3Cnet: pathogenicity prediction of human variants using multitask learning with evolutionary constraints
Problem
Interpreting human genetic variants for disease diagnosis is slow, costly, and existing tools suffer from overfitting due to limited clinical data.
Why it matters
Faster, more sensitive variant interpretation can reduce the cost of genome-based diagnosis for patients with genetic diseases.
What the AI did
3Cnet uses recurrent neural networks to analyze amino acid context of variants, trained on simulated variants reflecting evolutionary conservation plus clinical data.
Results
Finds disease-causing variants with 2.2 times greater sensitivity than currently available tools.
Impact
Improves discovery of pathogenic variants in patient genomes, per the stated sensitivity gain.
Record
| Venue | Bioinformatics |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Dhong‐gun Won, Dongwook Kim, Junwoo Woo, Kyoungyeul Lee |
| Citations | 39 (OpenAlex, at import time) |
| References | 51 |
| Model or method | 3Cnet (recurrent neural networks, multitask learning) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1093/bioinformatics/btab529 |
| Open access | hybrid |
| Keywords | pathogenicity prediction, recurrent neural networks, genetic variants, evolutionary conservation |
Importance rating
significant — thousands to millions affected
Genome-based diagnosis for genetic disease patients; 2.2x greater sensitivity in finding disease-causing variants improves diagnosis rates per abstract.