The human genome holds about 3 billion DNA letters. That adds up to more than 9 billion possible single-letter changes a scientist could study.
Testing each change by hand in a lab would take far too long. Google DeepMind built a new tool to help with that problem.
The tool is called the AlphaGenome Atlas. It became available today as a searchable resource for scientists around the world.
Google DeepMind worked with the Stowers Institute for Medical Research on the project. Investigator Julia Zeitlinger led the biology side of the work.
Žiga Avsec, Vice President of Science and Chief Scientist at Google DeepMind, led the technical team. Their goal was to map how DNA controls what happens inside cells.
The atlas holds one petabyte of data. It includes AI-generated predictions for the molecular effects of every possible genetic change across hundreds of cell types.
Pushmeet Kohli, VP of Science at Google DeepMind, said the tool shows how AI can expand scientific knowledge. He said the goal is to help researchers understand what happens when single letters in the genome change.
Scientists can use the atlas through a normal web browser. No coding skills are required, which opens it up to more researchers than before.
How the Atlas Maps DNA Motifs
Every cell in the body carries the same DNA. But cells use that DNA in different ways depending on their type.
Short DNA sequences called motifs act like instructions. They help decide which genes turn on, when they turn on, and how strongly.
Zeitlinger's team used the atlas to study these motifs across the genome. They sorted them by function, separating factors that open up DNA from those that switch genes on or off.
Doing this kind of testing in a lab across many cell types would take years. The atlas let the team spot patterns much faster.
Zeitlinger said the mapped motifs work like a searchable dictionary for parts of DNA that don't code for proteins. She said this can help researchers find disease-linked variants faster.
From Billions of Variants to Real Discoveries
The atlas also introduces a new measurement called the AlphaGenome Variant Impact score, or AVI. It combines predictions from AlphaGenome, AlphaMissense, and evolutionary data.
The AVI score gives researchers one number to rank how much a DNA variant might affect a person's biology. That includes effects on gene activity, RNA processing, and protein function.
Researchers at the Broad Institute already used the AVI score. They found a previously overlooked variant tied to an unsolved rare disease case.
At the University of Exeter, scientists applied the atlas to genetic data from more than 54,000 UK Biobank participants. They found new links between rare noncoding variants and protein levels in the body.
Stowers Institute leaders said the atlas does not replace lab work. Instead, it helps scientists decide which variants to test first, saving time and resources for the most promising leads.
