AlphaGenome

Summary

AlphaGenome takes up to one megabase of DNA sequence and predicts thousands of functional genomic tracks at resolutions down to a single base. It unifies gene expression, transcription, chromatin, splicing and three-dimensional contact predictions for regulatory variant interpretation.

Evaluation

Trained on human and mouse genomes, AlphaGenome matched or exceeded the strongest external models in 25 of 26 reported variant-effect evaluations. Its multi-modal outputs recapitulated mechanisms of clinically relevant variants near TAL1, illustrating how one model can connect sequence changes to several regulatory consequences.[1]

Application Case Studies

AlphaGenome's sequence-only foundation allows it to predict the effects of genetic variants that are "invisible" to standard databases because they lack deep evolutionary conservation or classic ENCODE overlap. For example, AlphaGenome was used to successfully predict that rs603424 disrupts a powerful, highly tissue-specific adipocyte enhancer regulating SCD, resolving a genetic association that standard proximity-based pipelines often misassign to PKD2L1.

Citations

[1] Avsec et al. (2026), "Advancing regulatory variant effect prediction with AlphaGenome"