Unprecedented genetic map: AI reveals the fate of every possible mutation in the human genome

Nine billion possibilities, one outcome sought by scientists: Is this genetic change benign, or is it the hidden root of a disease? This is precisely the question a new tool launched by Google DeepMind aims to answer.

According to a report in the scientific journal Nature, the company unveiled the “AlphaGenome Atlas,” a massive digital map that predicts the impact of every possible variant among the nine billion changes in the three billion letters that make up the human genome.

The tool is built on the “AlphaGenome” model launched by DeepMind last year and is based on a concept that previously succeeded with “AlphaFold,” the model that predicted the structures of over 240 million proteins and won the 2024 Nobel Prize in Chemistry.

To process this enormous volume of data, which amounts to approximately one petabyte, the research team developed a new metric called the “Variant Effect Score.” This helps researchers distinguish harmful mutations from benign ones, potentially aiding in the diagnosis of rare diseases and understanding the causes of common ones.

Nevertheless, molecular biologist Ben Lehner of the UK’s Wellcome Sanger Institute cautions that the tool is “useful but not revolutionary” and that it does not yet replace clinical medical decision-making.