Google DeepMind releases AlphaGenome Atlas mapping 9 billion DNA variants

Google DeepMind has released the AlphaGenome Atlas, a 1-petabyte database mapping the molecular effects of all 9 billion possible single-nucleotide variants in the human genome. By precomputing these effects, the Atlas removes the need for researchers to run computationally intensive models. The release introduces the AlphaGenome Variant Impact (AVI) score to rank mutations, which has already helped the GREGoR Consortium identify a rare epilepsy-linked variant in the DNM1 gene. The database is free for academic use, with commercial access coming soon to Google Cloud.

AlphaGenome Atlas release

• ▪Google DeepMind released the AlphaGenome Atlas, a comprehensive catalogue of precomputed predictions mapping the molecular effects of 9 billion single-nucleotide variants in the human genome
• ▪The AlphaGenome Atlas dataset is approximately 1 petabyte in size, which is more than 30 times larger than Google DeepMind's AlphaFold Database

Nine billion variant predictions

• ▪To compile the 9 billion predictions, Google DeepMind's genomics team accelerated calculation speeds by a factor of 80 using model distillation, GPU kernel optimization, and eliminating redundant calculations
• ▪Google DeepMind compiled the AlphaGenome Atlas by running its AlphaGenome AI model, originally introduced in 2025, across all 9 billion possible single-letter changes to a reference human genome
• ▪The AlphaGenome Atlas provides thousands of molecular effect predictions per variant, covering gene regulation across hundreds of human and mouse cell types and tissues

AVI scoring system

• ▪The AlphaGenome Variant Impact (AVI) score decomposes each score into additive feature attributions, showing researchers which specific biological processes, such as RNA splicing or chromatin accessibility, are predicted to be disrupted
• ▪In a collaboration with the GREGoR Consortium, researchers used the AVI score to identify a previously overlooked variant in the DNM1 gene linked to epileptic encephalopathy, which was experimentally validated to create an incorrect splice site
• ▪The AlphaGenome Variant Impact (AVI) score is a single number that ranks genetic variants by combining AlphaGenome's regulatory predictions with AlphaMissense, Google DeepMind's model for protein-altering variants

Noncoding DNA regulation

• ▪The AlphaGenome Atlas maps variants across both the 2% of the genome that codes for proteins and the remaining 98% of the genome that is noncoding
• ▪Gareth Hawkes of the University of Exeter applied the AlphaGenome Atlas to whole-genome data from over 54,000 UK Biobank participants, uncovering 22% more noncoding associations than conventional filters

Academic research portal access

• ▪The AlphaGenome Atlas is freely available for academic and non-commercial research through a web portal, the AlphaGenome API, and as a skill in Google Antigravity
• ▪Google DeepMind states that the AlphaGenome Atlas and the AVI score are research tools intended to prioritize laboratory experiments rather than clinical diagnostic tools

Debatable claims

• ▪DeepMind's AVI score oversimplifies the complex biological effects of genetic variants
• ▪Clinicians should use AlphaGenome Atlas predictions to guide patient diagnoses
• ▪Google should make the AlphaGenome Atlas entirely free for commercial use

8 sources