Google DeepMind Releases AlphaGenome Atlas: 1-Petabyte Precomputed Impact Predictions for All ~9 Billion Human DNA Variants

Google DeepMind launched AlphaGenome Atlas, a 1-petabyte database (30x the size of the AlphaFold Database) precomputing molecular-effect predictions for roughly 9 billion possible single-nucleotide variants in the human genome, with a unified AVI score that…

Google DeepMind released AlphaGenome Atlas, a 1-petabyte catalogue that precomputes functional-effect predictions for roughly 9 billion possible single-nucleotide variants across the ~3 billion base pairs of the human genome — about 27,000 prediction values per variant, and over 30 times the size of the AlphaFold Database. The Atlas introduces the AlphaGenome Variant Impact (AVI) score, a small neural network combining AlphaGenome regulatory predictions, AlphaMissense, and evolutionary conservation features (18 inputs versus CADD's 150+) into a single ranking covering both coding and non-coding variants. DeepMind reports best-in-class AVI performance: it ranked causal variants in the top 50 candidates for 29.5% of solved GREGoR rare-disease cases versus 12.5% for CADD. Early users at the Broad Institute, University of Exeter, and Stowers Institute demonstrated rare-disease variant reprioritization — including a previously unclear DNM1 splice variant linked to epileptic encephalopathy that lab experiments confirmed as likely disease-causing — and a UK Biobank analysis of 54,000+ genomes surfaced 22% more non-coding genetic associations, including 19 regions linked to BMI. The dataset also includes per-variant feature attributions and over 2,500 recurrent DNA sequence motifs. Access is available via a free academic (non-commercial) no-code web portal, an API, and a Google Antigravity skill, with a commercial version via Google Cloud coming soon; the tool is not clinically approved.

• AlphaGenome Atlas precomputes molecular-effect predictions for ~9 billion possible single-nucleotide variants in the human genome (~3 billion base pairs)
• Dataset size is 1 petabyte — more than 30x larger than the AlphaFold Database — with ~27,000 prediction values per variant
• AVI (AlphaGenome Variant Impact) score merges AlphaGenome regulatory predictions, AlphaMissense, and evolutionary conservation features into one ranking of coding and non-coding variants, using 18 input features versus CADD's 150+
• AVI ranked causal variants in the top 50 for 29.5% of solved GREGoR rare-disease cases versus 12.5% for CADD
• Broad Institute collaborators validated a DNM1 splice variant linked to epileptic encephalopathy, reclassified as likely disease-causing after lab confirmation
• UK Biobank analysis of 54,000+ genomes found 22% more non-coding genetic associations, including 19 regions linked to BMI
• Collaborating institutions include the Broad Institute, University of Exeter, and Stowers Institute
• Dataset includes per-variant feature attributions and over 2,500 recurrent DNA sequence motifs