DeepMind’s 9-Billion-Variant Atlas Opens the Genome’s Hardest Search

DeepMind’s 9-Billion-Variant Atlas Opens the Genome’s Hardest Search
Google DeepMind on Tuesday released AlphaGenome Atlas, a searchable AI database forecasting the molecular effects of roughly 9 billion possible single-letter changes in human DNA. The company’s pitch is straightforward: mutations that once demanded painstaking lab work or one-by-one model runs can now be investigated through a browser.

The scale reflects a longstanding genomics problem. Scientists mapped the human genome more than two decades ago, but, as DeepMind research vice-president Pushmeet Kohli put it, “we bought the book, but we did not understand how to read it.” Atlas runs predictions across gene regulation, expression and protein-related effects, including the largely non-coding portions of DNA that have been especially difficult to interpret.

Google is making the resource free for academic researchers, accessible without coding, while saying commercial access will follow through Google Cloud licensing. Sundar Pichai described it as an interactive map of all 9 billion variants and said he was “excited for the discoveries to come.”

Early testing offers some grounds for that optimism. In an unexplained case of epileptic encephalopathy, researchers used Atlas to flag a DNM1 variant that earlier blood-based RNA sequencing had not resolved; laboratory work subsequently confirmed the prediction and the variant was reclassified as likely pathogenic. In another analysis, University of Exeter researcher Gareth Hawkes said the tool could help “shrink the haystack” of possible disease-relevant variants.

But DeepMind itself draws a firm line between triage and diagnosis. Genomics lead Žiga Avsec said the predictions are “accurate enough to really point us in the right direction,” not “the universal truth.” The model performs better for some mutation types, such as splicing and promoters, than for enhancers, and its own paper says Atlas should form only one link in the evidence chain behind clinical decisions.

That restraint matters. The atlas may dramatically narrow the hunt, but biology—and the lab—still gets the final word.

Continue reading https://foxvector.com/stories/01a085fa-747c-0dde-739e-0dbb98331308