MVP predicts the pathogenicity of missense variants by deep learning
Problem
Accurately predicting whether missense genetic variants are pathogenic is critical for genetic studies and clinical diagnosis, and prior methods had limited performance.
Why it matters
Missense variant interpretation directly affects clinical genetic diagnosis and research into disease causes.
What the AI did
MVP uses a deep residual network trained on large datasets and many correlated predictors, with separate training for loss-of-function intolerant versus tolerant genes, benchmarked on cancer mutation hotspots and developmental-disorder de novo variants.
Results
Outperformed previous methods in prioritizing pathogenic missense variants, especially in LoF-tolerant genes; estimated de novo coding variant contribution of 7.8 (unit truncated in abstract).
Impact
Peer-reviewed benchmarking shows improved prioritization over prior methods, supporting MVP as a better prediction tool.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Hongjian Qi, Haicang Zhang, Yige Zhao, Chen Chen, John J. Long, Wendy K. Chung, Yongtao Guan, Yufeng Shen |
| Citations | 198 (OpenAlex, at import time) |
| References | 71 |
| Model or method | MVP (deep residual network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-020-20847-0 |
| Open access | gold |
| Keywords | deep learning, missense variants, pathogenicity prediction, MVP, de novo variants |
Importance rating
significant — thousands to millions affected
Abstract states variant interpretation is critically important for genetic studies and clinical diagnosis; MVP improved prioritization and nearly doubled estimated de novo contribution to congenital heart disease.