Deep structured learning for variant prioritization in Mendelian diseases
Problem
Identifying the disease-causing genetic variant among many candidates in Mendelian diseases is slow and difficult.
Why it matters
Faster variant prioritization speeds clinical diagnosis and research into monogenic diseases.
What the AI did
MAVERICK, an ensemble of transformer-based neural networks, classifies protein-altering SNVs and indels and ranks likely pathogenic variants under dominant or recessive inheritance.
Results
Ranks causative variant in top five in over 95% of 644 previously solved cases; 76% solved by top-ranked variant; outperforms all other major Mendelian pathogenicity programs.
Impact
Benchmarked on 644 patients with comparison against existing programs, per the abstract.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Matt Christopher Danzi, Maike Franziska Dohrn, Sarah Fazal, Danique Beijer, Adriana P. Rebelo, Vívian Pedigone Cintra, Stephan L. Zuchner |
| Citations | 43 (OpenAlex, at import time) |
| References | 62 |
| Model or method | MAVERICK (transformer-based ensemble, Keras) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-023-39306-7 |
| Open access | gold |
| Keywords | Mendelian diseases, variant prioritization, transformer, MAVERICK, pathogenicity prediction |
Importance rating
significant — thousands to millions affected
Automating causal variant identification for Mendelian diseases aids clinical diagnosis and research; abstract states it outperforms all major programs, ranks causative variant top-5 in >95% of 644 cases, and identified a novel disease gene.