Google DeepMind has released the AlphaGenome Atlas, a precomputed catalogue of molecular effect predictions for all 9 billion single-nucleotide variants in the human genome. The release introduces the AlphaGenome Variant Impact (AVI) score, which ranks variants by predicted impact across both coding and non-coding regions.
- The Atlas comprises a 1-petabyte dataset containing thousands of regulatory predictions per variant across hundreds of human and mouse cell types.
- The AVI score combines AlphaGenome's regulatory predictions with DeepMind's AlphaMissense model to provide a single impact number for all variants.
- Feature attributions decompose each AVI score into additive contributions from categories like chromatin accessibility, splicing, and conservation.
- A collection of over 2,500 recurrent DNA sequence motifs, including transcription factor binding sites, is included in the resource.
- Early results show collaborators identified a validated DNM1 splice variant and uncovered 22% more non-coding associations in UK Biobank data.
The Atlas provides a lookup table that removes bottlenecks for genome-scale studies by avoiding on-demand model inference. It is available as a free web portal and API for academic use, with commercial access coming soon via Google Cloud.