AlphaGenome Atlas: AI maps every possible single-letter DNA change

Google DeepMind released the AlphaGenome Atlas, a precomputed database that predicts the molecular consequences of all 9 billion possible single-letter DNA changes across the human genome. The Atlas is available free for academic researchers and provides variant-impact predictions across hundreds of cell types, with a summary AVI score that ranks variants by potential impact. For AI builders working in genomics, drug discovery, or biobank-scale analysis, this is a structured, open-access dataset that can accelerate hypothesis generation and reduce the search space for experimental validation.

What the AlphaGenome Atlas actually does

DeepMind built the Atlas by running the AlphaGenome model across a reference human genome and comparing each DNA base to the three possible alternatives. Each variant is linked to an average of about 27,000 individual predictions about how the mutation affects gene expression, splicing, and protein manufacture. The predictions span hundreds of human and mouse cell types and tissues.

The Atlas also includes a summary metric called the AlphaGenome Variant Impact (AVI) score, which combines regulatory predictions from AlphaGenome with protein-change predictions from the earlier AlphaMissense model. An AVI score of 10 places a variant among the 10% most impactful in the genome; a score of 30 places it among the strongest one in a thousand. Each score is broken down by mechanism (splicing, gene expression, protein change) to show what drives the impact. The Atlas also catalogs more than 2,500 recurring DNA motifs that transcription factors bind to.

Why this matters for AI builders

The Atlas demonstrates a pattern that matters for anyone building AI-driven drug discovery pipelines or variant interpretation tools: large-scale models can generate structured, multi-faceted scores that decompose complex biological effects into actionable signals. The open-access model (with commercial licensing via Google Cloud) sets a precedent for how such resources can be shared. Isomorphic Labs, DeepMind's sister company focused on drug discovery, will have access under licensing terms, which signals how this data might feed into proprietary pipelines.

For builders integrating genomic data into clinical or research workflows, the Atlas provides a precomputed feature set that can be queried without running expensive models one variant at a time. The AVI score, in particular, offers a single ranking that combines regulatory and protein-level effects, which could simplify prioritization in automated analysis pipelines.

Practical impact on research

Early beta testing shows concrete results. In a patient with epileptic encephalopathy, the AVI score pointed to a variant in the DNM1 gene where 69% of the impact came from splicing effects that were missed by standard blood RNA sequencing. Laboratory experiments confirmed the prediction, and the variant was reclassified as likely pathogenic. In a retrospective test on previously solved cases, AVI placed the known causal variant among a patient's top 50 candidates 29.5% of the time, compared to 12.5% for the existing CADD ranking method.

In a UK Biobank analysis of more than 54,000 participants, filtering candidates by predicted molecular effect yielded 22% more associations than the same analysis run without Atlas, and in one case narrowed a region from 526 candidates to four. The motif maps also enabled researchers to sort transcription factors by function across cell types, work that would have been impractical experimentally.

What the Atlas doesn't do (yet)

DeepMind is clear that the Atlas predictions are not a substitute for experimental evidence. AlphaGenome works well for variants affecting splicing or promoters but can miss others, particularly in enhancers. The overall accuracy is not on par with what AlphaFold achieved for protein structure prediction. The paper notes training-data gaps and a limited ability to capture effects that act indirectly through changes in regulatory proteins.

Commercial licensing terms are still evolving, and the Atlas is described as a research tool that can form only part of the evidence chain behind a clinical diagnosis. For builders, the Atlas is a useful resource to shrink the search space, but it requires careful interpretation and experimental follow-up. It's a tool to find the haystack, not a replacement for finding the needle.