Google DeepMind’s New Tool Maps All 9 Bn DNA Mutations in Humans

Google DeepMind has launched AlphaGenome Atlas, a platform that predicts the molecular effects of 9 billion possible single-letter changes in the human genome, giving researchers a way to rank and investigate genetic variants without testing each mutation individually in the laboratory.

The platform is available for academic research through a free website portal and includes precomputed predictions generated using DeepMind’s AlphaGenome AI model. The company said the resource is designed to provide researchers with a genome-wide view of how DNA variants may affect biological processes.

“DNA is the language of life,” Google DeepMind stated in a blog, asserting that interpreting how genetic variations affect biology at a molecular level remains a major challenge.

The dataset is about 1 petabyte in size, which DeepMind said is more than 30 times larger than the AlphaFold Database. AlphaGenome Atlas provides thousands of molecular effect predictions for each variant across multiple aspects of gene regulation, covering hundreds of human and mouse cell types and tissues.

The platform also introduces the AlphaGenome Variant Impact (AVI) score, which combines predictions from AlphaGenome with AlphaMissense, DeepMind’s model for predicting the effects of protein-altering DNA variants. The score assigns a single measure to each variant, allowing researchers to rank variants by potential impact.

Researchers can also access AVI feature attributions that show which biological processes contribute to a variant’s predicted impact, including RNA splicing and gene expression. The platform includes a collection of more than 2,500 recurring DNA sequence motifs and their locations.

DeepMind said the AVI score is designed to work across both coding and non-coding regions of the genome. Coding regions account for about 2% of the genome, while the remaining 98% is non-coding and contains regulatory elements that control gene activity.

The company said external collaborators have already used AlphaGenome Atlas to identify and experimentally validate variants relevant to rare diseases and to find rare variants associated with common traits.

AlphaGenome Atlas has also been used to study non-coding variants in population-scale genetic data. 

The platform is available through its website, the AlphaGenome API, and as a skill in Google Antigravity. DeepMind said Atlas is currently available for non-commercial use, while commercial access through Google Cloud is planned.

The company described AlphaGenome Atlas as a baseline rather than an endpoint, saying that improvements to AlphaGenome could make the genome-wide predictions more comprehensive and precise over time.

“Together, these tools will enable researchers and industry partners to accelerate the pace of biological discovery,” Google DeepMind said.