Mountain View, California – September 8, 2026
Google DeepMind released the AlphaGenome Atlas today, a comprehensive database containing AI-powered predictions for all 9 billion possible single-letter mutations in the human genome. The freely available resource maps how each genetic variant affects gene regulation across hundreds of cell types, offering researchers a powerful new tool for understanding disease and human biology.
The Human Genome Project completed the first full DNA sequence of humanity in 2003, but understanding how genetic variations affect biological function has remained a formidable challenge. With approximately 9 billion possible single-nucleotide variants across the human genome, experimentally testing each mutation in a laboratory would require more time than exists in human history. AlphaGenome Atlas addresses this bottleneck by using artificial intelligence to precompute predictions at scale.
Building on Previous AI Breakthroughs
Google DeepMind previously introduced the AlphaGenome model in 2025, which could predict how individual DNA variants impact biological processes. The new Atlas builds on this foundation by generating predictions for every possible genetic change across the entire genome, creating what the company describes as "the most comprehensive catalogue of how genetic mutations affect molecular biology."
The Atlas represents a 1-petabyte dataset—more than 30 times larger than DeepMind's AlphaFold protein structure database. DeepMind's AI for science team achieved this by improving computational efficiency by a factor of 80 through model distillation, GPU kernel optimization, and elimination of redundant calculations.
Comprehensive Molecular Effect Predictions
For each of the 9 billion variants, the Atlas provides thousands of molecular effect predictions across multiple aspects of gene regulation, including RNA splicing, chromatin accessibility, and gene expression levels. The predictions span hundreds of human and mouse cell types and tissues, enabling researchers to understand how genetic variants might function differently across different biological contexts.
To simplify interpretation, DeepMind introduced the AlphaGenome Variant Impact (AVI) score—a single number combining predictions from AlphaGenome and AlphaMissense, DeepMind's model for predicting protein-altering variant effects. This score allows researchers to rapidly rank variants based on their predicted biological impact. An AVI score of 10 puts a variant among the 10% most impactful in the genome, while a score of 30 places it among the strongest one in a thousand.
Real-World Research Applications
Early research collaborations have already demonstrated the Atlas's utility across multiple domains. The GREGoR Consortium, working on unsolved rare diseases, applied the AVI score to prioritize genetic variants among thousands of candidates. Researchers from the Broad Institute, including Laura Covill and Anne O'Donnell-Luria, identified a previously overlooked variant in the DNM1 gene associated with epileptic encephalopathy.
AlphaGenome predictions showed how the variant created an incorrect splice site, leading to abnormal protein extension—a prediction confirmed by experimental validation. This breakthrough illustrates how the Atlas helps researchers pinpoint causal variants hidden among thousands of candidates, a major hurdle in understanding rare diseases.
At the University of Exeter, Medical Research Council fellow Gareth Hawkes applied AlphaGenome Atlas to whole-genome data from over 54,000 UK Biobank participants. By grouping rare variants based on their predicted molecular effects, Hawkes uncovered 22% more non-coding genetic associations than statistical methods alone could detect. The approach identified specific regulatory variants affecting critical proteins including PLA2G7 (linked to aging) and EGLN1 (a cellular oxygen sensor).
At the Stowers Institute for Medical Research, Julia Zeitlinger and Melanie Weilert used the Atlas to map which transcription factors regulate DNA accessibility versus gene activation, helping identify regulatory DNA sequences across different cell types.
Open Access for Scientific Discovery
DeepMind made AlphaGenome Atlas immediately available for non-commercial academic research through an intuitive website portal, requiring no coding expertise. The AlphaGenome base model is available via the AlphaGenome API on GitHub, and Atlas is also accessible as a skill within Google's Antigravity platform for automated scientific workflows.
Commercial access will be offered through Google Cloud licensing arrangements, while DeepMind's sister company Isomorphic Labs—which applies AI to drug discovery—will have access through separate commercial agreements.
The release marks a significant expansion of open science resources, following DeepMind's pattern of making large-scale biological datasets freely available. The AlphaFold Database, released in 2022, made more than 200 million protein structure predictions available, driving widespread adoption across the life sciences.
Looking Ahead
DeepMind positions AlphaGenome Atlas as a baseline rather than an endpoint. As the underlying AlphaGenome model improves, the Atlas predictions will become increasingly comprehensive and precise. The company envisions integrating Atlas into broader agentic systems that could support end-to-end scientific discovery workflows, from variant identification to experimental design.
The Atlas includes over 2,500 de novo DNA sequence motifs—recurring patterns that control gene regulation—providing researchers with insights into how genetic changes might affect biological function at the molecular level.
Source: Google DeepMind Blog, September 8, 2026. ProPakistani, September 9, 2026. IEEE Spectrum, September 8, 2026.
FIRAT Editorial Board
Institutional Research Desk · Foresight Institute of Research and Translation
The collective editorial and research translation board of FIRAT, synthesising peer-reviewed evidence, policy briefs, and division milestones across our seven foundational research pillars.
