Prioritizing non-coding regions based on human genomic constraint and sequence context with deep learning
Problem
Identifying which non-coding regions of the human genome are functional and disease-relevant is a key challenge in genomics.
Why it matters
Prioritizing disease-relevant non-coding variants can improve diagnosis of Mendelian and other diseases.
What the AI did
JARVIS, a deep learning model integrating intolerance-to-variation (gwRVIS from 62,784 individuals), functional annotations and genomic sequence, prioritizes non-coding regions.
Results
Outperformed other human lineage-specific scores; performed comparably to or better than conservation-based scores for classifying pathogenic single-nucleotide and structural variants.
Impact
Provides a score that outperforms existing methods for prioritizing disease-relevant non-coding variants.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Dimitrios Vitsios, Ryan S. Dhindsa, Lawrence Middleton, Ayal B. Gussow, Slavé Petrovski |
| Citations | 88 (OpenAlex, at import time) |
| References | 50 |
| Model or method | JARVIS (deep learning model) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-021-21790-4 |
| Open access | gold |
| Keywords | deep learning, non-coding regions, genomics, JARVIS, variant prioritization, gwRVIS |
Importance rating
significant — thousands to millions affected
Non-coding functional annotation is a key human genomics challenge affecting disease-variant prioritization; JARVIS outperforms existing scores using 62,784 genomes, but no disease-population scale is stated.