Human DNA contains around 3 billion base pairs, but we still understand far less of it than it may seem. Science has a relatively good understanding of the 2% that contains instructions for making proteins. The other 98%, known as non-coding DNA, still hides many of the answers about diseases and biological traits.
Google DeepMind has introduced AlphaGenome Atlas, a database that predicts the impact of every possible single-letter variant in the human genome. The system uses the AlphaGenome AI model to anticipate how these variants might alter molecular processes, including in regions that do not directly produce proteins.
A map of 9 billion genetic changes
Each single-nucleotide variant, known as an SNV, occurs when one letter in the DNA changes. Although it may seem like a tiny modification, it can affect protein production, gene activity, or the way RNA is processed.
AlphaGenome Atlas calculated the regulatory effects of around 9 billion possible genetic changes. The result takes up approximately 1 petabyte of data, a volume so large that analyzing it manually or through traditional searches would be impractical.
So, what is the solution? The Atlas lets you query this information through a web portal without any programming. This allows clinical researchers, biologists, and other specialists to explore relevant variants without first building complex computing infrastructure.
AVI summarizes a variant’s impact
To make interpretation easier, the platform introduces AlphaGenome Variant Impact, or AVI. This indicator combines the model’s predictions for coding and non-coding regions into a single score.
Instead of reviewing thousands of independent signals, a researcher can use AVI to prioritize the variants most likely to have important biological effects. It does not replace experimental validation, but it helps determine where time and resources should be focused.
A high score alone does not prove that a variant causes a disease. It indicates that the variant deserves more detailed investigation.
From rare variants to complex traits
The Atlas is already being used to support research in different areas of genetics. At the Broad Institute, Laura Covill’s team used AVI to prioritize variants linked to cases of rare diseases that still had no explanation.
The tool highlighted a variant in the DNM1 gene and predicted that it could create an incorrect splice site, a process known as splicing. This evidence helped resolve the case and showed how a variant outside the best-known regions can have significant consequences.
The second example is related to complex traits, such as body mass index. Gareth Hawkes applied AlphaGenome Atlas to data from more than 54,000 UK Biobank participants.
By grouping variants according to their predicted molecular effects, the analysis identified 22% more non-coding genetic associations. In addition, by focusing on the 1% of variants with the highest estimated impact, it found 19 genetic regions linked to body mass index.
AI as a filter for genetic research
The main value of AlphaGenome Atlas is not simply that it accumulates data. Its contribution lies in turning a space that is almost impossible to review into a list of hypotheses that scientists can study more quickly.
This is especially important because non-coding DNA is not simply empty space. It can contain signals that regulate when a gene is activated, in which tissue it is expressed, or how much RNA it produces. Understanding these signals could improve research into rare diseases, cancer, inherited disorders, and complex traits.
The platform is available through a web portal that does not require programming knowledge. This decision could expand access to advanced genomic tools, although AI predictions still need to be tested against experiments, clinical data, and other analytical methods.
AlphaGenome Atlas does not automatically turn every change in DNA into a diagnosis. What it offers is a high-resolution map to guide exploration. And in a field with billions of possibilities, knowing which ones to study first can significantly speed up the path toward an answer.
Original source
https://blog.google/innovation-and-ai/models-and-research/google-deepmind/alphagenome-atlas