Accurate proteome-wide missense variant effect prediction with AlphaMissense
Problem
The vast majority of human missense genetic variants have unknown clinical significance, hindering genetic disease diagnosis.
Why it matters
Interpreting these variants is critical for diagnosing genetic disease and enabling precision medicine.
What the AI did
AlphaMissense, an AlphaFold adaptation fine-tuned on human and primate variant frequency databases, predicts missense variant pathogenicity from structural context and evolutionary conservation.
Results
State-of-the-art across a wide range of genetic and experimental benchmarks; 89% of missense variants classified as likely benign or likely pathogenic; predictions released for all possible human single amino acid substitutions.
Impact
Released a community database of pathogenicity predictions covering all possible human missense variants.
Record
| Venue | Science |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Jun Cheng, Guido Novati, Joshua Pan, Clare Bycroft, Akvilė Žemgulytė, Taylor Applebaum, Alexander Pritzel, Lai Hong Wong et al. |
| Citations | 2,095 (OpenAlex, at import time) |
| References | 105 |
| Model or method | AlphaMissense (adaptation of AlphaFold) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1126/science.adg7492 |
| Open access | bronze |
| Keywords | missense variants, pathogenicity prediction, AlphaFold, protein structure, clinical genetics |
Importance rating
significant — thousands to millions affected
Abstract states the vast majority of human missense variants are of unknown clinical significance, hindering genetic disease diagnosis; model classifies 89% of missense variants and releases predictions for all possible substitutions.