DeepMind's AlphaGenome Atlas Maps Effects of 9 Billion Possible DNA Variants

Google DeepMind has introduced AlphaGenome Atlas, a free online platform predicting the molecular effects of more than 9 billion single-nucleotide variants — every possible single-letter change in human DNA. DeepMind calls it the most comprehensive catalogue yet of how genetic mutations affect molecular biology, built for academic research.

The Atlas precomputes predictions from AlphaGenome, DeepMind's existing AI model for forecasting how genetic variants affect biology. A single-nucleotide variant is a change to just one DNA letter, and there are roughly 9 billion possible ones in the human genome — too many to test individually in a lab. Precomputing them at genome-wide scale, DeepMind says, makes the model's insights far more accessible.

DeepMind is also releasing the AlphaGenome Variant Impact (AVI) score, which combines AlphaGenome with AlphaMissense — its model for scoring protein-altering DNA variants — into a single number ranking each variant's likely impact. AVI covers both the 2% of the genome that codes for proteins (coding DNA) and the remaining 98% (non-coding DNA), which regulates gene activity and holds most trait-linked variants. 'AVI feature attributions' show which molecular processes, such as RNA splicing or gene expression, a variant is predicted to disrupt most.

The Atlas also includes a catalogue of more than 2,500 recurring DNA sequence patterns, called motifs, and their locations across the genome. The full dataset is 1 petabyte, more than 30 times larger than DeepMind's AlphaFold Database, which grew from about 190,000 experimentally determined protein structures to more than 200 million predicted structures after its 2022 expansion. DeepMind says its own testing shows the AVI score performs best-in-class on several variant-pathogenicity and rare-disease benchmarks.

According to DeepMind, its trusted external collaborators have already used AlphaGenome Atlas to identify and experimentally verify key variants in unsolved rare-disease research, and to find rare variants associated with common traits. The Atlas is available through a website portal, the AlphaGenome API, and as a skill in Google Antigravity.