On September 8 (local time), Google DeepMind announced the release of "AlphaGenome Atlas," a database that predicts the molecular-level impact of approximately 9 billion single-base mutations (single-character variations) in the human genome. The dataset spans 1 petabyte, making it more than 30 times larger than the AlphaFold Database, which aggregates protein structure predictions.

The platform introduces the "AVI (AlphaGenome Variant Impact) score," a single numerical value representing the impact of a mutation. By integrating predictions from two models, AlphaGenome and AlphaMissense, the score enables the ranking of high-impact mutations not only in protein-altering regions but also in non-coding regions (approximately 98% of the genome) that regulate gene function. Additionally, details on molecular processes such as RNA splicing and gene expression are provided.

DeepMind stated that the tool will be useful for researching rare diseases of unknown cause and identifying mutations associated with complex traits. In early use cases, the Broad Institute in the U.S. utilized the tool to identify genetic mutations related to epileptic encephalopathy, and the University of Exeter in the U.K. reported success in identifying genomic regions related to BMI.

Access is provided through three formats: a no-code web portal, the AlphaGenome API, and Google Antigravity skills. Non-commercial use is available for free starting from the date of announcement, with commercial access expected to be provided via Google Cloud soon.

Source: