Genome-wide prediction of disease variant effects with a deep protein language model
Problem
Existing deep-learning methods could not predict the effects of all possible human coding variants, limiting genetic diagnosis and interpretation.
Why it matters
Predicting which DNA variants cause disease is essential for diagnosing rare genetic conditions and understanding disease mechanisms.
What the AI did
The ESM1b protein language model was used to predict effects of all ~450 million possible human missense variants, released on a public web portal.
Results
Outperformed existing methods classifying ~150,000 ClinVar/HGMD variants and across 28 deep mutational scan datasets; annotated ~2 million isoform-specific damaging variants.
Impact
Peer-reviewed Nature Genetics paper with 420 citations; predictions made publicly available on a web portal.
Record
| Venue | Nature Genetics |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Nadav Brandes, Grant Goldman, Charlotte H. Wang, Chun Ye, Vasilis Ntranos |
| Citations | 421 (OpenAlex, at import time) |
| References | 76 |
| Model or method | ESM1b (650-million-parameter protein language model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41588-023-01465-0 |
| Open access | hybrid |
| Keywords | protein language model, ESM1b, missense variants, ClinVar, variant prediction, genomics |
Importance rating
significant — thousands to millions affected
Abstract states prediction of all ~450 million possible human missense variant effects, outperforming existing methods on ~150,000 ClinVar/HGMD variants; a genuinely new capability for genetic diagnosis, publicly released.