Google DeepMind has launched AlphaGenome Atlas, a comprehensive resource that predicts how every possible single-nucleotide variant in the human genome affects molecular biology. The Atlas contains precomputed predictions for all 9 billion possible single-letter genetic changes, forming a 1-petabyte dataset that is more than 30 times larger than the AlphaFold Database.
Google DeepMind launched AlphaGenome Atlas, a 1-petabyte resource predicting the molecular impact of all 9 billion possible single-nucleotide variants in the human genome, forming the centerpiece of the article's introduction.
At the core of the Atlas is the AlphaGenome Variant Impact (AVI) score, a single metric that combines predictions for both coding regions (which make up about 2% of the genome and code for proteins) and non-coding regions (the remaining 98% that regulate gene activity). This score allows researchers to rapidly rank variants by their predicted molecular impact without analyzing thousands of individual data points.
The Atlas is accessible through an intuitive website portal requiring no coding skills, as well as via an API and as a skill in Google Antigravity. It provides several interconnected resources: thousands of molecular effect predictions across hundreds of human and mouse cell types and tissues for each variant, AVI feature attributions that link scores to specific biological processes like RNA splicing or chromatin accessibility, and a compendium of over 2,500 recurrent DNA sequence motifs.
Early applications demonstrate the Atlas’s utility. At the Broad Institute, researchers used the AVI score to prioritize variants in unsolved rare disease research, identifying a critical variant in the DNM1 gene that predicted an incorrect splice site. In population genetics, Dr. Gareth Hawkes applied the Atlas to UK Biobank data from over 54,000 participants, uncovering 22% more non-coding genetic associations and identifying 19 genetic regions linked to body mass index (BMI) by focusing on the top 1% of impactful variants.
While the Atlas offers a genome-wide view of variant effects, discussion among experts highlights ongoing challenges in variant prediction, including the need for comparative data from other species and large-scale laboratory mutagenesis experiments to improve accuracy. Questions remain about how well the model captures complex regulatory elements like promoter sequences and joint probabilities of genetic features. The DeepMind team positions the Atlas as a tool to accelerate biological discovery by making vast genomic datasets usable for researchers without specialized computational expertise.