Google DeepMind released a free research map of every possible one-letter change in the human genome. It can help scientists choose what to test, but laboratory and clinical proof still matter.
New to this? Read it in simple words
- Google DeepMind published AlphaGenome Atlas, a free map of predictions about DNA changes.
- Human DNA has about 3 billion letters. The Atlas covers about 9 billion possible one-letter changes.
- It can help scientists choose what to test. This makes a hard first step much faster.
- A prediction cannot show by itself that a change causes a disease. Lab tests and patient data are still needed.
- Genome
- The full set of DNA in a living thing.
A very large map of small changes
Human DNA has about 3 billion letters. At each place, one letter can change into three other letters. AlphaGenome Atlas stores predictions for almost all 9 billion possible single-letter changes.
The dataset is about one petabyte. Google DeepMind says this is more than 30 times the size of the AlphaFold database. Researchers can explore it on a free website instead of running every prediction themselves.
Most human DNA does not directly write proteins. Changes in this large non-coding area can still affect when a gene turns on or off. This makes their effects hard and expensive to study one at a time.
The Atlas ranks possible one-letter DNA changes. Laboratory and clinical evidence must confirm any important result.
The atlas can help choose what to test
AlphaGenome predicts how a DNA change may affect several biological processes. The Atlas also includes an AVI score. This score combines signals from AlphaGenome and AlphaMissense to rank changes that may deserve attention.
A researcher can begin with a gene, a DNA area, or a possible condition. The tool can show which changes may affect gene activity or proteins. This can reduce a huge list before laboratory work begins.
Nature reports that outside scientists see value in the broad map, especially for rare changes. However, a useful ranking is not the same as knowing the real effect inside one person or one type of cell.
Prediction must lead to evidence
The model learned from existing biological data. Gaps or bias in that data can shape its predictions. A score may also miss effects that depend on age, environment, ancestry, or several DNA changes working together.
Scientists should use the Atlas to form a question, not to close one. Laboratory tests, patient data, and clinical review are still needed. Doctors should not use a prediction alone to diagnose or treat a person.
The release is important because it makes a difficult first step much faster. Its value will grow when researchers publish both successful checks and clear cases where the model was wrong.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1Primary source · September 8, 2026AlphaGenome Atlas: Molecular predictions for 9 billion human DNA variants Google DeepMind
- 2Research · September 8, 2026Google DeepMind releases map of 9 billion possible human DNA changes Nature
- 3Research · September 8, 2026New AlphaGenome Atlas could transform our understanding of genetic diseases Scientific American
We used Google DeepMind for the dataset size, access, and model design. We checked the meaning and limits with Nature and Scientific American. We clearly separate an AI prediction from laboratory evidence, diagnosis, or treatment advice.