Google DeepMind said today that it has created a massive new AI-powered map of the human genome.
The company has launched AlphaGenome Atlas, a 1-petabyte database containing AI predictions for the effects of 9 billion possible single-letter changes in human DNA – every such change possible across the genome. DeepMind calls it the most comprehensive catalog yet of how genetic mutations affect molecular biology.
The Atlas is built on AlphaGenome, DeepMind’s existing AI model for predicting how genetic variants affect biological processes. The advance is scale: rather than requiring researchers to run AlphaGenome on individual variants, DeepMind has precomputed the model’s predictions across the entire genome and assembled them into a resource scientists can search and analyze.
“Just as an atlas is a collection of maps, linking together features of the land like altitude and location, AlphaGenome Atlas charts the molecular effects of DNA variants across the genome.” the company said in a blog post.
It’s a strategy reminiscent of AlphaFold. DeepMind researchers Demis Hassabis and John Jumper won the 2024 Nobel Prize in Chemistry for developing the AI system, which predicts the three-dimensional structures of proteins. DeepMind subsequently made predictions for more than 200 million protein structures available through the AlphaFold Database, transforming access to structural information used by researchers around the world.
AlphaGenome Atlas takes a similar approach to genetic variation. At 1 petabyte, the new dataset is more than 30 times larger than the AlphaFold database.
The goal is to tackle a fundamental bottleneck in genetics. With roughly nine billion possible single-letter mutations in the human genome, experimentally testing each one is practically impossible. The Atlas instead gives researchers precomputed predictions of how those changes could affect molecular processes, potentially helping them narrow enormous numbers of variants to those most worth investigating.
The Atlas is free for non-commercial use. It is also available through AlphaGenome API, and as a skill in Google Antigravity.