DeepMind's AlphaGenome Atlas: Mapping 9 Billion DNA Changes for Disease Insights

Google DeepMind has launched the AlphaGenome Atlas, a public catalog that pre-computes the molecular effects of approximately 9 billion possible single-letter DNA changes across the human genome, pairing these predictions with a new single-number Variant Impact (AVI) score to accelerate disease insights. For broader context, explore our AI News. For broader context, explore our Top 100 AI Tools.

Understanding the AlphaGenome Atlas

The AlphaGenome Atlas is built upon the AlphaGenome model, which was initially introduced in 2025. This advanced AI model provides a comprehensive map of potential genetic impacts, offering insights into how individual nucleotide variants might influence molecular processes across various cell types and tissues. The dataset is significantly larger than previous efforts, exceeding the size of the AlphaFold protein database by more than 30 times.

A key focus of the AlphaGenome Atlas is the noncoding regions of the human genome. While about 98 percent of the human genome does not contain instructions for building proteins, a substantial portion of disease-linked variants are found in these areas. Interpreting the effects of these noncoding variants has historically been challenging, making the Atlas a potentially valuable resource for researchers.

The AlphaGenome Variant Impact (AVI) Score

Central to the AlphaGenome Atlas is the new AlphaGenome Variant Impact (AVI) score. This single-number score is designed to quantify the potential impact of a genetic variant. The AVI score utilizes 18 input features, combining predictions from AlphaGenome and AlphaMissense models with evolutionary conservation measures. This contrasts with the established CADD tool, a benchmark in genetic variant analysis, which uses more than 150 input features.

In retrospective tests, the AVI score demonstrated improved performance in identifying causal variants in rare-disease cases. The AVI score ranked the causal variant among its top 50 candidates in 29.5 percent of solved rare-disease cases, compared to 12.5 percent for the CADD tool. This suggests that the AVI score can offer a more focused and efficient approach to pinpointing disease-causing mutations.

Scope and Data Volume

The AlphaGenome Atlas covers approximately 9 billion single-nucleotide variants. For each variant, the atlas provides an average of about 27,000 individual prediction values. This vast amount of data allows for detailed analysis of how specific DNA changes might affect various biological functions.

Beyond individual variant predictions, the DeepMind team also identified 2,601 recurring short DNA patterns within the genome. These patterns are described as regulatory

Sources

• AlphaGenome Atlas: Molecular predictions for 9 Billion human DNA variants — Google DeepMind
• AlphaGenome — Google DeepMind
• alphagenome/README.md at main · google-deepmind/alphagenome · GitHub
• Google’s Atlas of the human genome could pave the way for new treatments | The Verge
• Client Challenge