Back

DeepMind releases AlphaGenome Atlas covering about 9 billion DNA letter changes

DeepMind AlphaGenome Atlas abstract press image

Google DeepMind published AlphaGenome Atlas, a roughly 1 petabyte catalogue of predicted effects for every single-nucleotide human DNA variant, plus an AVI impact score and a free research portal.

Google DeepMind on Sept. 8, 2026, introduced AlphaGenome Atlas: precomputed molecular-effect predictions for about 9 billion possible single-letter DNA changes in the human genome. The company says the resource is about 1 petabyte and more than 30 times the AlphaFold Database by its comparison.

The release includes an AlphaGenome Variant Impact score that combines AlphaGenome and AlphaMissense signals into one ranking for coding and non-coding variants, plus feature attributions and motif resources.

DeepMind says academic partners already used the score to prioritize rare-disease candidates, including a DNM1 splicing finding later screened experimentally, and to surface more UK Biobank non-coding associations. Those outcomes are company-reported collaborator case studies.

Access is via a website portal, an API and a Google Antigravity skill. The company frames the atlas as research support, not clinical advice.

A short official DeepMind video features collaborators explaining the project's "thesaurus for the language of life" framing.

Sources

Source moments

  1. 0:00-0:36 : genome-as-language framing; Dr. Gareth Hawkes on WGS / phenotypes
  2. 1:01-1:25 : what AlphaGenome predicts for a single mutation
  3. 1:30-1:59 : AVI score; precomputed for ~9B single-letter changes
  4. 2:12-2:31 : Atlas as ~1 PB resource + accessible website
  5. 2:56-3:29 : "thesaurus" framing; agentic AlphaFold+AlphaGenome aspiration