Google’s Map of Every Possible DNA Typo Could Speed Up Rare Disease Research
September 8, 2026 – 5:15 pm
Credit: Google
DeepMind has released AlphaGenome Atlas, a petabyte of precomputed predictions for all 9 billion possible single-letter changes in the human genome, free for academic use. Its impressive capabilities have been validated by UK Biobank whole genomes at Exeter and the Broad Institute. Commercial access is heading to Google Cloud.
DeepMind’s AlphaGenome Atlas
This dataset, more than 30 times the size of the AlphaFold database, includes:
- Predictions for every one of the 9 billion possible single-letter changes in the human genome.
- An AlphaGenome Variant Impact score combining AlphaGenome with AlphaMissense.
- A catalogue of over 2,500 recurrent DNA motifs.
Validations and Implications
Gareth Hawkes, a Medical Research Council fellow at Exeter, ran the Atlas against UK Biobank data and found 22% more non-coding associations. Narrowing to highly impactful variants, he identified 19 regions linked to body mass index.
At the Broad Institute, Laura Covill’s team used the score to discover a variant in the DNM1 gene causing an incorrect splice site, confirmed through experimental screens, which is linked to epileptic encephalopathy.
Licensing and Legal Considerations
Currently, the Atlas is free for non-commercial use. DeepMind plans to offer commercial access on Google Cloud, but no pricing has been announced yet. European law treats academic and commercial uses differently under the AI Act, with exemptions for research purposes.
As noted in a previous npj Digital Medicine paper, distinguishing between research and commercial goals can be challenging due to evolving project objectives and frequent overlaps.