When the world’s scientists finally pieced together a first draft of the human genome in 2003, one of the biggest surprises was just how little of it — only about 20,000 genes — are involved in the business of producing proteins. At first, the remaining 98% appeared not to do much of anything at all. With better tools, researchers began to discover that this “junk DNA” actually exerts a tremendous amount of influence on how and where and when protein-coding genes get expressed. But more than two decades later, making sense of these complicated interactions — and how they contribute to disease — remains one of biology’s most perplexing puzzles.
Now, a growing number of researchers are turning to an artificial intelligence developed by Google’s AI research company DeepMind to predict how DNA encodes gene regulation, with an eye toward possible applications in therapeutic development. The model, called AlphaGenome, was first described in a preprint and blog post last June. At the time, DeepMind’s vice president of science, Pushmeet Kohli, called it a step toward understanding the “semantics of DNA,” but far from a complete solution.
Since its launch seven months ago, nearly 3,000 scientists from 160 countries have started using it to advance research into cancer, neurodegenerative disorders, and infectious diseases, Kohli told reporters at a press briefing Tuesday. Initially, the company limited the use of AlphaGenome to researchers doing noncommercial work, who could only access the model through DeepMind’s servers via a free API. With its growing user base, Kohli said AlphaGenome currently has an API call volume of about 1 million per day.
This article is exclusive to STAT+ subscribers
Unlock this article — and get additional analysis of the technologies disrupting health care — by subscribing to STAT+.
Already have an account? Log in
