A new deep learning model called AlphaGenome, developed by researchers at Google DeepMind, analyzes up to one million base pairs of DNA sequence to forecast thousands of functional genomic features at single-base resolution. These predictions span 11 modalities, from gene expression and RNA splicing to chromatin accessibility, histone modifications, transcription factor binding, and spatial chromatin contacts, drawing on extensive human and mouse datasets. In benchmark tests, the model equals or surpasses leading alternatives on 25 out of 26 variant effect prediction tasks, including those for expression quantitative trait loci (eQTLs), splicing changes, and chromatin interactions. AlphaGenome also reconstructs the regulatory mechanisms behind clinically significant variants near the TAL1 oncogene.

ARTICLE

Avsec Ž, Latysheva N, Cheng J, Novati G, Taylor KR, Ward T, et al. Advancing regulatory variant effect prediction with AlphaGenome. Nature. 2026;649(8099):1206-1218. https://doi.org/10.1038/s41586-025-10014-0.

AVAILABILITY

AlphaGenome's primary website is hosted by Google DeepMind at https://deepmind.google.com/science/alphagenome

It is also on GitHub. There are two key repositories:

https://github.com/google-deepmind/alphagenome – This provides the Python SDK and programmatic access to the hosted API.
https://github.com/google-deepmind/alphagenome_research – This contains the model source code, weights, variant scoring tools, and related research materials, as stated in the Nature paper.

Additional resources include detailed documentation at https://www.alphagenomedocs.com and a community forum.