The human genome is the complete set of DNA instructions in a person. Ninety-eight percent of the human genome is non-coding DNA. It does not make proteins directly, but it can control when and how genes are used. Changing a single letter in these non-coding regions can disturb processes such as protein production.
Google DeepMind has introduced AlphaGenome Atlas, a database of every possible change of one DNA letter at one position. At each of the three billion positions there are three other letters that could appear, which gives about nine billion possible one-letter changes. The database shows how each change may alter the controls that turn genes up or down. The stored results fill about one petabyte. Researchers can query this store instead of calculating each variant from scratch.
How early users have applied the map
To make the data easier to use, the Atlas adds the AlphaGenome Variant Impact score, or AVI score. This is a single number that combines predictions for both protein-coding and non-coding regions, so a person does not have to scan thousands of separate measurements to decide which variants deserve a closer look. At the Broad Institute, Laura Covill and colleagues used the score while studying unsolved rare diseases. It flagged a change in the DNM1 gene and predicted that the change created an incorrect splice site, a cut-and-join point that cells use when they process genetic messages. That prediction gave supporting evidence that helped close the case. In other work, Gareth Hawkes applied the Atlas to records from more than fifty-four thousand people in the UK Biobank, a large health research collection. Grouping variants by their predicted molecular effects revealed twenty-two percent more links between non-coding DNA and complex traits, which are features influenced by many genes and by the environment. When the analysis was limited to the most impactful one percent of variants, it pointed to nineteen genetic regions tied to body mass index, a common measure of body weight relative to height.
The Atlas is available through a website that needs no computer programming. The authors present it as a way to give clinicians and biologists faster, more grounded starting points for genomic research. The results are predictions from a model, not laboratory measurements of every variant.