Žiga Avsec, Natasha S. Latysheva, Jun Cheng, Guido Novati, Kyle R. Taylor, Tom Ward, Clare Bycroft, Lauren Nicolaisen et al.
AlphaGenome is a unified model that takes 1 Mb DNA sequences as input and predicts diverse genomic functions (gene expression, transcription initiation, chromatin accessibility, histone modifications, transcription factor binding, chromatin contact maps, splicing, etc.) at single-base-pair resolution.
Existing deep learning models for genomic function prediction face a trade-off between input sequence length and prediction resolution, limiting their ability to capture long-range regulatory interactions or integrate diverse modalities.
AlphaGenome is a transformer-based model that takes 1 Mb DNA sequences as input and predicts thousands of functional genomic tracks at single-base-pair resolution. It is trained on human and mouse genomes and can simultaneously score variant effects across all modalities.
AlphaGenome matches or exceeds the strongest available external models in 25 of 26 evaluations of variant effect prediction. It accurately recapitulates the mechanisms of clinically relevant variants near the TAL1 oncogene and provides tools for making genome track and variant effect predictions from sequence.