Google’s Atlas of the human genome could pave the way for new treatments

Details

DeepMind has released AlphaGenome Atlas, a resource containing predictions for all 9 billion possible single-letter DNA substitutions in the human genome. Critically, coverage extends beyond protein-coding regions to the non-coding stretches that control when and how genes are expressed—territory that has historically been harder to analyze systematically. For each variant, Atlas predicts molecular effects such as changes to protein production levels. The platform is accessible via a public web portal, as a skill in Google's Antigravity agentic development platform, and through the AlphaGenome API. Alongside Atlas, DeepMind is releasing a Variant Impact Score (AVI) that combines multiple models to rank variants by likely biological importance—enabling researchers to prioritize which of the billions of candidates warrant closer investigation. Atlas builds on two prior DeepMind tools: AlphaGenome (released last year to identify genetic disease drivers) and AlphaMissense (focused on predicting effects of small protein-coding mutations). Atlas goes substantially further in scale and genome-wide scope. DeepMind's genomics lead Ziga Avsec noted that while the underlying AlphaGenome model was already released, compiling genome-wide predictions into an accessible catalogue required additional work. For drug discovery and disease research, the practical value is prioritization at scale: teams can rapidly rank millions of candidate variants without manually running individual predictions, potentially shortening the path from genetic association studies to therapeutic target identification.