Google DeepMind maps 9 billion genomic variations with AlphaGenome Atlas
The predictive map covers non-coding regions of human DNA to help researchers identify cancer drivers and rare diseases.
Original cover art, generated for this story. THE VISSION does not republish third-party press imagery.
- Google DeepMind has introduced AlphaGenome Atlas, a high-resolution map of human DNA designed to predict the molecular effects of mutations across the entire genome.
- The Atlas maps and predicts the effects of all 9 billion single-letter DNA changes, offering critical insights into the 98% of non-coding regions once considered genetic 'dark matter.'
- To help researchers navigate the 1-petabyte dataset, DeepMind introduced the AlphaGenome Variant Impact (AVI) score to prioritize significant variants.
Google DeepMind has released the AlphaGenome Atlas, a comprehensive, high-resolution predictive map of the human genome. The open platform is designed to model and predict the molecular effects of all 9 billion possible single-letter DNA variations (single nucleotide variants) across the entire human genetic sequence. While traditional genomics has heavily focused on the 2% of the genome that codes for proteins, AlphaGenome Atlas provides comprehensive insights into the remaining 98% of the genome—the non-coding regions historically referred to as genomic 'dark matter'—revealing how variations there affect gene expression and protein production.
Built upon a lineage of Google DeepMind breakthroughs including AlphaFold, Enformer, and AlphaMissense, the AlphaGenome Atlas marks a massive leap in scaling AI for biology. The dataset generated by the models exceeds one petabyte in size. To make this vast repository navigable for medical researchers, Google DeepMind introduced the AlphaGenome Variant Impact (AVI) score. The AVI score is a unified metric that combines predictions from both coding and non-coding regions, allowing clinical scientists to quickly isolate and prioritize the most significant genetic variations for further study.
The platform has already demonstrated tangible medical utility during pre-launch trials. Researchers at the Broad Institute used the AVI score to identify previously undiagnosed rare disease cases, pinpointing critical pathogenic variants in the DNM1 gene. Google is also collaborating with international bodies, such as the Centre for Population Genomics in Australia, to ensure the dataset includes diverse genomic backgrounds to benefit under-represented populations. The entire AlphaGenome Atlas has been made freely available to the global scientific community to accelerate oncology research and personalized medicine.
By providing a predictive map for the 98% non-coding region of DNA, AlphaGenome Atlas solves a major bottleneck in genomics, shifting the field from slow, empirical mutation screening to proactive, AI-driven diagnostics. The open availability of this petabyte-scale dataset guarantees that global clinical labs can immediately apply these molecular predictions to oncology and rare disease discovery, cementing deep learning as the core engine of next-generation medicine.
Will global regulatory bodies or health insurance agencies formally integrate AVI scores into diagnostic standards for oncology and prenatal screenings?
Still open. When the paper finds out, it will say so here and on the open questions page — including if it got this wrong.