Google DeepMind releases AlphaGenome Atlas mapping nine billion DNA variants

Google DeepMind has released the AlphaGenome Atlas, a one-petabyte dataset that provides predictions for approximately nine billion single-letter changes to human DNA. The Atlas covers every possible variant, assigning each an AlphaGenome Variant Impact score that estimates the potential damage caused by a specific genetic alteration.

By precomputing these results, DeepMind has addressed a major bottleneck in genomics. Previously, running the AlphaGenome model was a slow and computationally intensive process for individual laboratories. The Atlas allows researchers to perform a simple lookup of precomputed data rather than running the intensive model themselves.

The release follows a strategic business model similar to the AlphaFold Database. While the Atlas website is free for non-commercial use and its API is available on GitHub for academic research, commercial access is provided through Google Cloud’s Model Garden. This approach allows Google to provide a significant scientific contribution to the research community while monetizing the infrastructure and hosting through its cloud services.