Enhancing missense variant pathogenicity prediction with protein language models using VariPred
Problem
Predicting whether rare and novel genetic missense variants are pathogenic is difficult, limiting clinical interpretation of sequencing results.
Why it matters
Accurate variant interpretation affects diagnosis and care for patients with rare genetic disease.
What the AI did
VariPred uses a pre-trained protein language model (ESM-1b) in an end-to-end classifier requiring only the protein sequence as input, with no structural features or multiple sequence alignments.
Results
Claims outperformance of state-of-the-art methods including 3Cnet, PolyPhen-2, REVEL, MetaLR, FATHMM; specific metrics not stated in abstract (truncated).
Impact
Provides an improved sequence-only tool for prioritizing genetic variants, per peer-reviewed benchmark comparisons.
Record
| Venue | Scientific Reports |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Weining Lin, Jude Wells, Zeyuan Wang, Christine A. Orengo, Andrew C.R. Martin |
| Citations | 34 (OpenAlex, at import time) |
| References | 45 |
| Model or method | ESM-1b protein language model (VariPred) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41598-024-51489-7 |
| Open access | gold |
| Keywords | protein language model, missense variant, pathogenicity, ESM-1b, genomics |
Importance rating
significant — thousands to millions affected
Sequence-only pathogenicity prediction aids clinical interpretation of rare/novel variants; genuinely new capability (no MSAs or structures) but abstract states no scale of affected patients.