FULL STORY

DeepMind Launches AlphaGenome Atlas

Google DeepMind released AlphaGenome Atlas, precomputing the effects of ~9 billion DNA variants, and opened the 1PB lookup resource free for researchers.

2026-09-08 ~ 2026-09-15 · 2 episodes · 49 posts

Episode 1 · DeepMind Launches AlphaGenome Atlas: Predicted Effects of All 9 Billion DNA Variants Now Searchable (2026-09-08, 47 posts)

On September 8, Google DeepMind released AlphaGenome Atlas: a 1PB searchable open database that precomputes the molecular effects of all 9 billion possible single-nucleotide variants in the human genome using the AlphaGenome model released last year. It turns the rare-disease challenge of finding disease-causing "typos" among billions of bases into a queryable map, and is widely regarded as another AI-for-Science milestone after AlphaFold. CEO Demis Hassabis framed it as applying AI to mapping the human genome after the protein atlas, and Sundar Pichai highlighted that it is accessible from any web browser with zero code.

Confirmed

  • The dataset is 1PB, more than 30 times larger than the AlphaFold database according to DeepMind
  • It covers every possible single-letter DNA change and is described as the most comprehensive catalog of variant-to-molecular-effect mappings to date; any variant's predicted impact can be queried with zero code
  • Predictions span effects on gene expression, RNA splicing, and proteins, including the 98% non-coding regulatory genome that has been hard to study
  • Per Eric Topol, the atlas is free for non-commercial use and can aid interpretation of disease-causing variants
  • At its core is the AlphaGenome Variant Impact (AVI) score, which integrates AlphaMissense and AlphaGenome to rank variants by likely harm and links variants to the molecular mechanisms they disrupt rather than returning black-box scores
  • The release includes an online genome browser portal, a blog post, and a preprint, with collaborations applying the resource
  • It builds on the AlphaFold Database and its free-for-researchers model
  • According to DeepMind team member pushmeet, early users have already used Atlas to accelerate rare-disease research and identify causal phenotypes

Why it matters

  • The significance of most possible genomic variants was previously unknown; Atlas shifts variant-effect lookup from one-at-a-time experiments to precomputed, instantly searchable predictions, potentially accelerating identification of pathogenic mutations and genetic-disease research
  • AVI scores not only rank harmfulness but also explain the molecular mechanism of damage, giving researchers interpretable evidence

27 more related posts →

Episode 2 · DeepMind Launches AlphaGenome Atlas, Predicting All 9 Billion Human DNA Variants (2026-09-15, 2 posts)

Google DeepMind launched AlphaGenome Atlas, a free 1PB atlas predicting the functional impact of all 9 billion possible single-base DNA variants in the human genome, a major resource for genetics and disease research.