High-throughput deep learning variant effect prediction with Sequence UNET
Problem
The vast space of coding mutations cannot be experimentally characterised, and existing variant effect predictors are computationally intensive and difficult to scale.
Why it matters
Understanding coding mutations is important for many applications in biology and medicine.
What the AI did
Introduced Sequence UNET, a fully convolutional compression/expansion deep learning architecture using multi-scale sequence representations to classify variants and predict variant frequency at high throughput.
Results
Comparable pathogenicity prediction to recent methods; scalability shown by analysing 8.3B variants across 904,134 proteins; runs on modest hardware.
Impact
Enables high-throughput variant analysis on modest hardware via a simple Python package.
Record
| Venue | Genome biology |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Alistair S. Dunham, Pedro Beltrão, Mohammed AlQuraishi |
| Citations | 30 (OpenAlex, at import time) |
| References | 55 |
| Model or method | Sequence UNET |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1186/s13059-023-02948-3 |
| Open access | gold |
| Keywords | deep learning, variant effect prediction, Sequence UNET, genomics, scalability |
Importance rating
significant — thousands to millions affected
Variant interpretation matters broadly for biology and medicine; abstract demonstrates scale via 8.3B variants across 904,134 proteins, a genuinely new high-throughput capability.