Google DeepMind on Sept. 8, 2026, introduced AlphaGenome Atlas, a set of ready-made predictions of the molecular effects of about 9 billion possible single-letter changes to the human genome. The company says the resource is about 1 petabyte, roughly a million gigabytes, and more than 30 times the size of the AlphaFold Database by its comparison. Genetics labs can look up ranked predictions for these changes instead of computing each one from scratch.
The release includes an AlphaGenome Variant Impact (AVI) score, which combines signals from AlphaGenome and AlphaMissense into one ranking. It covers changes in both coding DNA, which carries instructions for proteins, and non-coding DNA. The release also comes with feature attributions and motif resources, which point to the DNA letters and patterns behind each prediction.
DeepMind says academic partners have already used the score to prioritize candidate DNA changes in rare-disease cases. One finding, in the DNM1 gene, involved splicing, the step where cells cut and rejoin a gene's message, and was later tested in lab experiments. Partners also used the score to find more links involving non-coding DNA in the UK Biobank, a large health study. DeepMind presents these as case studies from its collaborators.
Researchers can use the atlas through a website portal, through an API, a connection that lets developers plug a service into their own apps, and through a Google Antigravity skill. DeepMind frames the atlas as support for research, not clinical advice. In a short DeepMind video, collaborators explain the project's framing of the atlas as a "thesaurus for the language of life."