Google DeepMind Introduces AlphaGenome Atlas for DNA

Google DeepMind has created a predictive map of possible changes in human DNA to help scientists identify which genetic variants may cause disease or influence traits such as body mass index.

The tool is called AlphaGenome Atlas and works as an artificial intelligence database. Its AlphaGenome model has precomputed the possible effects of the 9 billion single-letter changes that can occur in the human genome.

The result occupies around 1 petabyte of data, far too much for a researcher to examine variant by variant. The Atlas lets you query that information quickly from a website, without needing to know how to code.

98% of DNA remains difficult to interpret

The human genome contains around 3 billion base pairs, the units that make up DNA. Scientists understand the roughly 2% that contains the instructions for making proteins fairly well, but the remaining 98% does not code for proteins and has long been considered more difficult to interpret.

That does not mean it is useless. Many of these regions regulate when, where and how much a gene is activated. An apparently small change can alter protein production or affect how a cell functions.

AlphaGenome Atlas aims to organize this territory. For each variant, the system predicts its impact in coding and non-coding regions, then summarizes the result in a measure called AlphaGenome Variant Impact, or AVI. The stronger the signal, the higher priority a researcher can give it among thousands of possibilities.

Two uses researchers are already testing

At the Broad Institute, Laura Covill's team used the AVI score to study a rare disease case that still had no explanation. The system flagged a variant in the DNM1 gene and predicted that it could create an incorrect splice site in RNA, the intermediate process that helps make proteins.

That prediction provided important evidence for resolving the case. It does not replace laboratory testing or prove on its own that a variant causes a disease, but it helps determine where to look first.

The Atlas was also tested with data from more than 54,000 UK Biobank participants, a large repository of biomedical information. By grouping variants according to their predicted molecular effects, researcher Gareth Hawkes found 22% more non-coding genetic associations linked to complex traits.

By focusing on the 1% of variants with the highest predicted impact, he identified 19 genetic regions linked to body mass index. These regions will serve as a starting point for more specific research.

What this changes for research

Until now, analyzing non-coding variants required combining large amounts of data with extensive manual work. AlphaGenome Atlas offers an initial classification so teams can focus sooner on the changes most likely to be relevant.

For you, this does not yet mean an automatic diagnosis or a new medical test. The Atlas is a research tool. Its value lies in speeding up the search for clues that must later be confirmed through experiments and clinical studies.

The next step will be to check how many of its predictions hold up when validated in patients and in the laboratory. If those predictions prove reliable, the map could help researchers better understand rare diseases and other traits whose connection to DNA remains hidden in the lesser-known 98% of the genome.