Google mapped every possible typo in your DNA

TL;DR: Yesterday, Google DeepMind released a massive, searchable AI-powered database of every possible one-letter change in human DNA (all 9 billion of them). It also predicts how each change could affect you. The database could help speed up rare disease and cancer research, but some experts are cautioning against using predictions as definitive answers.

What happened: The AlphaGenome Atlas database, which is available for free to academic researchers, has 1 petabyte’s worth of data. (That’s equivalent to 500 hard drives with 2TB of storage each…so, uh, no need to clear space on your laptop). It’s also more than 30 times larger than the team’s database of protein structures, AlphaFold—the work that won former DeepMind chief Demis Hassabis a share of the 2024 chemistry Nobel Prize.

Why this matters: A single-letter change in DNA can be the cause of diseases like sickle cell anemia and cystic fibrosis. More than half of rare disease patients go undiagnosed, partly because clinicians still struggle to identify the underlying DNA changes. Some mutations have also been tied to certain cancers—finding them normally involves sifting through billions of data points, which the Atlas could now help with.

DeepMind also released the AlphaGenome Variant Impact (AVI) score, which rates how much a genetic mutation is likely to matter. It blends AlphaGenome’s predictions with an earlier model that predicts whether small changes in proteins are harmful. The AVI score could help researchers sort DNA changes by importance and investigate what each one could do.

But wait: For now, Atlas is just a starting point for researchers. Some experts are questioning how accurate the predictions are and how researchers interpret the results.

While the AVI score’s tidiness is useful, it could be “easily misinterpreted,” says University of British Columbia genomicist Carl de Boer. Others say Atlas won’t replace experiments or account for individual cases in diagnoses—and warn that it shouldn’t be driving clinical decisions on its own.

DeepMind’s genomic lead acknowledged that the Atlas predictions are not as accurate as AlphaFold, and can miss some changes to regions like enhancers (DNA that makes other genes more active).

Bottom line: For the first time, researchers can look up any single-letter change in the human genome for free and get a quick read on whether it matters. What they do with that information will determine its usefulness. —LC