AlphaMissense
Proteome-wide missense variant pathogenicity scores, adapted from AlphaFold and fine-tuned on population variation.
About
AlphaMissense (Cheng et al., Science 2023) predicts whether single amino acid substitutions in human proteins are likely benign or pathogenic. It adapts the AlphaFold architecture and fine-tunes it on human and primate population variant frequencies, treating variants common in healthy populations as benign and rare ones as putatively pathogenic, combined with protein language modeling and structural context. Each possible substitution gets a pathogenicity score between 0 and 1.
DeepMind ran it across the human proteome and released predictions for all 71 million possible single amino acid substitutions: 89 percent were classified as likely benign (57 percent) or likely pathogenic (32 percent) at thresholds set for about 90 percent precision on ClinVar variants. Genomics England has tested it with the NHS. The prediction database is openly downloadable, but the model weights were never released, and DeepMind cautions the predictions are not for clinical use on their own.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported (Cheng et al., Science 2023): state-of-the-art across genetic and experimental benchmarks without training on them; 89% of 71M substitutions classified at ~90% precision on ClinVar thresholds; gene-level scores predictive of cell essentiality
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Science 2023 paper (DOI 10.1126/science.adg7492) and DeepMind's research page confirm the method, the 71M-variant resource, and the classification statistics; Genomics England testing is documented by the BBC. Strengths: massive open resource, state-of-the-art variant-effect prediction, real clinical-research uptake. Docked for: (1) model weights never released, so the model itself cannot be run or audited; (2) missense-only and human-only coverage; (3) predictions are not clinical-grade and DeepMind cautions against clinical use. License: predictions database CC BY 4.0 (non-commercial restriction lifted March 2024); model proprietary.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 6.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · 71M-variant open resource; state-of-the-art variant-effect prediction (Science 2023); tested by Genomics England.
- · Model weights never released, so the model cannot be run or audited.
- · Missense-only, human-only; not clinical-grade per DeepMind's own caution.