Sequential regulatory activity prediction across chromosomes with convolutional neural networks
Problem
Predicting which DNA sequences act as regulatory elements and how genetic variants influence gene expression in specific cell types, from sequence alone.
Why it matters
Such predictions help explain genomic function and support fine-mapping of disease loci.
What the AI did
A convolutional neural network system predicts cell-type-specific epigenetic and transcriptional profiles from DNA sequence, identifying promoters and distal regulatory elements to make gene expression predictions.
Results
Predictions of variant influence on expression align well with causal eQTL variants in human populations; quantitative metrics not stated in source.
Impact
The model's predictions support mechanistic hypotheses and fine-mapping of disease loci, per the authors.
Record
| Venue | Genome Research |
|---|---|
| Published | 2018 |
| Work type | preprint |
| Authors | David R. Kelley, Yakir Reshef, Maxwell L. Bileschi, David Belanger, Cory Y. McLean, Jasper Snoek |
| Citations | 534 (OpenAlex, at import time) |
| References | 71 |
| Model or method | convolutional neural networks |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1101/gr.227819.117 |
| Open access | green |
| Keywords | CNN, gene regulation, epigenetics, eQTL, disease loci |
Importance rating
significant — thousands to millions affected
Predicting cell-type-specific regulatory activity and variant effects from sequence aids understanding genomic function and fine-mapping disease loci; foundational capability; no scale stated.