Analysis of missense variants in the human genome reveals widespread gene-specific clustering and improves prediction of pathogenicity
Problem
Predicting whether missense gene variants are pathogenic is difficult, hindering diagnosis of hereditary disease and cancer.
Why it matters
Interpreting genetic variants is essential for diagnosing hereditary conditions and cancer risk.
What the AI did
A machine learning analysis revealed widespread within-gene clustering of pathogenic missense variants, and this was used to build MutScore, a random forest pathogenicity predictor.
Results
Non-random variant distribution detected in 387/840 (46%) genes for pathogenic and 172 (20%) for benign variants; MutScore identified pathogenic mutations with very high accuracy, outperforming existing tools.
Impact
Peer-reviewed predictor outperforming existing tools on 840 ClinVar genes.
Record
| Venue | The American Journal of Human Genetics |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Mathieu Quinodoz, Virginie G. Peter, Katarina Cisarova, Béryl Royer‐Bertrand, Peter D. Stenson, D.N. Cooper, Sheila Unger, Andrea Superti‐Furga et al. |
| Citations | 89 (OpenAlex, at import time) |
| References | 80 |
| Model or method | MutScore (random forest) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1016/j.ajhg.2022.01.006 |
| Open access | hybrid |
| Keywords | missense variants, pathogenicity prediction, random forest, ClinVar, MutScore |
Importance rating
significant — thousands to millions affected
Pathogenicity prediction underpins diagnosis of hereditary disease and cancer; abstract shows clustering across 840 ClinVar genes and a tool outperforming existing predictors, but states no population scale.