Google DeepMind's AlphaGenome Atlas scores ~9 billion human DNA variants, with AVI beating CADD in clinical benchmarks

DeepMind released the AlphaGenome Atlas, a 1-petabyte dataset of precomputed functional-effect predictions for roughly 9 billion human genome variants, plus a new AVI score that outperformed CADD on clinically classified variants and helped reclassify an…

Google DeepMind released the AlphaGenome Atlas, which precomputes functional-effect predictions for approximately 9 billion human genome variants, with about 27,000 prediction values per variant, in a one-petabyte dataset more than 30 times larger than the AlphaFold database. The underlying AlphaGenome AI system predicts gene expression, transcription factor binding, chromatin accessibility, and splice site usage for genomic sequences, but is currently limited to human and mouse sequences and a limited set of well-studied cell types. Alongside the atlas, DeepMind introduced the AlphaGenome Variant Impact Score (AVI), a small neural network combining AlphaGenome, AlphaMissense, and evolutionary conservation features using only 18 inputs versus CADD's 150+. AVI outperformed existing tools on clinically classified variants, ranking causal variants among the top 50 candidates in 29.5% of solved GREGoR cases compared with 12.5% for CADD. In one GREGoR epilepsy case, AVI elevated a previously unclear DNM1 splice variant that lab experiments subsequently confirmed as likely disease-causing. A UK Biobank analysis of 54,000 genomes using the atlas found 22% more links to noncoding variants. The atlas is available for noncommercial use via a web portal, API, and a Google Antigravity skill, with a commercial version also mentioned. Ars Technica notes the system's predictions generally match or exceed specialized bioinformatics tools, though its scope is constrained to human and mouse sequences and well-studied cell types.

• AlphaGenome Atlas precomputes impact predictions for ~9 billion human genome variants (~27,000 prediction values per variant) in a 1-petabyte dataset
• The atlas dataset is more than 30 times the size of the AlphaFold database
• AVI uses 18 input features (AlphaGenome, AlphaMissense, evolutionary conservation) versus CADD's 150+
• AVI ranked causal variants in the top 50 candidates for 29.5% of solved GREGoR cases versus 12.5% for CADD
• A DNM1 epilepsy splice variant flagged by AVI was confirmed by lab experiments as likely disease-causing
• UK Biobank analysis of 54,000 genomes found 22% more noncoding-variant links using the atlas
• AlphaGenome predicts gene expression, transcription factor binding, chromatin accessibility, and splice site usage; performance generally matches or exceeds specialized bioinformatics tools
• System is limited to human and mouse sequences and a limited set of well-studied cell types