Google DeepMind launches AlphaGenome Atlas to map predicted effects of 9 billion DNA variants

Google DeepMind has launched AlphaGenome Atlas, a large AI-generated resource that predicts the molecular effects of 9 billion single-nucleotide variants across the human genome. Announced on September 8, 2026, the platform is designed for academic researchers and is available through a free website portal, the AlphaGenome API and Google Antigravity.

The release addresses one of the central bottlenecks in modern genomics: scientists can sequence DNA at enormous scale, but interpreting what each genetic change does inside a cell remains much harder. The human genome contains about 3 billion base pairs, and a single-letter substitution can sometimes disrupt gene regulation, protein production or disease-related biological processes. Testing every possible variant in a laboratory is not practical, so researchers often need computational systems that can help rank which changes deserve closer study.

AlphaGenome Atlas builds on Google DeepMind's AlphaGenome model, which predicts how genetic variants may affect biological processes. Instead of requiring researchers to run the model only on specific variants, DeepMind has precomputed predictions across the genome and organized them into an atlas. The company says the dataset is about 1 petabyte, more than 30 times larger than the AlphaFold Database, and is intended to give researchers a broad map of molecular effects rather than isolated predictions.

The atlas includes several connected resources. For each variant, it can provide thousands of molecular effect predictions across gene regulation signals, cell types and tissues. It also introduces the AlphaGenome Variant Impact score, or AVI, which combines AlphaGenome with AlphaMissense, DeepMind's model for protein-altering variants. The goal is to give researchers a single score that helps prioritize variants while still letting them inspect which molecular features appear to drive the prediction.

That distinction matters because much of the genome does not directly code for proteins. DeepMind notes that the roughly 2 percent of the genome that codes for proteins is better understood than the remaining non-coding regions, even though those regions play an important role in controlling gene activity and are linked to many traits. By applying the AVI score to both coding and non-coding variants, AlphaGenome Atlas aims to help researchers investigate changes that might otherwise be difficult to interpret.

Google DeepMind says external collaborators have already used the atlas to identify and experimentally verify variants in unsolved rare disease research and to find rare variants associated with common traits. Those claims do not mean the system replaces laboratory work or clinical judgment. Instead, the platform is meant to narrow the search space, helping scientists decide which variants to test, how to interpret possible mechanisms and where to look for disease-relevant signals.

The broader importance of the release is that AI biology tools are moving from individual predictions toward navigable scientific infrastructure. AlphaFold changed protein structure research by making a large database broadly accessible. AlphaGenome Atlas attempts something similar for variant interpretation, giving biologists a way to explore a genome-scale set of AI predictions without needing to run every computation themselves. If the platform proves reliable across more real-world studies, it could accelerate rare disease research, trait discovery and the long effort to connect DNA changes with biological function.