AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome

Details

Google DeepMind released AlphaGenome Atlas, a pre-computed catalogue of predictions for every possible single-letter DNA change in the human genome — all 9 billion single-nucleotide variants. The dataset is 1 petabyte, more than 30 times the size of the AlphaFold Database at launch. It is freely available for academic research via a web portal, an API, and as a skill in Google Antigravity. The Atlas builds on AlphaGenome, DeepMind's AI model for predicting how genetic variants affect molecular biology. Rather than requiring researchers to run queries on individual variants, the Atlas pre-computes all predictions across the entire genome — turning a query-by-query research tool into a comprehensive indexed reference. The analogy to AlphaFold is intentional: just as AlphaFold unlocked structural biology by making protein structure predictions universally accessible, AlphaGenome Atlas aims to do the same for variant effect prediction. A new composite metric, the AlphaGenome Variant Impact (AVI) score, merges predictions from both AlphaGenome (molecular effects) and AlphaMissense (protein-altering variant impact) into a single ranked number per variant. Researchers can now sort all 9 billion SNVs by predicted importance and inspect molecular mechanism simultaneously — a significant improvement over running two separate queries. DeepMind reports that external collaborators have already used the Atlas to identify and experimentally verify key variants in unsolved rare disease cases and find rare variants linked to common traits. The scope — covering every possible mutation, not just observed ones — makes this qualitatively different from clinical databases like ClinVar, which catalog only variants that have appeared in research or clinical contexts.