Trusence Every claim has a source
Last updated 18 November 2026 Search Türkçe
Research

DeepMind's AlphaGenome Atlas ranks genetic variants to guide disease research

Researchers can now screen billions of possible DNA changes and focus lab work on the few most likely to matter.

A magnifying glass rests over a tray of countless small glass beads, with a few bright ones left highlighted.
AI-generated illustration

Google DeepMind has released AlphaGenome Atlas, a dataset built with its AlphaGenome model that predicts the outcomes of all roughly 9 billion possible single-letter changes in the human genome. Researchers can use the resulting AVI score to rank genetic variants by likely disruptive impact, which helps them decide which variants to study in the lab. Early users include a University of Exeter researcher who narrowed 1 billion variants to 19 groups linked to body mass index, and a Broad Institute team that used the score to identify a variant tied to neurodevelopmental disorders. The team has also introduced AlphaGenome 2, which aims to predict how variants affect individual cell types rather than whole tissues, and is researching ways to quantify how confident the model is in each prediction.

Why it matters

For people studying the causes of disease, the main shift is in where effort goes. Instead of testing variants one by one, researchers can use the AVI score to rank the roughly 9 billion possible single-letter changes in the genome by likely impact and choose which to study first. Early results, such as narrowing 1 billion variants to 19 groups linked to body mass index, show how much that filtering can cut. The next model, AlphaGenome 2, targets individual cell types rather than whole tissues, which would make such predictions more specific.

Sources · 1