Google DeepMind has introduced an artificial intelligence tool called AlphaGenome Atlas to examine the relationships between DNA changes in the human genome and biological processes. The company said the tool could accelerate research by making it easier to understand the genetic causes of diseases and could contribute to the development of new treatments.
Atlas provides a map containing predictions for every possible DNA letter change in the human genome. The genome consists of four chemical letters, abbreviated as A, C, G and T, and contains approximately three billion letter pairs. Changes in a single letter may be harmless, contribute to ordinary differences between people, or be associated with diseases. The predictions prepared for each of the approximately nine billion possible single-letter changes show how these changes could affect molecular processes such as protein production.
Google is also releasing the Variant Impact Score (AVI) tool to help researchers identify mutations that require further investigation. The predictions can be examined through a web portal, a skill on the Antigravity platform, and the AlphaGenome interface.
The project expands on AlphaGenome, introduced last year, and the AlphaMissense studies, which focus on small mutations that could affect proteins. The Atlas also covers regions that do not encode proteins but can regulate the behavior of genes. DeepMind’s genomics lead, Ziga Avsec, said that analyzing and precomputing billions of variants takes time.
The dataset generated by AlphaGenome, trained on publicly available human and mouse genome data, is approximately 1 petabyte in size. As of today, Atlas is available for non-commercial use and will soon be available for commercial use through Google Cloud. The project joins Google’s science-focused AI efforts such as AlphaFold, weather forecasting, and “co-scientist.” AlphaFold had earned Demis Hassabis and John Jumper the 2024 Nobel Prize in Chemistry.
Why it matters
This scope makes it possible to assess not only regions that produce proteins in genetic research, but also the regions that regulate how genes function. For researchers, the main change is that candidates that could be prioritized in the laboratory or as part of disease studies can be selected more systematically from among billions of possible variants. However, the results provided by Atlas are predictive; which changes are actually associated with disease and how they affect cellular processes still need to be separately validated. While making the data available for non-commercial use provides access to a broad group of researchers, the fact that commercial use will be offered through Google Cloud shows that access to the tool will differ depending on the purpose of use.
Background
Google is not a new name in the FikirPilot archive: we have published 30 news stories mentioning this name in the past 90 days; the latest is dated 9 September 2026.