{"architecture":{"family":"AlphaFold-derived network fine-tuned for variant classification","input":"protein sequence + substitution","output":"pathogenicity score","pretraining":"AlphaFold structure prediction, then weak labels from population allele frequencies (gnomAD, primates)"},"article":null,"cancer_slugs":["pan-cancer"],"category":"genomics","confidence":"high","datasets":[],"developer":"Google DeepMind","evaluation":[{"benchmark":"ClinVar held-out variants","external":true,"metric":"AUROC","source":"https://www.science.org/doi/10.1126/science.adg7492","value":"0.94 (paper)"}],"hf":null,"kind":"task-model","license":"Apache-2.0 (code); predictions CC-BY-NC-SA 4.0","limitations":"- Missense only; no indels, splice or non-coding variants.\n- Germline-oriented training signal; somatic driver status is a different question.","links":{"demo":null,"docs":null,"doi":"10.1126/science.adg7492","github":"https://github.com/google-deepmind/alphamissense","huggingface":null,"paper":"https://www.science.org/doi/10.1126/science.adg7492","pmid":null},"modalities":["genomics","protein-sequence"],"name":"AlphaMissense","notable_uses":"Classified 89% of all missense variants (32% likely pathogenic, 57% likely benign) versus ~0.1% annotated by human experts at the time; used as evidence in germline and tumour variant interpretation.","openness":"open-weights","regulatory":{"intended_use_en":"Research evidence for variant interpretation; not a diagnostic.","intended_use_pl":"Dow\u00f3d badawczy w interpretacji wariant\u00f3w; nie jest testem diagnostycznym.","source_url":"https://github.com/google-deepmind/alphamissense","status":"research-only"},"regulatory_status":"research-only","release_date":"2023-09-19","run_snippet":"","settings":["diagnosis","basic-research"],"slug":"alphamissense","sources":[{"label":"Cheng J et al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 2023","url":"https://www.science.org/doi/10.1126/science.adg7492"},{"label":"DeepMind blog \u2014 A catalogue of genetic mutations","url":"https://deepmind.google/blog/a-catalogue-of-genetic-mutations-to-help-pinpoint-the-cause-of-diseases/"}],"summary":"Classifies the pathogenicity of every possible single amino-acid substitution in the human proteome \u2014 71 million missense variants \u2014 by fine-tuning AlphaFold on population variant frequencies; a resource for interpreting variants of uncertain significance.","tasks":["variant-effect"],"training":{"institutions":"Google DeepMind","size":"predictions cover 71 million possible missense variants across the human proteome","summary_en":"AlphaMissense is an adaptation of AlphaFold fine-tuned on population frequency databases of human and primate variants. The authors state the model reaches its results on genetic and experimental benchmarks 'all without explicitly training on such data' \u2014 that is, without clinical pathogenicity labels as the training target; the training signal comes from which variants are observed, and at what frequency, in human and primate populations. ClinVar variants held out from training are used for evaluation (AUROC 0.94), not for training. Practical consequence for anyone reusing the model: the signal is oriented towards germline variation, so somatic driver status in a tumour is a different question and is not what the model was trained to answer.","summary_pl":"AlphaMissense to adaptacja AlphaFolda dostrojona na bazach cz\u0119sto\u015bci wyst\u0119powania wariant\u00f3w u ludzi i naczelnych. Autorzy zaznaczaj\u0105, \u017ce model osi\u0105ga swoje wyniki na testach genetycznych i eksperymentalnych \u201ebez jawnego trenowania na takich danych\" \u2014 to znaczy bez klinicznych etykiet patogenno\u015bci jako celu uczenia; sygna\u0142em treningowym jest to, kt\u00f3re warianty i jak cz\u0119sto wyst\u0119puj\u0105 w populacjach ludzkich i naczelnych. Warianty z ClinVar wy\u0142\u0105czone z treningu pos\u0142u\u017cy\u0142y do OCENY modelu (AUROC 0,94), a nie do uczenia. Konsekwencja praktyczna dla ka\u017cdego, kto model wykorzystuje: sygna\u0142 jest nastawiony na warianty germinalne, wi\u0119c status somatycznego drivera w guzie to inne pytanie i nie na nie model by\u0142 uczony."},"updated_at":"2026-09-07T15:15:28.912658","url":"/ai-oncology/models/alphamissense","usage":{"library":"predictions table (TSV) / repo code","notes_en":"Most users download the precomputed table rather than run the model; note the non-commercial licence on the predictions."},"verified_at":"2026-09-07T15:15:28.911174","verified_by":"Occe3C","version":null,"what_it_does":"Provides a pathogenicity score (0\u20131) and a likely benign / ambiguous / likely pathogenic call per variant; the full prediction table is downloadable and integrated into variant annotation pipelines."}