MMSplice: modular modeling improves the predictions of genetic variant effects on splicing
Problem
Predicting how genetic variants affect splicing (exon skipping, splice site choice, pathogenicity) is difficult but highly relevant to human genetics.
Why it matters
Mis-predicting variant effects on splicing can miss disease-causing genetic variants in patients.
What the AI did
MMSplice combines neural network modules scoring exons, introns, and splice sites, trained on distinct large-scale genomics datasets, to predict variant effects directly from VCF files.
Results
Winning model of the CAGI5 exon skipping prediction challenge; matched or higher performance than state-of-the-art.
Impact
Provides a top-ranked, independently benchmarked tool for predicting splicing effects of variants, including indels.
Record
| Venue | Genome biology |
|---|---|
| Published | 2019 |
| Work type | article |
| Authors | Jun Cheng, Thi Yen Duong Nguyen, Kamil J. Cygan, Muhammed Hasan Çelik, William Guy Fairbrother, Žiga Avsec, Julien Gagneur |
| Citations | 234 (OpenAlex, at import time) |
| References | 67 |
| Model or method | MMSplice (modular neural networks) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1186/s13059-019-1653-z |
| Open access | gold |
| Keywords | splicing, genetic variants, neural networks, CAGI5, exon skipping, pathogenicity |
Importance rating
significant — thousands to millions affected
Variant effects on splicing are highly relevant to human genetics and missing disease-causing variants; abstract gives no quantified scale of affected patients.