DeepMind releases AlphaGenome Atlas with predictions for 9 billion DNA variants

Google DeepMind on Tuesday released AlphaGenome Atlas, a database of precomputed predictions for the effects of nine billion single variants in the human genome. The atlas is built on AlphaGenome, an artificial intelligence model DeepMind released last year that analyzes stretches of noncoding DNA, the part of the genome that primarily regulates gene activity rather than encoding proteins. The model predicts how much a given variant changes molecular function. With the atlas, that analysis has been run in advance across the full set of possible single-letter changes, so researchers can look up a prediction rather than compute it themselves. Scientific American reports that this could streamline genetics research for many diseases, since much of the difficulty in linking DNA differences to disease lies in the regulatory regions AlphaGenome models. The audience is primarily genetics researchers and clinicians studying the variants behind inherited conditions. The atlas provides predictions, not experimental confirmation, so its value depends on how well those predictions hold up against laboratory results. DeepMind's own announcement describes the release as a predictive map of every possible DNA letter change in the human genome.