Google DeepMind has unveiled the AlphaGenome Atlas, a comprehensive database covering 9 billion potential single-letter mutations in the human genome. The project utilizes the AlphaGenome model to predict thousands of effects for each mutation, including changes in gene activity and chromatin structure. The resulting dataset spans approximately 1 petabyte of information.
The atlas incorporates a specific metric, AlphaGenome Variant Impact (AVI), designed to help researchers prioritize mutations for further study. The tool has already assisted the Broad Institute in identifying a non-coding DNA variant linked to severe epilepsy. Additionally, the database includes over 100 million short insertions and DNA fragments.
A significant feature of the atlas is its ability to analyze non-coding DNA, which constitutes 98% of the human genome. DeepMind has mapped thousands of short sequence motifs, suggesting their roles in gene activation or suppression across various cell types. The database is now available for free for non-commercial use.
Despite its capabilities, the system currently analyzes mutations individually against a reference genome and does not account for combinations of variants or unique individual genetic differences. Developers and experts emphasize that the atlas is intended for research prioritization rather than clinical diagnosis or replacing experimental validation.
