DeepMind

AlphaGenome Atlas: Molecular predictions for 9 Billion human DNA variants

Sep 8, 2026AlphaGenome Atlas team
AlphaGenome Atlas: Molecular predictions for 9 Billion human DNA variants
SummaryAI, from the source post

This post introduces AlphaGenome Atlas, a 1-petabyte dataset containing predictions for the molecular impact of 9 billion single-nucleotide variants in the human genome. It details the creation of the AlphaGenome Variant Impact (AVI) score, which combines predictions from AlphaGenome and AlphaMissense models, and introduces AVI feature attributions to explain the drivers of these scores. The post highlights the application of this resource in identifying variants for rare disease research and mapping rare variants associated with complex traits, demonstrating a significant expansion in the scale and accessibility of genomic variant prediction.

More from the AI for Biosecurity and Drug Discovery trail