Google DeepMind released the AlphaGenome Atlas on September 8, 2026, launching a petabyte-scale computational resource containing molecular predictions for all 9 billion possible single-letter DNA variants in the human genome alongside a companion technical paper. According to the company’s announcements and published research, the catalogue is roughly 30 times larger than the AlphaFold Database expanded in 2022 and provides researchers with precomputed sequence-to-function predictions across hundreds of cell types and tissues.
What the AlphaGenome Atlas Contains
The platform maps genetic mutations by performing in silico saturation mutagenesis across the human genome. According to Google DeepMind, the resource generates predictions for every single-nucleotide variant in the hg38 reference assembly, plus over 100 million insertions and deletions drawn from population datasets including gnomAD, the UK Biobank, and All of Us. Each variant carries an average of 27,000 experiment-specific scalar predictions covering hundreds of human and mouse cell types and tissues.
Alongside raw predictions, the platform introduces the AlphaGenome Variant Impact (AVI) score. According to the technical documentation, the AVI score condenses AlphaGenome and AlphaMissense predictions, conservation metrics, and protein loss-of-function features into a single metric per variant. Each score is paired with feature attributions that break down the underlying molecular drivers, such as splicing alterations. The release also includes a compendium of 2,601 recurrent DNA sequence motifs mapped across various cell types to help researchers analyze regulatory genomic elements.
Early Research Applications in Genomics
External research teams have already applied the resource to clinical genetics and population studies. Working through the GREGoR Consortium, Laura Covill and Anne O’Donnell-Luria of the Broad Institute utilized the AVI score to investigate unsolved rare disease cases, prioritizing a deep intronic variant in the DNM1 gene linked to epileptic encephalopathy. The underlying AlphaGenome predictions indicated the variant created a brain-specific cryptic splice site that extended the resulting protein by 13 amino acids, a finding validated by experimental screens.
At the University of Exeter, Medical Research Council fellow Gareth Hawkes applied the atlas to whole-genome data from more than 54,000 UK Biobank participants. Grouping rare variants by predicted molecular effects allowed researchers to uncover 22 percent more non-coding genetic associations with circulating protein levels than would otherwise have been detectable, pinpointing regulatory variants impacting proteins such as PLA2G7 and EGLN1. Meanwhile, at the Stowers Institute for Medical Research, Julia Zeitlinger and Melanie Weilert used the motif resource to distinguish transcription factors that alter DNA accessibility from those that switch genes on and off.
Availability and Access Terms
The AlphaGenome Atlas website is available for non-commercial research use, accompanied by the AlphaGenome API and a dedicated skill in Google Antigravity, the company’s agentic development platform. According to Google DeepMind, commercial access via Google Cloud is slated for release soon.
Related reading