Google DeepMind has released a free catalog of predicted molecular effects for every possible single-letter substitution in human DNA. Researchers can use AlphaGenome Atlas to prioritize variants for laboratory testing, including those that earlier searches for the causes of rare diseases missed.
Released on September 8, the Atlas contains predictions calculated in advance using the existing AlphaGenome model. An accompanying Variant Impact (AVI) score combines its predictions with those of AlphaMissense, which assesses DNA changes that alter proteins. The score ranks variants by their predicted impact and shows which molecular processes they may disrupt.
Each variant comes with thousands of predicted molecular effects across hundreds of cell types and tissues. The Atlas covers the non-coding 98% of the genome, which contains most variants associated with human traits, as well as the 2% that encodes proteins.
In research with the GREGoR Consortium, a Broad Institute team identified a previously overlooked variant affecting DNM1, a gene linked to epileptic encephalopathy. AlphaGenome predicted that the variant created an incorrect splice site, disrupting how the cell processes genetic instructions and producing an abnormally extended protein. Laboratory tests confirmed the predicted effect and identified nearby variants with similar effects.
In a separate analysis, a University of Exeter researcher applied the Atlas to whole-genome data from more than 54,000 UK Biobank participants. Grouping rare variants by their predicted molecular effects revealed 22% more non-coding genetic associations than the analysis could otherwise detect.
Academic researchers can access the Atlas free through its web portal, the AlphaGenome API or Google Antigravity. Commercial access through Google Cloud is planned. AlphaGenome has not been validated or cleared for clinical use.