A convolutional neural network combined with greedy matching to detect, track and measure pre-existing nanovoids in in-situ TEM videos of irradiated copper. The analysis revealed non-monotonic, random-walk-like size fluctuations and shrinkage during irradiation; phase-field simulations were used to investigate the mechanism. Follow-up manual measurements and additional experiments confirmed the size-fluctuation observation.

Technology: Convolutional neural network for void detection, with greedy matching for tracking and migration 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 convolutional neural network combined with greedy matching to detect, track and measure pre-existing nanovoids in in-situ TEM videos of irradiated copper.
  • Still unknown: The source reports the finding for pre-existing nanovoids in irradiated copper and uses phase-field simulations to explain it. The observed fluctuation was additionally checked by manual measurements and experiments, but generalization to other materials or irradiation regimes is not established by this study.
  • AI role: Automated defect recognition and tracking across high-volume in-situ microscopy video, making it possible to measure nanovoid evolution at scale; phase-field simulations then analyzed the observed dynamics.

Evidence sources: https://www.sciencedirect.com/science/article/pii/S0022311522006687

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 27 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 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.

Machine-learning-guided PBMC proteomics identifies SLE diagnosis and activity biomarker panels

A random-forest and OPLS-DA prioritization pipeline applied to peripheral-blood mononuclear-cell proteomics identified candidate protein combinations for distinguishing systemic lupus erythematosus (SLE) from healthy and rheumatoid-arthritis controls and for separating active from inactive SLE. ELISA in a separate 80-person cohort supported a six-protein diagnosis panel (IFIT3, MX1, TOMM40, STAT1, STAT2 and OAS3; AUC 0.723 versus healthy controls and 0.815 versus rheumatoid arthritis) and a nine-protein activity panel (AUC 0.990). Single-cell RNA-sequencing analysis mapped several associated interferon-signature proteins to specific immune-cell populations.

Technology: POC-SLE: random-forest feature ranking with 1,000 bootstrap iterations, combined with OPLS-DA variable-importance 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: ELISA in a separate 80-person cohort supported a six-protein diagnosis panel (IFIT3, MX1, TOMM40, STAT1, STAT2 and OAS3; AUC 0.723 versus healthy controls and 0.815 versus rheumatoid arthritis) and a nine-protein activity panel (AUC 0.990).
  • Still unknown: The study is cross-sectional and the ELISA validation cohort contained 80 people from the same clinical setting; it does not establish longitudinal flare prediction, treatment response, causal mechanisms or clinical utility. The six-protein panel had moderate discrimination versus healthy controls (AUC 0.723) and better but still cohort-specific discrimination versus rheumatoid arthritis (AUC 0.815). The activity-panel AUC of 0.990 was reported for active versus inactive groups in this validation cohort and needs broader independent validation. Authors note limited sample size/ethnic diversity and need for mechanistic study.
  • AI role: The random forest ranked differentially expressed proteins from discovery-cohort PBMC proteomics; its selected proteins were intersected with OPLS-DA features to construct candidate diagnosis and activity panels. Researchers evaluated those panels with ELISA in a separate cohort and used single-cell transcriptomic data to examine immune-cell sources.

Evidence sources: https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2022.969509/full

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, 73 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 selects Eu²⁺-activated green and blue-green phosphors

A bagged gradient-boosted regression model trained on 129 Eu²⁺-activated phosphors predicted emission wavelengths and helped select 20 host candidates for synthesis. The authors report three Eu²⁺-activated phosphors with green or blue-green emissions: Li₂Ca₄Si₄O₁₃:Eu²⁺, Na₂Ca₂Si₂O₇:Eu²⁺, and SrLaGaO₄:Eu²⁺. The host compounds were known; novelty concerns their activator-specific luminescence. Some powders contained impurities, and selected-particle diffraction/spectroscopy characterization does not establish phase-pure bulk yields or device suitability. This is one campaign.

Technology: Bagged gradient-boosted regression trees

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: Twenty candidates were synthesized and characterized; three showed designed green or blue-green Eu²⁺ emission.
  • Still unknown: Some samples contain impurity phases; selected-particle evidence and host-specific results do not establish phase-pure bulk/device performance or complete prior-art priority.
  • AI role: A model trained on known phosphors ranked host candidates by predicted emission wavelength for experimental synthesis and characterization.

Evidence sources: https://mdr.nims.go.jp/filesets/fcaeedc2-0180-470c-af42-ec87e7356505/download

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 article/PDF, date, affiliations, methods/results, candidate measurements and stated limitations..

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-guided search discovers and synthesizes the rare-earth-free magnet Fe₃CoB₂

A feedback workflow combining machine learning, an adaptive genetic algorithm, first-principles calculations, and experiments identified and synthesized Fe₃CoB₂, a rare-earth-free magnetic compound. Magnetic measurements reported saturation polarization Js=1.39 T and anisotropy K1=1.2 MJ/m³, properties the authors describe as suitable for rare-earth-free permanent-magnet applications. This is a discovery and synthesis of a material compound; commercial performance or device readiness was not demonstrated.

Technology: Machine-learning-guided adaptive feedback with an adaptive genetic algorithm and first-principles calculations

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 feedback workflow combining machine learning, an adaptive genetic algorithm, first-principles calculations, and experiments identified and synthesized Fe₃CoB₂, a rare-earth-free magnetic compound.
  • Still unknown: The paper reports an experimentally synthesized compound and magnetic measurements, but suitability is for potential rare-earth-free permanent-magnet applications; no commercial device demonstration is reported. The workflow combines ML with adaptive genetic search, DFT and human synthesis/measurement.
  • AI role: Machine learning screened and informed the material search, while adaptive genetic structure search and first-principles calculations guided candidates; researchers synthesized the selected compound and measured its magnetic properties. The discovery is attributed to the integrated workflow, not the ML model alone.

Evidence sources: https://doi.org/10.1073/pnas.2204485119

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, 65 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.

Generative deep learning discovers the RIPK1 inhibitor RI-962 with an unreported scaffold

A target-conditioned generative deep-learning model based on a distribution-learning conditional recurrent neural network generated a tailored compound library that was screened for RIPK1 inhibition. The study reports RI-962 as a potent inhibitor with a previously unreported scaffold, validated in vitro for protection from necroptosis and in vivo in two inflammatory models.

Technology: Generative deep-learning distribution-learning conditional recurrent neural network (GDL cRNN)

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 RI-962 as a potent inhibitor with a previously unreported scaffold, validated in vitro for protection from necroptosis and in vivo in two inflammatory models.
  • Still unknown: Evidence is preclinical: cellular necroptosis assays and two inflammatory mouse models do not establish human efficacy or safety. The paper identifies one leading molecule from a generated library; it does not validate generative-model performance across targets generally.
  • AI role: The GDL cRNN generated a target-tailored virtual compound library; virtual screening and subsequent chemical synthesis and bioactivity assays selected RI-962. This provides direct experimental testing of a model-generated molecule rather than a prediction-only hit.

Evidence sources: https://pubmed.ncbi.nlm.nih.gov/36371441/

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, 114 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 screening guides ionic-liquid polymer electrolytes for lithium-metal batteries

A machine-learning and quantum-chemistry workflow screened ionic liquids for conductivity and electrochemical-window properties, then guided selection for rigid-rod polyelectrolyte membranes. The 2022 preprint reports ~50 μm membranes and lithium-cell testing, including symmetric-cell cycling up to 800 hours at room temperature and reversible operation at 6 mA cm−2 at 80°C. Its preprint abstract reports 146 mAh g−1 at 5C/80°C; the later 2023 journal abstract lists 3C, so those version-specific figures are not conflated.

Technology: Graph-convolutional neural network and ensemble SVM/random-forest/XGBoost screening with Psi4 quantum-chemistry 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: The 2022 preprint reports a machine-learning/quantum-chemistry workflow that screened 2,356 ion pairs and recommended 40 for testing in rigid-rod polyelectrolyte membranes. It reports symmetric-cell cycling up to 800 hours at room temperature and reversible operation at 6 mA cm−2 at 80°C; its abstract reports 146 mAh g−1 at 5C/80°C.
  • Still unknown: The 2023 journal abstract reports the same study but gives 146 mAh g−1 at 3C rather than 5C. The record keeps the version-specific preprint metric separate and does not transfer it to the journal version. A publisher correction changes plot labels, not the core study claim.
  • AI role: Machine-learning models and quantum-chemistry calculations screened ionic-liquid properties and guided selection for experimental polymer-electrolyte membranes and cell testing.

Evidence sources: https://assets-eu.researchsquare.com/files/rs-2220925/v1/0341f570-c3ce-430f-8b00-a72d3fa48184.pdf?c=1684221171 · https://www.nature.com/articles/s41467-023-38493-7 · https://www.nature.com/articles/s41467-023-39621-z

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 Research Square v1 full-text PDF, including version date, affiliations, Methods and Results; primary Nature Communications journal article indexed text and metadata; primary publisher correction notice..

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.

Autonomous microscopy discovers an edge plasmon in layered MnPS₃

Deep-kernel-learning-guided STEM-EELS measurements identified a low-energy edge-localized plasmon response in the layered antiferromagnet MnPS₃, distinct from the higher-energy bulk plasmon response. The system used local structure to predict spectra and uncertainty, then autonomously selected where to measure based on a physics-defined objective. After 10–25 spectra, about 1% of the image-space points, it localized the edge mode to the flake boundary.

Technology: Deep kernel learning (feed-forward neural-network feature embedding with a Gaussian-process kernel and physics-based acquisition/scalarizer)

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-kernel-learning-guided STEM-EELS measurements identified a low-energy edge-localized plasmon response in the layered antiferromagnet MnPS₃, distinct from the higher-energy bulk plasmon response.
  • Still unknown: The paper demonstrates the mode in the reported exfoliated MnPS₃ flake and STEM-EELS setup. The physical search objective was supplied by researchers (a spectral feature below the bulk-plasmon peak), so the AI did not independently formulate the scientific question. The authors report that the edge response varied with local boundary shape and present this as a proof-of-concept discovery workflow, not a complete theory of MnPS₃ plasmon behavior.
  • AI role: The model learned a structure-to-electron-energy-loss-spectrum relationship from sequential measurements and used its predicted response and uncertainty to choose subsequent STEM probe locations. A user-specified spectral scalarizer prioritized a low-energy feature relative to the bulk plasmon; the closed-loop instrument then discovered and mapped an edge-localized response without pretraining on MnPS₃ sample data.

Evidence sources: https://advanced.onlinelibrary.wiley.com/doi/10.1002/advs.202203422

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, 80 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.

Machine-learning-assisted neutron scattering shows hydrostatic pressure tunes interactions in spin ice

Machine-learning-assisted analysis of diffuse neutron scattering from Dy₂Ti₂O₇ at ambient pressure and about 1.3 GPa extracted pressure-dependent Hamiltonian parameters and found that hydrostatic pressure modifies further-neighbor magnetic interactions and the material’s nanoscale magnetic-glass ordering. The analysis supports using pressure to tune the spin-ice glass state. A transition at substantially higher pressure is a model-based prediction, not an observed transition in this experiment.

Technology: Nonlinear autoencoder, radial-basis generative surrogate and hierarchical clustering for neutron-scattering 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: Machine-learning-assisted analysis of diffuse neutron scattering from Dy₂Ti₂O₇ at ambient pressure and about 1.3 GPa extracted pressure-dependent Hamiltonian parameters and found that hydrostatic pressure modifies further-neighbor magnetic interactions and the material’s nanoscale magnetic-glass ordering.
  • Still unknown: The reported pressure comparison is between ambient pressure and an estimated hydrostatic pressure of about 1.3 GPa, using separate measurement setups. The change in magnetic couplings and glass morphology is inferred by fitting a spin model to scattering with ML-assisted analysis, not measured as a direct microscopic coupling. The authors predict that substantially higher pressure may induce a phase transition; no such transition was observed here. Some analysis and experimental steering remained manual.
  • AI role: The nonlinear autoencoder compressed measured and simulated diffuse neutron-scattering volumes; a trained generative surrogate enabled fast forward prediction of scattering from spin-Hamiltonian parameters, while clustering and inverse mapping constrained parameters against the experiment. Those ML components made it practical to analyze pressure-dependent measurements and infer the change in further-neighbor couplings during the neutron experiment workflow.

Evidence sources: https://www.nature.com/articles/s43246-022-00306-7

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 88 of works of the same type, year and subfield (OpenAlex work metrics, 26 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.

Meta AI

ESM Metagenomic Atlas: over 600 million protein structures

Meta AI published predicted structures for almost the entire MGnify90 database (over 600 million), generated in about two weeks on roughly 2,000 GPUs with a 15-billion-parameter language model, up to 60 times faster than previous methods. These are predictions, and high-confidence structures are a much smaller subset.

Technology: ESMFold (15-billion-parameter language model)

Family ESM / ESMFold: 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 preprint reports structure prediction up to 60 times faster while maintaining accuracy versus state-of-the-art methods and an atlas of 617 million structures; the blog describes no experimental checks.
  • Still unknown: The blog lists no limitations and notes some sequences fold into disordered forms; high-confidence predictions are a subset.
  • AI role: Language model (15B parameters) predicted protein structures from sequence.

Evidence sources: https://ai.meta.com/blog/protein-folding-esmfold-metagenomics/ · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1101/2022.07.20.500902&resultType=core&format=json

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.

Semi-supervised machine learning identifies two strong gravitational lenses in the Deep Lens Survey

A ResNetV2-based semi-supervised workflow ranked images from the 20-square-degree Deep Lens Survey and surfaced two Grade-A lens systems for follow-up. Keck spectroscopy secured a source redshift of z=1.81 for DLS212072337, while DLS432021848 has only a tentative single-line redshift but supporting high-resolution Hubble morphology; the paper counts both as lensing systems, with the latter less secure. The broader visually inspected list contains additional candidates, not confirmed lenses.

Technology: ResNetV2 neural network with semi-supervised MixMatch or Pi-Model and GAN-based image augmentation

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 ResNetV2-based semi-supervised workflow ranked images from the 20-square-degree Deep Lens Survey and surfaced two Grade-A lens systems for follow-up.
  • Still unknown: This is a paper-level result about a survey search and validated lens identification, not a claim that all 22 visually graded systems are confirmed. DLS212072337 has secure spectroscopic source and deflector redshifts; DLS432021848 has an ambiguous single emission line and is judged likely from HST morphology rather than conclusive spectroscopy. arXiv version 1 is a preprint; current full text was updated in 2026.
  • AI role: Ranked galaxy-survey images for strong-lensing morphology; the top-ranked systems were then human-inspected and followed up spectroscopically/imaged to validate lensing.

Evidence sources: https://arxiv.org/abs/2211.00047

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 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.

Ligand-based discovery of coronavirus main protease inhibitors using MACAW molecular embeddings

MACAW molecular embeddings and ligand-based models prioritized purchased compounds for assays. The paper reports inhibition of recombinant coronavirus main protease for eight of ten selected molecules; compound 7 also reduced hCoV-229E infectivity in a cell assay. This is not a clinical COVID-19 result or proof of SARS-CoV-2 patient benefit.

Technology: MACAW molecular embeddings used in ligand-based machine learning virtual 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 article reports 8/10 selected compounds inhibiting recombinant Mpro and follow-up cellular testing of compound 7 against HCoV-229E.
  • Still unknown: Independent replication and exhaustive prior-art review are not established.
  • AI role: MACAW molecular embeddings used in ligand-based machine learning virtual screening

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC9621234/

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, stated AI contribution, date 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.

Prospective machine-learning screen identifies six previously unreported OATP1B1 inhibitors

Researchers trained and compared seven machine-learning classifiers for OATP1B1 transporter inhibition. In prospective CHO-cell testing, 15 of 19 compounds predicted active by SVC models inhibited the transporter; six (abamectin, asiaticoside, berbamine, doramectin, mobocertinib and umbralisib) were described as previously unreported OATP1B1 inhibitors.

Technology: Seven molecular-fingerprint machine-learning classifiers, with prospective selection by support vector classifier (SVC)

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 prospective CHO-cell testing, 15 of 19 compounds predicted active by SVC models inhibited the transporter; six (abamectin, asiaticoside, berbamine, doramectin, mobocertinib and umbralisib) were described as previously unreported OATP1B1 inhibitors. The new candidate screen used contracted Eurofins assays; external execution is not a separate replication study.
  • Still unknown: The 84% refers to a small prospective test set of 19 compounds, and transporter inhibition was measured in vitro in engineered CHO cells. These findings flag possible drug–drug interaction liability; they do not show clinical interaction or patient outcome. The authors qualify six inhibitors as appearing novel at publication.
  • AI role: Models trained on in-vitro OATP1B1 inhibition data ranked compounds for prospective testing. Researchers tested 19 model-selected compounds in transporter-expressing CHO cells and observed inhibition for 15, including six compounds the authors described as novel OATP1B1 inhibitors.

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC9873312/

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 89 of works of the same type, year and subfield (OpenAlex work metrics, 31 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.

Novo Nordisk A/S

Mass spectrometry and machine learning identify bioactive peptide NHPD-50

A predicted-peptide-variant model trained using mass-spectrometry observations from seven mouse tissues helped prioritize candidate peptides. The authors report that NHPD-50, a secretogranin-1-derived peptide, lowered blood glucose and increased plasma insulin in specified diabetic-mouse experiments. The authors state that the data do not establish the peptide’s exact mechanism. These are early preclinical findings, not a human diabetes treatment; other predicted peptides had different or absent outcomes.

Technology: Predicted-peptide-variant machine-learning prioritization from mass-spectrometry 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: Primary paper reports the stated mouse assay results.
  • Still unknown: Independent replication and exhaustive prior-art review are not established; the paper states that the mechanism is undetermined.
  • AI role: A predicted-peptide-variant model prioritized candidate sequences from mass-spectrometry data; biological activity was established through subsequent assays.

Evidence sources: https://www.nature.com/articles/s41467-022-34031-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: 16 Nov 2026. Review scope: Primary Nature full article and bibliographic metadata reviewed for identity, AI role, date, affiliation and bounded mouse 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.

FLEET-guided observing campaign classifies superluminous supernovae

FLEET used transient light curves and host-galaxy features to prioritize telescope follow-up. Its first two-year campaign spectroscopically classified 21 superluminous supernovae. This is one campaign-level result, not 21 separately counted discoveries or proof that FLEET first detected every object. The result was public in the October 2022 preprint before the 2023 journal article.

Technology: FLEET machine-learning classifier

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: Spectra supported classifications in the AI-prioritized follow-up sample.
  • Still unknown: First-discovery attribution for individual objects and completeness beyond the studied alerts were not established.
  • AI role: FLEET machine-learning classifier supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://arxiv.org/abs/2210.10811 · https://pureadmin.qub.ac.uk/ws/portalfiles/portal/555515225/Fleet.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: Primary preprint abstract and submission history; indexed institutional publisher-PDF header and abstract for affiliations and journal 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.

Generative learning selects experimentally active EGFR inhibitor analogs

A recurrent generative model refined by reinforcement learning selected compounds for an EGFR enzyme assay. Four of 15 predicted-active compounds were active, including two with nanomolar inhibition; five predicted-inactive controls were inactive. The experimentally tested compounds were commercially obtainable analogs of a known privileged scaffold, not a wholly new drug scaffold or demonstrated cellular therapy.

Technology: Recurrent neural generator with reinforcement 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: An independent contract assay tested selected compounds and negative controls.
  • Still unknown: Cellular efficacy, selectivity, safety and novel-scaffold discovery are not established.
  • AI role: Recurrent neural generator with reinforcement learning supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://www.nature.com/articles/s42004-022-00733-0 · https://www.nature.com/articles/s42004-022-00733-0.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 date: 16 Nov 2026. Review scope: Primary publisher PDF abstract, selected experimental Results and affiliations; primary Crossref online-date metadata. Supplement 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.

Machine-learning screening finds previously unreported antibacterial compounds active against Burkholderia and ESKAPE pathogens

A directed-message-passing neural network trained on a 29,537-compound Burkholderia cenocepacia screen prioritized compounds from a natural-product library. Laboratory testing found five previously uncharacterized compounds that inhibited B. cenocepacia; 22 of 43 tested prioritized compounds inhibited at least one of five ESKAPE pathogens.

Technology: Directed-message-passing neural network (D-MPNN) implemented with Chemprop

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: Laboratory testing found five previously uncharacterized compounds that inhibited B. cenocepacia; 22 of 43 tested prioritized compounds inhibited at least one of five ESKAPE pathogens.
  • Still unknown: The compounds were active in vitro; no clinical effect, safety or therapeutic readiness is established. Several confirmed compounds were known antimicrobials, while five natural-product hits were described as previously uncharacterized. An inferred RpoB mechanism for STL558147 was suggested by CRISPRi sensitivity rather than directly demonstrated binding.
  • AI role: The D-MPNN ranked FDA-approved and natural-product library molecules for predicted antibacterial activity; researchers tested high-ranked compounds in bacterial growth assays and confirmed activity. It was a screening-prioritization tool, not evidence of clinical efficacy.

Evidence sources: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1010613

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, 59 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.

LIGHTHOUSE identifies riboflavin 5′-monophosphate as a PPAT inhibitor

LIGHTHOUSE, a deep-learning model using protein sequence and chemical representations, nominated riboflavin 5′-monophosphate as an inhibitor of PPAT, a target without reported inhibitors in the paper. A biochemical assay measured PPAT inhibition with an IC50 of 7.5 μM, close to the model’s predicted range; the same study also tested predicted antiviral compounds in cell models.

Technology: LIGHTHOUSE (ensemble graph-based deep-learning 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: LIGHTHOUSE, a deep-learning model using protein sequence and chemical representations, nominated riboflavin 5′-monophosphate as an inhibitor of PPAT, a target without reported inhibitors in the paper.
  • Still unknown: The result is an in-vitro enzyme assay, not a validated anticancer treatment. The authors note LIGHTHOUSE can produce false positives; COVID-19 results were also cell-based and had no clinical evidence.
  • AI role: The model scored protein–compound relationships without requiring a 3D target structure and prioritized the compound from >333,000 candidates for experimental follow-up. The authors assayed PPAT activity and confirmed inhibition in vitro.

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC9549714/

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 57 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 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.

Bayesian language-model design improved measured scFv binding

An end-to-end Bayesian language-model workflow designed diverse single-chain antibody fragment libraries that were measured for binding. In the 2023 journal version, the best generated scFv bound 28.7-fold better than the best directed-evolution comparator, and 99% of designs in the most successful library improved on the initial candidate. The 2022 bioRxiv v1 abstract already reported this result chain, giving 28.8-fold for the best generated scFv; the small difference is version-specific. These are in-vitro binding results, not clinical efficacy.

Technology: Bayesian optimization with a language-model-based antibody sequence representation

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: Designed scFvs were empirically measured. The publisher abstract reports the best generated scFv had 28.7-fold greater binding than the best directed-evolution scFv and 99% of the designed scFvs in the most successful library improved on the initial candidate.
  • Still unknown: The journal version reports the best generated scFv at 28.7-fold better binding while the v1 abstract reports 28.8-fold; this minor version discrepancy is retained. The measured binding result does not establish clinical efficacy.
  • AI role: A Bayesian, language-model-based workflow designed candidate scFv libraries, whose binding was experimentally measured.

Evidence sources: https://api.biorxiv.org/details/biorxiv/10.1101/2022.10.07.502662 · https://www.nature.com/articles/s41467-023-39022-2.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 date: 17 Nov 2026. Review scope: Official bioRxiv API v1 date/abstract and primary Nature journal PDF for full result 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.

Machine learning and DFT discover high-entropy Invar alloys

A closed-loop active-learning framework combining machine learning, density-functional theory and thermodynamic calculations screened millions of compositions. Of 17 new alloys processed and characterized, two high-entropy Invar alloys showed very low thermal expansion, about 2 x 10^-6 per kelvin at 300 K. Only the top-ranked candidates were synthesized, so the hit rate is small.

Technology: Closed-loop active learning combining machine learning, density-functional theory and thermodynamic calculations

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: 17 new alloys were processed and characterized; two high-entropy Invar alloys showed very low thermal expansion of about 2 x 10^-6 per kelvin at 300 K.
  • Still unknown: The method works on very sparse data, and the abstract notes that thermodynamic design rules alone often fail in high-dimensional spaces; no further limits are stated.
  • AI role: A closed-loop active-learning framework integrating machine learning with density-functional theory and thermodynamic calculations.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/science.abo4940&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, 803 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.

Google DeepMind

AlphaTensor: new algorithms for matrix multiplication

A reinforcement-learning agent found matrix-multiplication algorithms better than known ones for many sizes (for example, 76 multiplications instead of 80 for 4×5 by 5×5). Algorithms optimized for hardware were 10–20% faster on V100 GPUs and TPU v2. The improvement over Strassen applies in a finite field.

Technology: AlphaTensor (based on AlphaZero)

Family AlphaTensor: 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: Algorithms are provably correct by construction and their multiplication counts are compared with known methods; hardware-targeted algorithms were benchmarked on V100 GPU and TPU v2.
  • Still unknown: The page lists no formal limitations; it notes the search space is very large and the work is specific to matrix multiplication.
  • AI role: Reinforcement-learning agent searched for tensor decompositions giving new matrix-multiplication algorithms.

Evidence sources: https://deepmind.google/blog/discovering-novel-algorithms-with-alphatensor/ · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-022-05172-4&resultType=core&format=json

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, 613 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 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.

Machine learning and synthesis clarify metastable phases of carbon

An algorithm constructed metastable phase diagrams of carbon and identified structural features of n-diamond that were then verified experimentally. The group also synthesized predicted phases and compared structures under a microscope; the work concerns carbon only and does not establish useful properties of the new materials.

Technology: Machine learning and atomistic simulations

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: Workflow found 505 candidate carbon structures; a neural network trained on 273 phases, test error 37.1 meV/atom. Diamond-anvil laser-heating experiments on graphite with electron microscopy were used to compare predicted metastable phases, including n-diamond.
  • Still unknown: The search is not guaranteed exhaustive; the paper calls cubic-diaphite a candidate structure for n-diamond. The model had higher errors at high pressure; kinetics are only partly covered. Argonne synthesis of unreported phases was not seen in the paper text read.
  • AI role: A deep neural network predicted Gibbs free energy; a support vector machine drew phase boundaries; an evolutionary search proposed structures. Humans ran the experiments.

Evidence sources: https://doi.org/10.1038/s41467-022-30820-8 · https://www.alcf.anl.gov/news/scientists-use-machine-learning-accelerate-materials-discovery

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, 55 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.

Deep-learning screening identifies covalent SARS-CoV-2 main-protease inhibitor scaffolds

A two-stage deep-learning screen selected cysteine-directed compounds for testing against SARS-CoV-2 3CL protease. Of 32 top-ranked compounds experimentally assayed, 12 inhibited the enzyme; six had IC₅₀ values below 12 μM, with the strongest reported at 1.4 μM. Five of those six exhibited time-dependent inhibition consistent with covalent binding.

Technology: Deep-learning COVCL and NOVCL classifiers integrating a directed message-passing neural network and feed-forward network

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 two-stage deep-learning screen selected cysteine-directed compounds for testing against SARS-CoV-2 3CL protease.
  • Still unknown: Evidence in the primary abstract is biochemical enzyme inhibition and inhibition-mechanism assays. It does not establish antiviral efficacy in cells/animals or clinical benefit. Potency values are for the reported enzyme assay; most tested compounds were inactive, and the six sub-12 μM hits are early leads.
  • AI role: The COVCL model first prioritized cysteine-targeted covalent candidates, then NOVCL ranked those candidates for predicted 3CL-protease inhibition. Researchers tested 32 selected molecules in enzyme assays and performed follow-up tests of time-dependent inhibition for active hits.

Evidence sources: https://pubmed.ncbi.nlm.nih.gov/36209629/

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, 25 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.

Crystal-graph learning and DFT identify 19,512 additional thermodynamically stable materials

The authors trained crystal-graph attention networks on an expanded, composition- and symmetry-balanced dataset and used them to search 2,500 binary and ternary structure prototypes covering about one billion inorganic compounds. Density-functional-theory validation identified 19,512 additional compounds on the calculated thermodynamic convex hull and about 150,000 more within 50 meV/atom of it; the paper also analyzed predicted superconducting and superhard materials and compounds with large gap deformation potentials.

Technology: Crystal-graph attention neural networks trained on expanded density-functional-theory materials data

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 authors trained crystal-graph attention networks on an expanded, composition- and symmetry-balanced dataset and used them to search 2,500 binary and ternary structure prototypes covering about one billion inorganic compounds.
  • Still unknown: This is an arXiv preprint and the reported stable compounds are computationally predicted, with DFT convex-hull validation; the study does not report their experimental synthesis. Stability depends on the calculated structures and thermodynamic reference data. The additional ~150,000 compounds are near-hull candidates, not counted as thermodynamically stable discoveries in the title or summary.
  • AI role: The graph networks accelerated screening of roughly one billion prototype-generated candidates, after which density-functional-theory calculations validated stability for identified candidates; machine learning plus ab initio calculations also ranked materials for selected functional properties.

Evidence sources: https://arxiv.org/abs/2210.00579

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.

Robot-guided optimization identifies six non-aqueous electrolytes with improved fast-charge performance

A closed-loop platform paired Clio, an automated electrolyte-formulation and conductivity-testing robot, with Dragonfly Bayesian optimization. In 42 experiments over two workdays, it selected six non-aqueous electrolyte formulations from a LiPF₆ and EC/EMC/DMC design space. The team tested these formulations in 220 mAh graphite‖NMC pouch cells; all showed improved fast-charging capability relative to a baseline electrolyte selected in advance from the same design space. The search reached its reported target about six times faster than random search using the same automated setup.

Technology: Dragonfly Bayesian optimization coupled to the Clio robotic electrolyte experimentation 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: A closed-loop platform paired Clio, an automated electrolyte-formulation and conductivity-testing robot, with Dragonfly Bayesian optimization.
  • Still unknown: This is a search within a defined design space of LiPF₆ in EC/EMC/DMC solvents, not a claim that the constituent chemicals or all six electrolyte recipes were wholly unprecedented. The optimization target was bulk ionic conductivity, with fast-charge cell performance as a subsequent validation. Pouch-cell tests used nominal 220 mAh graphite‖LiNi0.5Mn0.3Co0.2O2 cells and a baseline chosen from the same design space; results do not establish commercial-scale manufacturing or long-term fleet performance. The paper compares optimization time with random search in the same automated experiment.
  • AI role: Dragonfly selected electrolyte formulations for experiments and used measurement feedback from Clio to adapt subsequent selections. The closed-loop system optimized measured ionic conductivity over a non-aqueous single-salt/ternary-solvent search space; selected formulations were then assembled and evaluated by the researchers in pouch cells.

Evidence sources: https://www.nature.com/articles/s41467-022-32938-1

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 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, 248 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.

AlphaFold helps interpret activation of PINK1, a Parkinson’s-linked protein

After experimentally determining different forms of PINK1, the WEHI group used AlphaFold to model the human protein and its dimer, obtaining structures consistent with experimental results. The predictions also helped interpret how mutations associated with early-onset Parkinson’s may affect dimerization; this does not amount to a validated therapy.

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: The WEHI group compared an AlphaFold model of human PINK1 dimer with its own experimental structure work and used it to model disease mutations, which it says confirmed their suspicions.
  • Still unknown: The article describes no specific limitations of the predictions and says plenty of questions remain.
  • AI role: AlphaFold modelled human PINK1 and its dimer and the effect of mutations on dimerisation.

Evidence sources: https://deepmind.google/blog/targeting-early-onset-parkinsons-with-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: 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.

ProteinMPNN: protein sequences designed in about one second

A neural network generates the amino-acid sequence for a given protein shape in about one second, over 200 times faster than the best previous software, without specialist tuning. A related study shows that proteins designed with ProteinMPNN fold correctly much more often, including nanoscale rings and complex assemblies. Researchers describe it as the beginning of machine learning in protein design.

Technology: ProteinMPNN

Family ProteinMPNN: 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: Speed is compared with prior software; a related Science study tested lab-made designs, which folded as intended more often, including nanoscale rings imaged by electron microscopy.
  • Still unknown: The article gives no quantitative success rates and notes protein data is scarce, requiring trial and error.
  • AI role: Neural network generated amino-acid sequences for given protein shapes.

Evidence sources: https://www.ipd.uw.edu/2022/09/proteinmpnn-excels-at-creating-new-proteins/ · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1126/science.add2187&resultType=core&format=json

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, 2354 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.

MIMaL predicts and experimentally validates five yeast gene functions from protein–metabolite profiles

MIMaL used tree-based regression and SHAP interpretation on yeast proteome/metabolome measurements, then clustered protein-control profiles across single-gene knockouts. It predicted functions for five genes, with experimental validation reported for all five. Two previously uncharacterized genes, YJR120W and YDL157C, were predicted to modulate mitochondrial translation; functions were also predicted and validated for SDH9, ISC1 and FMP52. The study additionally validated previously unrecognized protein links to citrate.

Technology: MIMaL: ExtraTrees regression of metabolite abundance from proteomic profiles, SHAP interpretation, and UMAP/OPTICS clustering of protein-control profiles

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: MIMaL used tree-based regression and SHAP interpretation on yeast proteome/metabolome measurements, then clustered protein-control profiles across single-gene knockouts.
  • Still unknown: This was a yeast study using a prior 174-single-gene-knockout multi-omics dataset, with 348 condition profiles; it does not establish corresponding human gene functions. Five gene functions are reported validated, and the article specifically states YJR120W and YDL157C modulate mitochondrial translation. Some experimental validation used existing knockout strains and measurements; no independent replication is implied.
  • AI role: A tree-based model predicted metabolite profiles from proteomic measurements; SHAP-derived protein-control values and clustering across knockout conditions were used to infer new protein–metabolite connections and gene functions. The researchers experimentally validated predicted citrate-control relationships and five gene-function predictions.

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC9801967/

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 81 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 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 learning detects pancreatic cancer on CT in a nationwide study

A deep-learning tool for contrast-enhanced CT reached 89.9% sensitivity and 95.9% specificity in an internal test set (not significantly different from the original radiologist reports, 96.1%). On 1,473 real-world Taiwanese CT studies it reached 89.7% sensitivity and 92.8% specificity, but sensitivity for cancers under 2 cm was 74.7%. The study is retrospective, with no prospective reader comparison in the abstract.

Technology: Segmentation CNN with an ensemble of five CNN classifiers for pancreatic cancer on CT

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 CT tool was tested on an internal retrospective test set against original radiologist reports and on 1,473 real-world Taiwanese CT studies, reaching 89.7% sensitivity and 92.8% specificity in the latter.
  • Still unknown: Sensitivity for tumours under 2 cm was 74.7%; the study is retrospective with no prospective reader comparison in the abstract.
  • AI role: An end-to-end computer-aided detection tool (segmentation CNN plus an ensemble of five CNNs) detected pancreatic cancer on contrast CT.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1148/radiol.220152&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, 207 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.

Machine learning screens perovskite oxygen-reduction electrodes

The study introduced ionic Lewis-acid strength as a physical descriptor and used machine learning to screen 6,871 perovskite compositions. Four oxides were synthesized and reported to have superior oxygen-reduction activity metrics in specified ceramic-fuel-cell tests. The record counts one paper-level study and does not imply 6,871 physical syntheses or commercial deployment.

Technology: Machine-learning composition screening; ionic Lewis-acid-strength 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: Primary publisher abstract reports four synthesized oxides screened from 6,871 and experimental activity results.
  • Still unknown: Exact formulation-level overlap and independent replication not exhaustively assessed.
  • AI role: ML plus physical descriptor screened composition candidates followed by synthesis and electrochemical tests.

Evidence sources: https://www.nature.com/articles/s41560-022-01098-3

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: 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.

AI detects Parkinson's disease from nocturnal breathing signals

An attention-based AI model extracting breathing from radio waves during sleep detected Parkinson's disease with an AUC of 0.90 on held-out data and 0.85 on external sets, across 7,671 individuals. Its severity estimates correlated with MDS-UPDRS scores (R = 0.94), and it worked for at-home assessment. Caveat: largely retrospective data and no prospective diagnostic trial reported.

Technology: Attention-based deep learning on contactless radio-wave breathing signals

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: Parkinson's detection was tested on held-out data (AUC 0.90) and external test sets (AUC 0.85) across 7,671 individuals, and severity estimates were compared with MDS-UPDRS scores (R = 0.94).
  • Still unknown: The authors describe the findings as initial evidence that the model may be useful for risk assessment before clinical diagnosis.
  • AI role: Attention-based AI model extracting breathing from radio waves during sleep to detect and grade Parkinson's disease.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-022-01932-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: 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, 321 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.

Deep learning helps experimentally map a ribozyme neutral network

After earlier experimental evolutionary rounds, the authors trained a deep classifier to guide the final in-silico generation and experimentally tested candidates. They measured all 2^16 intermediates between two active RNA ligase ribozymes separated by 16 mutations and report an extensive network of neutral paths. The result applies to this ribozyme and assay, not all RNA evolution. The paper also notes possible model overfitting and the restricted classifier’s accuracy.

Technology: Deep-learning classifier; evolutionary 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 study reports experimental activity measurements for all 2^16 connecting sequence variants and a screen of more than 120,000 ribozyme sequences.
  • Still unknown: The neutral network is bounded to one RNA ligase ribozyme and assay; generality across other landscapes is not established, and the paper notes potential classifier overfitting.
  • AI role: A deep-learning classifier trained on prior experimental generations guided in-silico selection, mutation and recombination in the final generation.

Evidence sources: https://www.nature.com/articles/s41467-022-32538-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: 16 Nov 2026. Review scope: Primary Nature Communications indexed full text, abstract, main text, publication metadata and author affiliation..

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-guided peptide design yielded non-hemolytic anticancer candidates

A recurrent neural network generated peptide sequences and classified predicted activity and hemolysis; a genetic-algorithm route also generated analogues. After sequence and physicochemical filters, researchers synthesized and tested 33 peptides. The paper reports 11 active anticancer peptides in its cell assays, four non-hemolytic under its red-cell assay. These in-vitro results do not establish cancer-cell specificity, clinical safety or therapeutic efficacy. A related 2020 modeling paper is separate from this experimental study; no claim of absolute sequence novelty is made.

Technology: Generative recurrent neural network, genetic algorithm, and recurrent-neural-network activity and hemolysis 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: The abstract reports that 33 peptides were experimentally evaluated, 11 were active in anticancer cell assays, and four of the active peptides were non-hemolytic in the reported assays.
  • Still unknown: In-vitro activity and red-cell hemolysis do not establish cancer-cell specificity, clinical safety or treatment efficacy. Detailed potency measurements and absolute sequence novelty were not reassessed in this bounded review.
  • AI role: RNN and genetic-algorithm generation with RNN activity and hemolysis classifiers guided peptide selection for synthesis and testing.

Evidence sources: https://pubmed.ncbi.nlm.nih.gov/35880810/ · https://chemistry-europe.onlinelibrary.wiley.com/doi/10.1002/cmdc.202200291

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 metadata and primary PubMed abstract/affiliation block; full PMC endpoint returned a browser verification challenge in this 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.

Machine learning generates active daunomycin-binding synthetic nucleic-acid polymers outside selected sequence families

The study combined in-vitro selection with a conditional variational autoencoder trained on selection data to generate high-affinity highly functionalized nucleic-acid polymer (HFNAP) sequences. Experimentally generated CVAE sequences bound daunomycin with reported dissociation constants of 9–26 nM despite having no apparent primary-sequence similarity to experimentally selected polymers; competition selections found selected generated variants performed as well as or better than sequences obtained after multiple rounds of high-stringency selection.

Technology: Conditional variational autoencoder (CVAE) trained on in-vitro selection-derived HFNAP sequence/fitness 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: Experimentally generated CVAE sequences bound daunomycin with reported dissociation constants of 9–26 nM despite having no apparent primary-sequence similarity to experimentally selected polymers; competition selections found selected generated variants performed as well as or better than sequences obtained after multiple rounds of high-stringency selection.
  • Still unknown: The finding is target binding by synthetic HFNAPs, not therapeutic efficacy or cellular activity. Binding results are specific to daunomycin and reported assays; the authors describe sequences as having no apparent similarity rather than proving wholly unrelated functional mechanisms. Some competition performance was assessed by in-vitro selection rather than a clinical or organism-level test.
  • AI role: The authors used the CVAE to condition generation of novel HFNAP sequences on predicted daunomycin-binding fitness, then synthesized/tested generated sequences in binding and competition-selection assays. The machine-learning model generated the active sequences; wet-lab selection and competition assays established their binding activity.

Evidence sources: https://www.nature.com/articles/s41467-022-31955-4

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: at percentile 79 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.

ATLAS boosted decision trees enable first observation of triple-W boson production

In 139 fb⁻¹ of 13 TeV LHC data, ATLAS observed simultaneous WWW production at 8.0σ (5.4σ expected) and measured an inclusive cross-section of 820 ± 100(stat) ± 80(syst) fb. The analysis selected same-sign dilepton-plus-jets and trilepton final states. Two boosted decision trees—one per channel—improved signal/background separation and supplied distributions for the fit. The measured rate is about 2.6σ above the Standard Model prediction, not a significant discrepancy.

Technology: Two boosted decision tree classifiers for the dilepton and trilepton channels

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 139 fb⁻¹ of 13 TeV LHC data, ATLAS observed simultaneous WWW production at 8.0σ (5.4σ expected) and measured an inclusive cross-section of 820 ± 100(stat) ± 80(syst) fb.
  • Still unknown: The final peer-reviewed paper reports 8.0σ (5.4σ expected), superseding the earlier 2021 conference-note/press account that quoted 8.2σ and 850 fb. CMS had previously observed combined inclusive production of three massive gauge bosons; this ATLAS paper is the first observation specifically of WWW production. The measured cross-section is about 2.6σ above the theory prediction, below a significant deviation, with data-driven background and systematic uncertainties included.
  • AI role: The BDTs were trained separately for the two- and three-lepton channels, using 12 and 11 well-modelled event variables respectively, to discriminate WWW signal from background. Their output distributions entered the likelihood fit; the paper says the improved separation helped enable the observation.

Evidence sources: https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.129.061803

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 90 of works of the same type, year and subfield (OpenAlex work metrics, 33 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.

Deep learning and spectroscopy confirm four faint quasars at redshift about 5

A deep-learning and Bayesian-information-criterion search of Subaru Hyper Suprime-Cam photometry selected four faint high-redshift quasar candidates. Palomar spectroscopy found broad Lyman-alpha emission and breaks in all four, confirming redshifts between 4.7 and 5.6; one object also received a black-hole mass estimate.

Technology: Deep-learning photometric classifier combined with Bayesian information criterion

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 deep-learning and Bayesian-information-criterion search of Subaru Hyper Suprime-Cam photometry selected four faint high-redshift quasar candidates.
  • Still unknown: The paper spectroscopically confirms four targeted candidates, not the full 35-object catalog. The selection method combines deep learning with a Bayesian information criterion and conventional candidate filtering; AI was one component rather than the sole discovery mechanism.
  • AI role: The classifier shortlisted high-redshift quasar candidates from survey photometry; human researchers selected four for observation and used telescope spectra to confirm all four as quasars.

Evidence sources: https://doi.org/10.5303/JKAS.2022.55.4.131

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.

Machine learning isolates aerosol-driven cloud-cover increase after the Holuhraun eruption

Using MODIS satellite observations and a random-forest counterfactual trained on meteorology and non-eruption years, researchers estimated cloud responses to the 2014 Holuhraun volcanic aerosol plume. They found cloud fraction increased by about 10% across the studied region; this change contributed more to the estimated aerosol cooling than the smaller radiative effect from brighter clouds with smaller droplets. The study supplies an observational constraint on aerosol–cloud climate forcing.

Technology: Random-forest ML-MODIS surrogate trained on meteorological reanalysis and MODIS satellite data from 2001–2020, excluding the 2014 eruption year

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 MODIS satellite observations and a random-forest counterfactual trained on meteorology and non-eruption years, researchers estimated cloud responses to the 2014 Holuhraun volcanic aerosol plume.
  • Still unknown: The result concerns one volcanic natural experiment, Holuhraun in 2014, and satellite observations of a North Atlantic/Northern European region. The cloud response is estimated by comparing observations with an ML counterfactual and is not a randomized intervention; it depends on model and MODIS retrieval assumptions. The paper reports about a 10% cloud-cover increase and attributes the leading radiative contribution to cloud fraction, while the droplet-size effect is also present and the liquid-water-path contribution is weak.
  • AI role: The random forest estimated the cloud properties expected under observed meteorological conditions without volcanic aerosol, creating a counterfactual baseline. Comparing these predictions with 2014 MODIS observations let the authors isolate cloud changes associated with the Holuhraun aerosol perturbation from meteorological co-variability and estimate the relative radiative contributions of cloud-fraction, liquid-water-path, and droplet-size changes.

Evidence sources: https://www.nature.com/articles/s41561-022-00991-6

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, 144 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.

AlphaFold Database: over 200 million structures

The predicted-protein-structure archive, maintained with EMBL-EBI, grew from nearly 1 million to over 200 million structures, covering practically all cataloged proteins, with free access and bulk download from Google Cloud. More than 500,000 researchers in 190 countries had already used it. These are predictions, not experimental measurements.

Technology: AlphaFold 2

Family AlphaFold: 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 post gives no accuracy metrics for the new structures; it cites two Nature AlphaFold papers and uses in experiments as evidence.
  • Still unknown: The post states no explicit limitations or caveats on prediction accuracy.
  • AI role: AlphaFold predicted protein structures added to the AlphaFold Database.

Evidence sources: https://deepmind.google/discover/blog/alphafold-reveals-the-structure-of-the-protein-universe/

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 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.

ProtGPT2: a language model generating novel protein sequences

ProtGPT2, a 738-million-parameter autoregressive transformer trained on UniRef50, generates de novo protein sequences in seconds. The authors report the generated sequences are distantly related to natural ones, sample unexplored regions of protein space, and match natural sequences in amino acid and disorder propensities. Caveat: the evaluation is computational (sequence searches and similarity networks), with no experimental validation in the sources checked.

Technology: ProtGPT2 (transformer protein language 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: Generated sequences were assessed computationally: composition and disorder propensities, sequence searches, and AlphaFold-predicted folds; no experiments are described in the abstract read.
  • Still unknown: Not reported in the abstract read; no experimental validation is described.
  • AI role: ProtGPT2, a 738-million-parameter autoregressive transformer trained on UniRef50, generated de novo protein sequences.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-022-32007-7&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, 883 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.

Random forest prioritizes dual-stage antimalarial compounds

A random forest trained on parasite activity data screened commercial compounds; 100 predicted actives and 20 predicted inactives were tested. The primary manuscript reports 18 dose-response-validated compounds with liver- and blood-stage activity among 26 liver-stage hits. The selected sets are not population samples, and assays used P. berghei liver-stage and P. falciparum blood-stage parasites.

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 indexed Methods/Results report 18 dual-stage validated compounds and screening denominators.
  • Still unknown: Chemical prior art, complete supplemental details and clinical translation remain open; do not present 18/26 as general hit rate.
  • AI role: Classifier prioritizes compounds from >1.5 million virtual candidates; human structural-diversity selection and assays follow.

Evidence sources: https://doi.org/10.1021/acsinfecdis.2c00189

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: PMC-indexed primary manuscript text; publisher primary metadata/index. 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.

TREWS: early sepsis alerts evaluated in five hospitals

A machine-learning system monitored about 590,000 patients in five hospitals, flagging sepsis risk from medical records. In the studied cohort, timely clinical confirmation of alerts was associated with lower mortality, organ failure and length of stay. Observational study: the association does not show that AI caused better outcomes.

Technology: TREWS (Targeted Real-time Early Warning System)

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: A prospective multi-site cohort of 590,736 monitored patients (five hospitals) compared outcomes of sepsis alerts confirmed by a provider within 3 hours with outcomes otherwise, adjusting for severity.
  • Still unknown: The association is observational and does not show that the AI caused better outcomes.
  • AI role: Machine-learning early warning system (TREWS) flagged sepsis risk from medical records.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41591-022-01894-0&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, 416 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.

Interpretable machine learning discovers ultrapermeable polymer membranes for gas separation

Multitask machine-learning models trained on experimental gas permeability data screened over 9 million hypothetical polymers. They identified hundreds of never-before-seen ultrapermeable membranes, with molecular dynamics simulations confirming the predicted permeabilities of the top candidates. The validation is computational rather than fabricated membranes, so the candidates remain unproven experimentally.

Technology: Multitask machine learning with model interpretation, plus molecular dynamics

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 dynamics simulations checked the predicted permeabilities of the top candidates among over 9 million screened hypothetical polymers; no membranes were fabricated.
  • Still unknown: The authors suggest many candidates could be translated to reality, which is not tested; no limitations section is reported.
  • AI role: Interpretable multitask machine-learning models trained on experimental gas permeability data screened hypothetical polymers.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1126/sciadv.abn9545&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, 257 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.

Bayesian search and synthesis reveal a high-anisotropy [Fe/Co/Fe/Ni]₁₃ multilayer

A first-principles and Bayesian-optimization workflow selected [Fe/Co/Fe/Ni]₁₃ for experimental fabrication from four-element, four-layer-period magnetic multilayer candidates. The researchers grew the repeated monoatomic-layer structure and measured it with X-ray diffraction and SQUID magnetometry. They report a magnetic anisotropy constant of 374 kJ m⁻³ and saturation magnetization of 1313 emu cm⁻³, and identify the material as a new ferromagnetic multilayer with reported magnetocrystalline anisotropy exceeding L1₀ FeNi and FeCo benchmarks.

Technology: Gaussian-process Bayesian optimization with expected-improvement acquisition and a Matérn 5/2 kernel, coupled to first-principles magnetic-anisotropy calculations

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 first-principles and Bayesian-optimization workflow selected [Fe/Co/Fe/Ni]₁₃ for experimental fabrication from four-element, four-layer-period magnetic multilayer candidates.
  • Still unknown: This is a proof-of-concept search over a restricted four-element, four-layer-period design space; the best candidate required 52 monoatomic layers in its repeated stack. The sample’s measured spontaneous easy axis remained in-plane and its shape-corrected effective anisotropy was negative, even though the authors report a large intrinsic magnetocrystalline-anisotropy value. Their proposed local perpendicular domains were a hypothesis requiring microscopy follow-up. The study does not establish a practical permanent magnet or device.
  • AI role: The model sequentially used first-principles magnetocrystalline-anisotropy calculations to rank discrete layer sequences and search for high-anisotropy multilayers. The top three candidates were fabricated and characterized; the resulting [Fe/Co/Fe/Ni]₁₃ experimental sample gave the paper’s highest measured anisotropy and saturation magnetization among the three. This ML-guided candidate selection and iterative calculation materially directed the new synthesis.

Evidence sources: https://www.tandfonline.com/doi/pdf/10.1080/27660400.2022.2094698

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, 14 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.

Semi-supervised learning selects a lithium thioborate for vacancy engineering

Semi-supervised learning screened lithium-containing structures and selected Li3BS3 for experimental characterization. The parent material was a poor room-temperature conductor. Researchers then introduced vacancies by Si substitution and amorphized the substituted material; the paper reports approximately four orders of magnitude higher ionic conductivity for a-Li2.95B0.95Si0.05S3. The model selected the framework, while human-led defect engineering produced the high-conductivity sample. This does not establish that pristine Li3BS3 is superionic.

Technology: Semi-supervised learning; agglomerative 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 selected Li3BS3 for characterization and reports increased ionic conductivity after Si substitution and amorphization, with the combined vacancy-engineering treatment improving conductivity by roughly four orders of magnitude.
  • Still unknown: The ML model selected a framework candidate; vacancy engineering and synthesis were subsequent researcher-led steps. The primary page was checked through indexed full-text results because direct article opening returned 403.
  • AI role: Semi-supervised learning used literature-derived conductivity labels and structural descriptors to rank lithium-containing structures and select Li3BS3 for experimental characterization.

Evidence sources: https://authors.library.caltech.edu/records/evkz6-5jd40 · https://pubs.rsc.org/en/content/articlehtml/2023/ee/d2ee03499a

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: CaltechAUTHORS primary manuscript record and deposited manuscript metadata; RSC indexed primary full-text results and methods..

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 selected fractionally doped perovskites for synthesis

A function-confined gradient-boosting model trained on 632 records was used to rank fractionally doped perovskite oxides for a solar thermochemical hydrogen-production application. The authors synthesized 11 selected compositions, seven of which the model classified as relevant to that application. This is a composition-discovery and synthesis result; it does not show that all seven were experimentally validated for hydrogen-production performance.

Technology: Function-confined gradient-boosting 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 authors report synthesizing 11 model-selected fractionally doped perovskite oxide compositions; seven were classified as relevant to solar thermochemical hydrogen production. The model was evaluated on 632 training records and achieved reported 95.4% accuracy; this classification metric is not a performance result for the 11 synthesized materials.
  • Still unknown: The evidence reviewed does not establish experimentally measured hydrogen-production performance for all seven materials. The entry does not claim each composition is novel outside the authors’ checked data or that model accuracy generalizes beyond the reported dataset.
  • AI role: A function-confined gradient-boosting classifier ranked compositions for synthesis based on target application criteria.

Evidence sources: https://www.nature.com/articles/s43246-022-00269-9 · https://par.nsf.gov/servlets/purl/10356865

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 publisher record and full article author manuscript; composition synthesis separated from predicted application relevance..

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.

Active learning improves a cell-free carbon-fixation cycle

An ensemble learned from experimental feedback to select compositions for an engineered carbon-fixation cycle. Laboratory measurements supported improved glycolate production compared with the study's reference cycle. This is an in-vitro pathway result, not carbon capture by living organisms or a demonstrated industrial process. Other applications in the same paper are not additional catalog entries.

Technology: METIS XGBoost 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: Selected pathway compositions received biochemical measurements.
  • Still unknown: Independent replication, in-vivo function and scale-up remain unestablished.
  • AI role: METIS XGBoost active learning supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://www.nature.com/articles/s41467-022-31245-z · https://www.nature.com/articles/s41467-022-31245-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 date: 16 Nov 2026. Review scope: Indexed primary publisher PDF model Methods, date and affiliations; prior primary experimental Results review retained. Direct PDF access blocked..

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 selection identifies a six-metabolite signature for one-year cardiovascular events after PCI in type 2 diabetes

VSURF feature selection prioritized six serum metabolites—ADPR, 1-methylnicotinamide, D-ribose 5-phosphate, nicotinamide, L-tryptophan and phosphatidylcholine PC(36:2)—for one-year major adverse cardiovascular events (MACEs) after PCI in patients with type 2 diabetes. Random-forest and XGBoost models were tested in a nested case-control study and an independent prospective cohort of 301 patients; in the external cohort, reported AUCs were 0.92 and 0.94 for MACEs, respectively. The study also linked NAD+ biosynthesis perturbation to altered bioenergetics, inflammation and proliferation in high-glucose-cultured human aortic smooth-muscle cells.

Technology: VSURF random-forest variable selection; random forest and XGBoost 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: Random-forest and XGBoost models were tested in a nested case-control study and an independent prospective cohort of 301 patients; in the external cohort, reported AUCs were 0.92 and 0.94 for MACEs, respectively. For the prognostic panel, the external cohort was collected before the development cohort. Separate cell experiments support the reported pathway mechanism, not prospective clinical performance of the panel.
  • Still unknown: The discovery component was a matched nested case-control sample (108 cases and 108 controls) from a single center; external validation was in a separate Chinese cohort of 301 patients, so transportability to other populations is unknown. AUCs describe retrospective/prospective cohort validation, not demonstrated clinical utility or improved patient outcomes. HASMC experiments do not prove the metabolites cause clinical events, and the paper states further mechanistic work is needed.
  • AI role: VSURF selected a six-metabolite feature set from serum metabolomic profiles. Random-forest and XGBoost models used that panel to predict one-year post-PCI outcomes; targeted metabolite testing evaluated the panel in a separate prospective cohort.

Evidence sources: https://link.springer.com/article/10.1186/s12933-022-01561-1

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 scope and date are not recorded; source checked does not establish full-text review.

Citations: at percentile 51 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.

BirdNET: citizen science through bird-song identification

The app recognizes over 3,000 species from their calls, with more than 2.2 million people contributing since 2018; in four test cases, its data reproduced known patterns such as song distribution and the migration ranges of two species. Published in PLOS Biology; only four cases, with no identification-error figures.

Technology: BirdNET (bioacoustics)

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 four test cases where conventional research had robust answers, BirdNET data replicated known patterns such as song-type distribution and migration ranges.
  • Still unknown: No identification-error rates are reported and the authors describe the validation as the first step.
  • AI role: Bird-song recognition app identified species from calls and produced citizen-science observations.

Evidence sources: https://news.cornell.edu/stories/2022/06/ai-powered-birdnet-app-makes-citizen-science-easier · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1371/journal.pbio.3001670&resultType=core&format=json

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, 88 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.

Machine-learning-guided Fe₂P-type alloys reach cryogenic magnetocaloric temperatures without rare earths

An artificial neural network trained on reported experimental compositions guided optimization of Fe₂P-type Mn–Fe–P–Si alloys for sub-77 K magnetocaloric use. Experiments measured a 97 K transition for Mn₁.₇₀Fe₀.₃₀P₀.₆₃Si₀.₃₇ at 1 T; replacing a small fraction of Fe with Co lowered the transition to 73 K. The Co-containing composition showed a 7.5 J kg⁻¹ K⁻¹ magnetic entropy change and a 1.4 K reversible adiabatic temperature change under a 0–5 T field.

Technology: Artificial neural network trained on a curated, literature-mined dataset of experimental Fe₂P-type magnetocaloric compounds

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 artificial neural network trained on reported experimental compositions guided optimization of Fe₂P-type Mn–Fe–P–Si alloys for sub-77 K magnetocaloric use.
  • Still unknown: The reported compositions are rare-earth-free but still contain transition metals and phosphorus/silicon; results characterize laboratory alloys and do not establish a working hydrogen-liquefaction refrigerator. The 73 K transition requires Co substitution and is reported with a modest reduction in magnetization. The 1.4 K reversible adiabatic change is measured over 0–5 T; stated performance and operating window should not be generalized beyond the investigated alloys.
  • AI role: The team mined published experimental composition/property results to train an artificial neural network for composition optimization toward lower transition temperatures. Predictions guided selection of Mn₁.₇₀Fe₀.₃₀P₀.₆₃Si₀.₃₇, then a minor Co-for-Fe substitution was tested experimentally to reduce its transition temperature below 77 K. AI guided composition selection; the reported magnetic and caloric properties came from measurements on synthesized samples.

Evidence sources: https://www.sciencedirect.com/science/article/pii/S1359645422003238

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, 52 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.

Machine learning and genetic search predict stronger refractory alloys

A random-forest model trained on published refractory high-entropy-alloy data was coupled to a stochastic genetic algorithm to optimize alloy compositions for yield strength at selected temperatures. For an AlMo0.5NbTa0.5TiZr base composition, the authors report modeled gains of 90 MPa at 25 °C and 13% at 1,000 °C for optimized compositions. Other base-alloy optimizations are also reported. The proposed alloys were not experimentally fabricated or mechanically measured in this study; these are computational design predictions.

Technology: Random-forest regression; stochastic genetic algorithm

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 validated against published experimental yield-strength data, including previously unseen datasets; the proposed optimized compositions and their target-temperature improvements were computational predictions.
  • Still unknown: The proposed optimized alloys were not experimentally fabricated or measured in this study; predicted yield strength depends on the source dataset and model assumptions.
  • AI role: A random-forest model predicted yield strength and a stochastic genetic algorithm searched compositions to maximize strength at specified temperatures.

Evidence sources: https://assets-eu.researchsquare.com/files/rs-1666028/v1_covered.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: Research Square primary v1 full text and posted-date metadata.

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.

EMBL

Model of the human nuclear pore with AlphaFold2 and electron microscopy

AlphaFold2 and ColabFold, combined with cryo-tomography and cryo-electron microscopy, produced a model of the human nuclear pore complex (about 30 proteins, about 1,000 copies, 120 MDa) covering over 90% of the complex, versus 46% for previous models. Flexible parts remain unresolved; a complete dynamic model is future work.

Technology: AlphaFold2 and ColabFold

Family AlphaFold: 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 integrative model (AlphaFold2 with cryo-ET and cryo-EM data) was validated against experimental data and covers more than 90% of the NPC core versus 46% for previous models.
  • Still unknown: Flexible peripheral parts are not modelled, no quantitative confidence scores are given, and a complete dynamic model is future work.
  • AI role: AlphaFold2 and ColabFold predicted protein structures that were assembled into the nuclear pore model.

Evidence sources: https://www.embl.org/news/science/puzzling-out-the-structure-of-a-molecular-giant/ · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1126/science.abm9506&resultType=core&format=json

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, 360 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.

X-ray machine-learning analysis reveals opposing Cd/Re displacements and Re 5d involvement in Cd₂Re₂O₇ structural order

Unsupervised X-ray diffraction temperature clustering (X-TEC) analyzed high-volume temperature-dependent diffraction data from Cd₂Re₂O₇. The extracted selection rules indicate that Cd and Re displacements are approximately equal in amplitude and opposite in phase, revealing a previously unrecognized involvement of Re 5d electrons in the structural order. The analysis also resolves X-ray diffraction evidence for fluctuations associated with the structural transitions. This is a data-driven interpretation of measured diffraction, not a new direct imaging of atom trajectories.

Technology: X-ray diffraction temperature clustering (X-TEC), an interpretable unsupervised machine-learning method using Gaussian-mixture clustering and reciprocal-space label smoothing

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: Unsupervised X-ray diffraction temperature clustering (X-TEC) analyzed high-volume temperature-dependent diffraction data from Cd₂Re₂O₇.
  • Still unknown: The displacement relation and Re 5d involvement are physical interpretations inferred from XRD selection rules and X-TEC clustering, not direct atom-by-atom imaging. The work reports XRD evidence for a structural Goldstone mode; it does not claim that all Goldstone-mode physics in Cd₂Re₂O₇ was previously unknown. The findings concern the analyzed crystal and temperature-dependent synchrotron diffraction data.
  • AI role: X-TEC clustered temperature trajectories of diffraction intensities across a high-volume reciprocal-space dataset to identify structural order parameters and weaker diffuse-scattering fluctuations. Its data-driven selection rules exposed the reciprocal-space pattern from which researchers inferred approximately equal, out-of-phase Cd and Re displacements and the Re 5d contribution; the manual crystallographic analysis would have required selecting only a small number of peaks.

Evidence sources: https://eunahkim.lassp.cornell.edu/sites/kim/files/publications/Harnessing%20interpretable%20and%20unsupervised%20machine%20learning%20to%20address%20big%20data%20from%20modern%20X-ray%20diffraction.pdf

Source 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, 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.

EQTransformer-based catalog reveals a 50-km-deep outer-rise fault at the Southern Mariana trench

Applying EQTransformer phase detection to near-field ocean-bottom seismometer recordings collected near the Challenger Deep yielded more than 1,975 earthquakes. Relocation and waveform-similarity clustering outlined a southern Mariana outer-rise fault extending approximately 20–50 km below the seafloor (about 50 km maximum depth), inferred to be a reactivated normal fault; the depth suggests more incoming-plate water input than prior estimates.

Technology: EQTransformer machine-learning seismic phase detector

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 EQTransformer phase detection to near-field ocean-bottom seismometer recordings collected near the Challenger Deep yielded more than 1,975 earthquakes.
  • Still unknown: The fault depth, fault type and hydration implications are inferred from a localized seismic cluster, waveform similarity, and regional tomography; focal mechanisms could not be robustly resolved, and the authors state that limited station coverage does not establish whether such deep faults are widespread.
  • AI role: Detected and picked P- and S-wave phases in near-field ocean-bottom seismometer data, producing a denser earthquake catalog. Researchers manually reviewed selected phases, relocated the events and used waveform-similarity clustering to delineate the fault.

Evidence sources: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022GL097779

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, 31 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.

ColabFold: free protein folding with AlphaFold2 or RoseTTAFold

ColabFold pairs AlphaFold2 or RoseTTAFold with fast MMseqs2 homology search, running free on Google Colab. Beyond benchmark comparisons, the paper reports predicting structures for 1,762 Methanocaldococcus jannaschii proteins in 48 hours on one GPU and using predicted complexes to help determine the 120 MDa human nuclear-pore structure with cryo-electron microscopy. The proteome structures are computational predictions, not experimentally validated structures. The nuclear-pore application overlaps the separate nuclear-pore entry; this is a methodological contribution, not an additional independent biological finding.

Technology: ColabFold (AlphaFold2, RoseTTAFold, MMseqs2)

Family AlphaFold: 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: Throughput was measured (close to 1,000 structures per day on one GPU) and predicted complexes were used alongside cryo-EM to help determine the human nuclear-pore structure.
  • Still unknown: Not reported in the abstract read.
  • AI role: AlphaFold2 or RoseTTAFold, paired with MMseqs2 homology search, run free on Google Colaboratory.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41592-022-01488-1&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: 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, 9430 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.

Machine learning suggests a more general theory of microlensing

An algorithm highlighted microlensing configurations not explained by usual interpretations, leading to a theory unifying two degeneracies. The authors report a better fit to data from two dozen previous discoveries; they describe the theory as probably incomplete.

Technology: Machine-learning 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: The machine-learning-derived theory was checked against the datasets of two dozen earlier microlensing discoveries and fit them better than the earlier interpretations.
  • Still unknown: The authors say the broader theory is likely still incomplete, and the multi-planet case is not yet worked out.
  • AI role: Algorithm highlighted microlensing configurations not explained by the usual interpretations.

Evidence sources: https://news.berkeley.edu/2022/05/24/ai-reveals-unsuspected-math-underlying-search-for-exoplanets/ · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41550-022-01671-6&resultType=core&format=json

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 88 of works of the same type, year and subfield (OpenAlex work metrics, 40 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.

Learned cluster expansion guides an acidic oxygen-evolution catalyst

A fitted energy model and quantum-inspired optimizer selected a mixed-oxide catalyst family for laboratory testing. The study reports a Ru-Cr-Mn-Sb oxide catalyst with higher ruthenium-normalized activity than its RuO2 comparator and sustained operation in the tested setup. This is one bounded materials result, not commercial electrolyzer validation or evidence that the optimizer itself is a learned model. The catalyst result already appears in the May 2022 preprint.

Technology: DFT-trained cluster expansion with quantum-inspired search

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: Selected oxide catalysts received electrochemical activity and stability measurements.
  • Still unknown: Scale-up, independent replication and general design success rates were not established.
  • AI role: DFT-trained cluster expansion with quantum-inspired search supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://arxiv.org/abs/2205.09007v1 · https://light.northwestern.edu/wp-content/uploads/2023/01/1-s2.0-S2590238522006622-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: Primary v1 preprint abstract/history; indexed publisher article and institutional publisher-PDF passages on model fitting and affiliations; supplements 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.

Machine learning model predicts mental health crises from health records

A model using electronic health records to flag mental health crises over 28 days reached an AUROC of 0.797, with 58% sensitivity at 85% specificity. In a six-month prospective workflow study at one NHS trust, clinicians rated predictions useful for managing caseloads or mitigating crisis risk in 64% of cases. This measures clinician-reported usefulness, not a demonstrated reduction in crises; generalizability remains limited by the single-trust cohort, and precision was modest (AUPRC 0.159). The study includes authors affiliated with the developer Koa Health.

Technology: Machine learning model on electronic health records

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: A six-month prospective workflow study at one NHS trust asked clinicians to rate predictions, which were judged useful for managing caseloads or mitigating crisis risk in 64% of cases; the AUROC was 0.797.
  • Still unknown: Not reported; the page gives no rating scale or AUROC for the prospective usefulness measure.
  • AI role: Model using electronic health records to monitor patients for risk of a mental health crisis over 28 days.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-022-01811-5&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, 268 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.

Machine-learned design of CuPd nanocubes for electrocatalytic nitrate reduction to ammonia

Interpretable machine learning identified ordered B2 CuPd intermetallics whose adsorption energies break the usual scaling relations. The team synthesized monodisperse CuPd nanocubes that reached a 92.5% Faradaic efficiency at -0.5 V RHE for nitrate reduction to ammonia, with a yield of 6.25 mol h-1 g-1 at -0.6 V. Performance was measured in the authors' own electrochemical tests, and the design rules come from a limited set of predicted intermetallics.

Technology: Interpretable machine learning on d-state Pauli repulsion descriptors (DFT-based)

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 synthesized monodisperse ordered B2 CuPd nanocubes and measured nitrate reduction performance in their own electrochemical tests (92.5% Faradaic efficiency at -0.5 V RHE).
  • Still unknown: Not reported.
  • AI role: Interpretable machine learning identified ordered B2 CuPd intermetallics whose adsorption energies break scaling relations and provided design rules.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41467-022-29926-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: 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, 466 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.

Knowledge-graph-guided discovery of antibiotic-resistance associations

KIDS combined an expert-curated knowledge graph with learned link-prediction models to rank gene–antibiotic resistance hypotheses for experiments. The authors report 93 validated associations involving 83 E. coli genes and 15 antibiotics, including six genes they describe as previously unknown resistance genes; five tested Salmonella homologs also showed resistance. These are laboratory gene–drug associations, not clinical validation. The article’s second-round percentage conflicts with its counts (29/90 is 32.2%, although the text states 28.8%). Count one study.

Technology: Knowledge graph with PRA, multilayer perceptron, AdaBoost-stacked model, TransE and TransD

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 wet-lab validation of 93 gene–antibiotic associations, including five tested Salmonella homologs.
  • Still unknown: Reported percentages are internally inconsistent for 29 of 90 second-round tests; these assays do not establish clinical relevance or treatment outcomes.
  • AI role: KIDS ranked gene–antibiotic links from an expert-curated graph with learned link-prediction models, directing experiments to candidate resistance associations.

Evidence sources: https://www.nature.com/articles/s41467-022-29993-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: 16 Nov 2026. Review scope: Primary Nature indexed full text, article dates, affiliations, methods/results, contributions and stated counting discrepancy..

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.

FAST-PETase: an AI-engineered enzyme degrades PET waste

A machine-learning model suggested PETase mutations, then tested on 51 post-consumer containers, fibers and fabrics. FAST-PETase degrades PET below 50 °C, in some cases down to monomers within 24 hours; production at scale and environmental use still needed development.

Technology: MutCompute / 3D convolutional network

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: FAST-PETase was tested on 51 post-consumer containers, fibers and fabrics, degrading PET below 50 °C, in some cases to monomers within 24 hours.
  • Still unknown: Scale-up, field use at polluted sites and cost are not demonstrated, and the work covers only PET.
  • AI role: Machine-learning model suggested enzyme mutations.

Evidence sources: https://news.utexas.edu/2022/04/27/plastic-eating-enzyme-could-eliminate-billions-of-tons-of-landfill-waste/ · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-022-04599-z&resultType=core&format=json

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, 1271 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.

Swarm learning predicts colorectal cancer mutations from histopathology slides

Swarm learning, which trains models across partner sites without moving patient data, was applied to over 5,000 patients to predict BRAF status and microsatellite instability from H&E colorectal cancer slides. Swarm-trained models outperformed most locally trained models and matched models trained on merged datasets, and were validated on two independent UK datasets. Caveat: the outcome is a research-grade prediction from slides, not a clinical test.

Technology: Swarm learning with deep learning on histopathology images

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: Swarm-trained models for BRAF status and microsatellite instability from H&E slides were compared with local and merged-data models and validated on two independent UK datasets.
  • Still unknown: Not reported; the abstract includes no limitations statement.
  • AI role: Swarm learning trained AI models across partner sites without moving patient data, predicting molecular alterations from colorectal cancer histopathology slides.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-022-01768-5&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, 237 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.

Murmurations: unexpected oscillations in elliptic-curve data

Classification experiments and analysis of elliptic-curve data show oscillations in the mean of Frobenius traces that vary with rank. A numerical observation, not proof of a general law; the 2022 preprint was later published in Experimental Mathematics in 2024.

Technology: Logistic regression and PCA

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: Classification experiments and analysis of elliptic-curve data show oscillations in the average Frobenius trace that vary with rank.
  • Still unknown: Not reported in the arXiv abstract read.
  • AI role: Machine-learning classification experiments were used to classify elliptic curves by rank.

Evidence sources: https://arxiv.org/abs/2204.10140

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 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-guided microscopy characterizes an atypical ataxia presentation

Unsupervised image analysis detected cellular patterns not apparent by inspection in an investigation of one family with an atypical ataxia presentation. Additional cell experiments found altered ataxin-7 behavior and mitochondrial bioenergetic deficits. The proposed combined role of ATXN7 and TOP1MT variants remains a hypothesis: unaffected parents carried individual variants, and fly models did not reproduce the patient phenotype.

Technology: PhenoRipper unsupervised image-pattern 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: Patient-derived cells, functional measurements and fly models were examined.
  • Still unknown: Clinical causality, variant interaction and diagnostic utility are not established by this small study.
  • AI role: PhenoRipper unsupervised image-pattern analysis supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://www.nature.com/articles/s41439-022-00188-8

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 date: 16 Nov 2026. Review scope: Publisher abstract, selected image-analysis Methods, Results, Discussion, dates and affiliations; supplement 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.

Random-forest alert screening leads to spectroscopic confirmation of four changing-look AGNs

A hierarchical random-forest classifier in the ALeRCE broker applied to Zwicky Transient Facility alert-stream light curves helped select candidate changing-look active galactic nuclei from objects previously classified as type 2. New spectra confirmed significant changes in broad Balmer emission in four of six observed candidates, establishing new type 2-to-type 1 transitions. The other two observed objects had only tentative evidence and are not counted as confirmed transitions.

Technology: ALeRCE light-curve classifier (hierarchical imbalanced random forest)

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 hierarchical random-forest classifier in the ALeRCE broker applied to Zwicky Transient Facility alert-stream light curves helped select candidate changing-look active galactic nuclei from objects previously classified as type 2.
  • Still unknown: Only six of the 30 visually retained candidates received new spectra in this paper, selected partly for observability; four showed clear >3σ broad-line changes and two were tentative. Thus the reported ≥66% success rate is a small, selected follow-up sample, not an unbiased prevalence estimate. The authors note possible 30–50% misclassification in the parent type-2 sample and that the ZTF alert threshold makes the method incomplete.
  • AI role: The ALeRCE random-forest classifier used variability features from ZTF light curves and ZTF/AllWISE photometric colors to classify alert sources as AGN-like. Its classifications were cross-matched against a pre-existing spectroscopic type-2 AGN sample to produce candidates with current type-1-like variability; this AI-assisted shortlist enabled follow-up spectroscopy that confirmed the transitions.

Evidence sources: https://academic.oup.com/mnrasl/article/513/1/L57/6563628

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.

Learned phase models identify synergistic photoanode mixtures

Learning from measured diffraction and photocurrent data identified Bi-Cu-V oxide phase mixtures whose photoactivity exceeded an additive non-interacting-phase model. BiVO4-like phases showed beneficial associations with copper-vanadate phases. This is an experimentally grounded composition-property result; the underlying electronic, catalytic and morphological mechanisms remain unresolved, and device-level solar-fuel efficiency was not established.

Technology: Deep Reasoning Networks and matrix factorization

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: Measured photocurrents and phase distributions were compared with a learned additive baseline.
  • Still unknown: Independent replication, detailed mechanisms and practical device performance were not assessed.
  • AI role: Deep Reasoning Networks and matrix factorization supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://www.nature.com/articles/s41524-022-00747-1

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: 16 Nov 2026. Review scope: Indexed primary publisher abstract, selected phase-model and measured-synergy Results, date and affiliations; supplementary material 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.

Active-learning experiments reveal distinct on-field and off-field polarization-switching behavior in ferroelectrics

An active-learning workflow paired with automated piezoresponse-force microscopy discovered relationships between local ferroelectric domain structure and polarization-switching hysteresis. The experimentally selected sampling paths and points differed substantially between on-field and off-field hysteresis loops, indicating that the two measurement modes are dominated by different mechanisms. This is a bounded result from the reported ferroelectric sample and spectroscopy workflow, not a universal mechanism established for all ferroelectrics.

Technology: Machine-learning active-learning framework for automated piezoresponse force microscopy and spectroscopy

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 active-learning workflow paired with automated piezoresponse-force microscopy discovered relationships between local ferroelectric domain structure and polarization-switching hysteresis.
  • Still unknown: The authors report experimental correlation and different on-field/off-field discovery paths, interpreting this as evidence that the measurement modes are dominated by different mechanisms. The study does not establish a universal causal account across materials; it reports a specific automated PFM/spectroscopy investigation.
  • AI role: The model learned correlations between high-dimensional local domain structure and hysteresis-loop switching characteristics, and used loop descriptors and uncertainty to guide the automated microscopy and spectroscopy measurements. Human physics insight was encoded in the acquisition function.

Evidence sources: https://arxiv.org/abs/2108.06037

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, 143 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.

Deep learning assesses heart allograft rejection from biopsy whole-slide images

A deep learning system analysing gigapixel endomyocardial biopsy slides detected allograft rejection with an AUC of 0.962, distinguished cellular and antibody-mediated rejection (AUCs 0.958 and 0.874) and graded rejection with an AUC of 0.833. In a human reader study it was non-inferior to conventional assessment and reduced interobserver variability and assessment time. Caveat: test cohorts from the US, Turkey and Switzerland are retrospective and the authors call for clinical trials.

Technology: Deep learning on whole-slide images for endomyocardial biopsies

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: Rejection detection (AUC 0.962) was measured on a curated US dataset and independent test cohorts from Turkey and Switzerland, and a human reader study found the system non-inferior to conventional assessment.
  • Still unknown: The authors state that clinical trials are needed to establish the efficacy of AI-assisted biopsy assessment.
  • AI role: Deep learning system analysed gigapixel endomyocardial biopsy whole-slide images to detect, subtype and grade rejection.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41591-022-01709-2&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 10% of works of the same type, year and subfield (OpenAlex work metrics, 105 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.

LSTM-designed peptides show antibacterial activity and mixed efficacy in a mouse sepsis model

A generative long short-term-memory recurrent neural network produced 198 candidate antimicrobial peptides; five were synthesized. PEP-38 and PEP-137 inhibited carbapenem-resistant Klebsiella isolates in vitro, and PEP-137 also inhibited Pseudomonas aeruginosa. In a small mouse K. pneumoniae sepsis experiment, PEP-137 yielded 50% survival, PEP-38 did not improve survival, and the in-vitro-inactive PEP-36 control unexpectedly yielded 66.7% survival.

Technology: Generative long short-term memory recurrent neural network trained on antimicrobial peptide sequences

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 small mouse K. pneumoniae sepsis experiment, PEP-137 yielded 50% survival, PEP-38 did not improve survival, and the in-vitro-inactive PEP-36 control unexpectedly yielded 66.7% survival.
  • Still unknown: Only 5/198 AI-generated peptides were synthesized. The animal experiment used one administration, small groups (12 mice per group) and a simple model; PEP-36's unexpected survival effect despite no in-vitro antibacterial activity complicates interpretation. PEP-38 did not improve survival; no mechanism, pharmacokinetics, pharmacodynamics or safety was established. These are exploratory preclinical results, not a treatment claim.
  • AI role: The LSTM RNN generated 198 peptide sequences, which were further filtered in silico; five selected candidates were synthesized and assayed. Two showed in-vitro activity against carbapenem-resistant isolates, and two were advanced into a mouse sepsis study.

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC8944797/

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, 26 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.

Machine-learning selection uncovers plant enzymes for alkaloid pathway branches

Support-vector-machine ranking selected poppy homologs whose experimental characterization identified aromatic acetaldehyde synthase activity in PsTyDC6 and phenylpyruvate-decarboxylase activity in PsPDC1. In vitro and engineered E. coli metabolomics detected pathway products including norcoclaurine and reticuline, demonstrating alternative routes from tyrosine into benzylisoquinoline alkaloids.

Technology: High-dimensional support vector machine (SVM) enzyme-sequence 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: Support-vector-machine ranking selected poppy homologs whose experimental characterization identified aromatic acetaldehyde synthase activity in PsTyDC6 and phenylpyruvate-decarboxylase activity in PsPDC1.
  • Still unknown: The study demonstrates enzyme activity and pathway products in vitro and engineered E. coli; it does not establish the exact physiological roles of these enzymes in living poppy plants. The paper describes several enzyme activities as potential and says natural pathway roles need further elucidation.
  • AI role: SVM models ranked more than 100 homologous candidate sequences for specialized aromatic-acetaldehyde-synthase and phenylpyruvate-decarboxylase functions. The team selected high-scoring sequences for in-vitro enzyme assays and metabolomics of reconstructed E. coli pathways.

Evidence sources: https://www.nature.com/articles/s41467-022-28883-8

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.

Citations: at percentile 89 of works of the same type, year and subfield (OpenAlex work metrics, 41 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.

Google DeepMind

Ithaca: restoring and dating ancient Greek inscriptions

A neural network reconstructs missing text in damaged Greek inscriptions, identifies their provenance and dates them: 62% accuracy in restoration, 71% in location, and dating within 30 years of reference ranges. With Ithaca, historians improved from 25% to 72% correct restorations. It is intended to assist historians.

Technology: Ithaca (deep neural network)

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: Restoration (62%), location (71%) and dating were checked on held-out dated inscriptions, and a historian study rose from 25% to 72% correct restorations with the tool.
  • Still unknown: Accuracy is far from perfect, dating is a range within about 30 years, and the model was trained only on Greek inscriptions.
  • AI role: Neural network restored missing text, identified provenance and dated Greek inscriptions.

Evidence sources: https://deepmind.google/discover/blog/predicting-the-past-with-ithaca/ · https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI%3A10.1038/s41586-022-04448-z&resultType=core&format=json

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, 268 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.

Unsupervised learning guides high-zT ScNiSb-based half-Heusler thermoelectrics

Iterative unsupervised clustering of half-Heusler compounds, using known thermoelectric materials as labeled reference points, guided selection of ScNiSb-based compositions for experimental testing. The team measured peak zT ≈0.5 at 925 K for p-type Sc₀.₇Y₀.₃NiSb₀.₉₇Sn₀.₀₃ and ≈0.3 at 778 K for n-type Sc₀.₆₅Y₀.₃Ti₀.₀₅NiSb. The paper reports these as compositionally optimized thermoelectric properties, not a new crystal structure or a device-level demonstration.

Technology: Iterative unsupervised clustering using prototype-based, density-based and hierarchical machine-learning 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: Iterative unsupervised clustering of half-Heusler compounds, using known thermoelectric materials as labeled reference points, guided selection of ScNiSb-based compositions for experimental testing.
  • Still unknown: The measured materials are doped/alloyed variants of the known ScNiSb half-Heusler parent; the article does not claim discovery of a new crystal structure. Reported zT values are laboratory measurements at elevated temperatures, not performance from a fabricated waste-heat device or a commercial deployment. The two compositions are part of one paper-level screening and experimental result; do not split them into separate records.
  • AI role: The authors applied unsupervised clustering to Materials Project descriptors for 456 half-Heusler compounds, iteratively using clusters containing 20 known thermoelectric materials as reference groups to shortlist candidates. The resulting ScNiSb-based choices guided synthesis and transport-property measurements; the model supported candidate selection, while the researchers performed synthesis, characterization and interpretation.

Evidence sources: https://www.nature.com/articles/s41524-022-00723-9

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.

Citations: in the top 10% of works of the same type, year and subfield (OpenAlex work metrics, 110 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.

Learned activity models improve viral diagnostic assay designs

Learned diagnostic-activity scores guided assay design across viral variation. Experiments on selected synthetic targets supported improved detection compared with the studied conservation-based designs. The larger cross-species design catalog is computational; it is not thousands of validated diagnostics. Patient-specimen sensitivity, clinical specificity and field performance were not established.

Technology: ADAPT learned sequence-activity classifier and regressor

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: Selected assay designs were tested on synthetic viral targets.
  • Still unknown: Clinical and environmental sample performance require further validation.
  • AI role: ADAPT learned sequence-activity classifier and regressor supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC9287178/

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 article metadata, affiliations, selected design Methods, experimental Results and limitations; supplements 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.

Machine-learning-guided screening identifies Bcl-2-family inhibitors IS20 and IS21

Machine-learning-guided virtual screening followed by surface-plasmon-resonance and biological assays identified IS20 and IS21 as compounds binding anti-apoptotic Bcl-2, Bcl-xL and Mcl-1 proteins. Both affected cancer-cell viability and induced apoptosis/autophagic flux in tested cell models; IS21 also reduced tumor growth in leukemia and melanoma mouse models.

Technology: Machine-learning-guided virtual screening (model family not named in PubMed abstract)

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: Machine-learning-guided virtual screening followed by surface-plasmon-resonance and biological assays identified IS20 and IS21 as compounds binding anti-apoptotic Bcl-2, Bcl-xL and Mcl-1 proteins.
  • Still unknown: The article calls the compounds potential/new leads; findings are biochemical, cell-based and mouse-model preclinical evidence. It does not establish clinical effectiveness. The PubMed abstract does not specify the machine-learning algorithm, so no specific model architecture is attributed here.
  • AI role: Machine-learning-guided virtual screening prioritized putative Bcl-2-family inhibitor compounds; the candidate hits were then evaluated by surface plasmon resonance and in vitro cell assays, and IS21 was evaluated in mouse tumor models. The reported outcome is the specific assay characterization of IS20 and IS21.

Evidence sources: https://pubmed.ncbi.nlm.nih.gov/35265218/

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, 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.

FourCastNet: high-resolution global forecasts with neural networks

The preprint presents global forecasts at 0.25° resolution and a one-week forecast in under two seconds. Comparison with IFS is favorable at short lead times for large-scale variables and some finer-scale variables; this does not mean general superiority or operational use.

Technology: Adaptive Fourier Neural Operators

Family FourCastNet: 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: Forecasts at 0.25 degrees are compared with ECMWF IFS, matching it at short lead times for large-scale variables and outperforming it for some fine-scale variables such as precipitation.
  • Still unknown: Not reported in the arXiv abstract read.
  • AI role: Adaptive Fourier neural operator produced global weather forecasts.

Evidence sources: https://arxiv.org/abs/2202.11214

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 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.

Tokamak plasma control with reinforcement learning

A single neural network controls the magnetic coils of Lausanne’s TCV tokamak and forms plasmas in different configurations (snowflake, ITER-like shape, double droplet). Trained mainly in simulation and then tested on the real machine. Stated limitations: plasma lasts about three seconds and the machine changes from day to day.

Technology: Deep reinforcement 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: On the TCV tokamak the controller produced and held elongated, conventional, negative-triangularity and snowflake configurations, with accurate tracking of location, current and shape, and sustained two separate 'droplet' plasmas.
  • Still unknown: Transfer to larger or reactor-scale devices is not reported in the abstract.
  • AI role: A deep reinforcement-learning architecture learned to command the full set of control coils from high-level objectives.

Evidence sources: https://doi.org/10.1038/s41586-021-04301-9

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, 907 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 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.

Knowledge-graph learning prioritizes experimentally examined Alzheimer genes

A learned knowledge-graph model nominated candidate Alzheimer-associated genes for neuronal, autopsy and cell-assay follow-up. The study highlights candidate associations rather than establishing causes of human disease. Small samples, noisy negative labels and an internal discrepancy in reported knockdown-hit counts constrain interpretation; those counts are not treated as reconciled.

Technology: MPxgb(AD) knowledge-graph XGBoost

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: Selected predicted genes received biological screening.
  • Still unknown: Independent validation, human causality and therapeutic benefit remain unestablished.
  • AI role: MPxgb(AD) knowledge-graph XGBoost supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://www.nature.com/articles/s42003-022-03068-7 · https://www.nature.com/articles/s42003-022-03068-7.pdf · https://pubmed.ncbi.nlm.nih.gov/35149761/

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 publisher PDF abstract and model description, publisher date and PubMed affiliations; detailed prior Methods review retained, direct PMC retrieval blocked..

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 subgroup rules guide higher-component oxidation catalysts

Rules learned from high-throughput catalyst measurements guided tests of more complex four- and five-component materials for propylene oxidation. Selected materials reached approximately 60% oxygenate yield in laboratory tests. The learned descriptor ranges are associative design guides, not proven atomic mechanisms or universal industrial performance rules.

Technology: Subgroup-discovery rule 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: More complex catalyst compositions were tested against rules learned from three-component measurements.
  • Still unknown: Independent replication and industrial durability were not assessed; temperature rules did not transfer unchanged.
  • AI role: Subgroup-discovery rule learning supported the stated study; experimental and interpretive work remained with researchers.

Evidence sources: https://doi.org/10.1021/acscatal.1c04793 · https://pure.mpg.de/pubman/item/item_3369647_4/component/file_3370185/acscatal.1c04793.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: Indexed primary publisher text and deposited publisher PDF abstract and selected experimental-validation Results; supplements 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.

HydrAMP generates antimicrobial peptide analogues validated in bacterial assays

HydrAMP, a conditional variational autoencoder with antimicrobial activity classifiers, generated peptide analogues and de novo candidates. The 2022 preprint reports wet-lab confirmation of increased activity for a generated Pexiganan-based analogue; the 2023 journal version reports testing across five bacterial strains, including nine analogues of clinical prototypes and six analogues of an inactive peptide. The tested panel was small and model/human filtered; it does not establish therapeutic efficacy.

Technology: Conditional variational autoencoder with classifier-guided peptide generation and molecular-dynamics-based candidate preselection

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: Generated peptide analogues were experimentally evaluated for antimicrobial activity across five bacterial strains; the journal abstract reports high activity for nine generated analogues of clinical prototypes and six analogues of an inactive peptide.
  • Still unknown: The tested panel is selected and small; these are in-vitro assays and do not establish safety, therapeutic efficacy or clinical benefit. The 2023 correction replaced a duplicated generation-statistics figure, not the wet-lab result stated here.
  • AI role: A conditional VAE generated peptide analogues and unconstrained candidates from antimicrobial activity conditions; ranking and molecular-dynamics preselection helped choose candidates for experimental validation.

Evidence sources: https://www.biorxiv.org/content/10.1101/2022.01.27.478054v1 · https://www.nature.com/articles/s41467-023-36994-z · https://www.nature.com/articles/s41467-023-40879-6.pdf

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/metadata; primary Nature Communications indexed full article including methods/results; primary publisher author 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.

Machine learning identifies new CYP2C9-inhibiting drugs and drug metabolites

A structure-, dynamics- and physicochemical-feature machine-learning approach predicted CYP2C9 inhibition and experimentally identified four drugs—vatalanib, piriqualone, ticagrelor and cloperidone—as strong inhibitors, plus sertindole, asapiprant, duvelisib and dasatinib as moderate inhibitors in reported assays. Vatalanib had a reported CYP2C9 inhibition IC50 of 0.067 μM. Separate metabolism assays characterized CYP2C9-produced metabolites for abemaciclib, cloperidone, vatalanib and tarafenacin.

Technology: Machine-learning inhibitor predictors combining CYP2C9 structural/dynamics information with inhibitor physicochemical 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: A structure-, dynamics- and physicochemical-feature machine-learning approach predicted CYP2C9 inhibition and experimentally identified four drugs—vatalanib, piriqualone, ticagrelor and cloperidone—as strong inhibitors, plus sertindole, asapiprant, duvelisib and dasatinib as moderate inhibitors in reported assays.
  • Still unknown: Inhibition and metabolite findings are from the reported in vitro systems and do not by themselves establish clinically relevant drug–drug interactions, exposure changes, or patient harm. Reported inhibition strengths are assay-specific; clinical relevance depends on dose, exposure and context and requires pharmacokinetic/clinical assessment.
  • AI role: The authors used machine-learning models to prioritize drug-like compounds for CYP2C9 inhibition testing. Follow-up in vitro assays identified the listed strong and moderate inhibitors; CYP2C9 metabolism assays separately characterized metabolites of four drugs.

Evidence sources: https://pubmed.ncbi.nlm.nih.gov/35081108/

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 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.

AMPlify prioritizes bullfrog-genome antimicrobial peptides active against priority pathogens

Researchers integrated the attentive deep-learning classifier AMPlify into a genome-mining pipeline for North American bullfrog sequences. Of 101 candidate mature peptides generated from genome-annotated precursors, AMPlify predicted 75 as antimicrobial; 11 candidates were synthesized and tested, with four showing activity against tested bacteria, including a multidrug-resistant carbapenemase-producing E. coli isolate.

Technology: AMPlify, an attentive deep-learning antimicrobial-peptide sequence 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: Of 101 candidate mature peptides generated from genome-annotated precursors, AMPlify predicted 75 as antimicrobial; 11 candidates were synthesized and tested, with four showing activity against tested bacteria, including a multidrug-resistant carbapenemase-producing E. coli isolate.
  • Still unknown: This is a small selected validation set: 11 of 75 model-predicted candidates were tested and four were active. Results are in vitro and do not establish therapeutic efficacy or safety. Candidate precursor identification also used homology/genome annotation; attribute AMPlify specifically to mature-peptide prioritization, not the whole discovery workflow.
  • AI role: After bioinformatics identified candidate peptide precursors from the bullfrog genome, AMPlify scored 101 mature peptide sequences and nominated putative AMPs. The team selected 11 predicted peptides for synthesis and antimicrobial susceptibility testing; four showed activity against one or more of six tested bacterial strains, including the carbapenemase-producing E. coli clinical isolate.

Evidence sources: https://pmc.ncbi.nlm.nih.gov/articles/PMC8788131/

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, 250 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.

Machine learning screening selects an energetic material for laboratory characterization

A high-throughput workflow combined learned molecular-property predictions and crystal-packing classification to screen 25,112 generated structures and select ICM-104. The authors synthesized and characterized the selected material, reporting that its measured properties agreed with the model-assisted predictions. This is one laboratory-scale, study-specific material result; the broad candidate space remains computational, and the study does not establish scale-up, deployment or independent replication.

Technology: Machine-learning property regressors and a convolutional neural network for crystal-packing 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 study screened 25,112 generated candidates, selected one target material, then synthesized and characterized it; the publisher reports measured properties in agreement with predictions.
  • Still unknown: Only the selected target was synthesized and characterized. Scale-up, independent replication and operational performance are not established. Some energetic-performance values in the paper are calculated rather than directly measured.
  • AI role: Learned molecular-property models and a CNN packing classifier helped rank and select the target for synthesis and characterization.

Evidence sources: https://www.engineering.org.cn/engi/EN/1159869469231210541 · https://www.engineering.org.cn/engi/CN/1159869469231210541

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: Official publisher English article text/indexed conclusions and Chinese primary metadata/abstract reviewed; a saved 2026-11-16 review notes an author-hosted primary PDF was also checked..

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.

Cell2location: mapping fine-grained cell types in spatial transcriptomics

Cell2location is a Bayesian model that resolves fine-grained cell types in spatial transcriptomics data by accounting for technical variation and sharing information across locations. Applied to three tissues, it revealed regional astrocyte subtypes in mouse brain, a rare pre-germinal center B cell population in human lymph node, and fine immune populations in human gut lymphoid follicles. Caveat: the reported results are biological findings from the method's application, with no single quantitative accuracy figure confirmed in the abstract.

Technology: Cell2location (Bayesian deep probabilistic 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: Cell2location was assessed in three tissues (mouse brain, human lymph node, human gut) and reported to give higher sensitivity and resolution than existing tools.
  • Still unknown: The abstract names no conditions under which the claimed higher sensitivity holds and no limitations.
  • AI role: A Bayesian model that resolves fine-grained cell types in spatial transcriptomics data.

Evidence sources: https://www.ebi.ac.uk/europepmc/webservices/rest/search?query=DOI:10.1038/s41587-021-01139-4&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, 1414 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.

Machine learning guides discovery of hard Fe–Nb–B metallic-glass films

Random-forest models for glass formation and hardness helped select Fe–Nb–B composition regions for sputtered-film experiments. The study reports a maximum measured hardness of 29 GPa for Fe₃Nb₂₅B₇₂ and an H/E ratio of 0.09 for Fe₂₀Nb₁₂B₆₈. H/E is used as an estimate of wear resistance, not a direct wear test. Samples were 500–800 nm films with substrate-effect corrections; bulk properties and industrial production were not demonstrated. This extends earlier machine-learning/high-throughput work on metallic glasses and is one study.

Technology: Random-forest models for glass formation and hardness

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: Nanoindentation and synchrotron characterization were used to test sputtered films and measure hardness and structure.
  • Still unknown: Thin-film results; H/E is an estimated wear-resistance metric, not direct wear testing or evidence of bulk/industrial performance.
  • AI role: Random-forest predictions helped select Fe–Nb–B composition regions for high-throughput sputtered-film synthesis and measurement.

Evidence sources: https://www.nrel.gov/docs/fy22osti/80853.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: Primary full article PDF, title/DOI/date, affiliations, methods, composition/result and measurement limitations..

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-learned multiscale simulations reveal lipid-dependent RAS organization at membranes

The MuMMI machine-learning multiscale simulation ensemble found that local membrane lipid fingerprints correlate with wild-type KRAS clustering and orientation, with many simulated RAS–RAS interfaces appearing nonspecific. Complementary liposome and diffusion experiments supported lipid-dependent RAS adhesion and dynamics, motivating the authors' conclusion that lipids can mediate RAS multimer organization.

Technology: MuMMI (Multiscale Machine-learned Modeling Infrastructure)

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: Complementary liposome and diffusion experiments supported lipid-dependent RAS adhesion and dynamics, motivating the authors' conclusion that lipids can mediate RAS multimer organization.
  • Still unknown: Most specific orientation and interface conclusions are simulation-derived; experimental comparisons provide validation of selected membrane adhesion/diffusion behaviors, not direct demonstration of these mechanisms in intact cells or cancer outcomes. Authors describe absolute diffusion values as semiquantitative and dependent on methods/force fields.
  • AI role: Machine learning selects diverse local membrane patches from a macroscale lipid model to launch and couple atomistic/coarse-grained simulations, producing an ensemble of more than 100,000 simulations. The resulting simulations identified lipid-composition relationships with RAS orientation and multimer organization; selected predictions were compared with surface-plasmon resonance, fluorescence-correlation and other biophysical experiments.

Evidence sources: https://www.pnas.org/content/pnas/119/1/e2113297119.full.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, 97 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.

Open the filtered catalog for charts and filters, or browse all organizations and model families.