Google DeepMind opens AlphaGenome Atlas for genetics research

Google DeepMind has released AlphaGenome Atlas, a freely available database designed to make its AlphaGenome model easier for geneticists to use. Announced on Sept. 8, the resource contains predictions for 9 billion possible changes to the human genetic code, estimating how each variant could affect tissues and cellular processes. It also provides an AlphaGenome Variant Impact score to help researchers prioritize genetic changes for further study.

AlphaGenome predicts how single-letter DNA changes may alter proteins, cells and biological outcomes, with coverage extending up to 1 million pairs of surrounding code. The Atlas packages those results in a browser-based portal and removes the need for many users to run demanding computations themselves. DeepMind computed the variant effects in advance and made the dataset, totaling 1 petabyte, or 1 million gigabytes, freely available.

Experts described the tool as useful but limited. Tuuli Lappalainen said it is not a holy grail and warned that predictions should be treated with caution. A Sept. 11 preprint led by Katie Pollard found AlphaGenome could identify causal mutations but often underestimated their impact, especially when regulatory elements and controlled genes were far apart. Researchers said wet-lab experiments remain essential because biology is still data-limited.