Basset: learning the regulatory code of the accessible genome with deep convolutional neural networks
Problem
The functional mechanisms of most disease-associated noncoding genetic variants are unknown, and the regulatory code of gene expression is incompletely understood.
Why it matters
Interpreting noncoding variants is essential to understanding genetic disease mechanisms and prioritizing causal mutations.
What the AI did
Basset, an open-source deep convolutional neural network, learns DNA sequence functional activity from DNase-seq data across 164 cell types and predicts variant effects on accessibility.
Results
Greater predictive accuracy than previous methods; variant-effect predictions far greater for likely-causal GWAS SNPs than linked nearby SNPs
Impact
Provides a trained tool that researchers can apply to their own cell type with a single sequencing assay.
Record
| Venue | Genome Research |
|---|---|
| Published | 2016 |
| Work type | preprint |
| Authors | David R. Kelley, Jasper Snoek, John L. Rinn |
| Citations | 1,047 (OpenAlex, at import time) |
| References | 64 |
| Model or method | Basset (deep convolutional neural network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1101/gr.200535.115 |
| Open access | bronze |
| Keywords | deep learning, CNN, DNase-seq, GWAS, regulatory variants, gene expression |
Importance rating
significant — thousands to millions affected
Many noncoding variants are statistically associated with human disease but have unknown mechanisms; tool annotates genome-wide variant effects, a significant but unquantified genetics advance.