The AlphaGenome Atlas maps the potential impact of nearly 9 billion genetic changes, helping researchers identify DNA variants that could influence diseases, biological processes and gene activity for further investigation.
Google DeepMind has introduced AlphaGenome Atlas, an AI-powered database designed to help scientists investigate the potential biological impact of genetic variations at unprecedented scale. The resource maps predicted effects for roughly 9 billion possible single-letter changes in human DNA, giving researchers a way to narrow down variants that may warrant closer experimental study.
The Atlas draws on predictions generated by DeepMind’s AlphaGenome model, which assesses how individual DNA changes could influence molecular processes across different tissues and cell types. Instead of experimentally examining every possible genetic alteration, researchers can use the database to identify changes predicted to have stronger biological effects and focus laboratory resources on those candidates.
DeepMind describes the resource as its most extensive catalogue of predicted molecular effects associated with genetic variants. The Atlas is available for non-commercial use through a web portal, while researchers can also access its capabilities through the AlphaGenome application programming interface (API). Commercial availability through Google Cloud is planned.
The database is approximately 1 petabyte in size, making it more than 30 times larger than DeepMind’s AlphaFold Database, which contains predicted protein structures. Researchers can search genetic variants through a standard web browser without having to write code, while academic users can access the Atlas without charge.
AI helps narrow the search for rare-disease clues
One of the early research applications of the technology involves rare diseases for which the underlying genetic causes remain difficult to establish. Researchers at the Broad Institute, working with the GREGoR Consortium, used an AlphaGenome-generated score to assess genetic variants and identified a previously overlooked change involving the DNM1 gene.
The gene has been associated with epileptic encephalopathy. According to DeepMind, AlphaGenome predicted that the variant could create an abnormal splice site, potentially altering the way genetic instructions are processed before a protein is produced. The resulting protein was predicted to be unusually extended.
Subsequent laboratory testing supported the computational prediction, with researchers also finding that nearby genetic variants could generate comparable effects. The work demonstrates how the Atlas could function as a prioritisation system, helping researchers decide which genetic changes should be investigated experimentally rather than replacing laboratory validation.
Exploring the vast landscape of non-coding DNA
Researchers are also applying AlphaGenome to genetic information collected from large populations. In one study involving more than 54,000 UK Biobank participants, Gareth Hawkes, a Medical Research Council fellow at the University of Exeter, grouped rare variants according to their predicted molecular effects.
DeepMind said the approach uncovered 22% more associations involving non-coding regions of DNA than would otherwise have been identified. Although non-coding DNA does not directly contain instructions for producing proteins, it can influence when and where genes are activated.
Among the associations identified were variants linked to proteins including PLA2G7, which has been associated with ageing, and EGLN1, a protein involved in sensing cellular oxygen levels.
Hawkes also used AlphaGenome to analyse hundreds of millions of non-coding variants for potential connections to body mass index. By concentrating on the 1% of variants that the model predicted would have the strongest effects, the research identified 19 genetic regions that could be investigated further.
DeepMind said such applications demonstrate how AI-based predictions could help researchers navigate the enormous volume of genetic variation present in human DNA. However, the company stresses that AlphaGenome’s predictions are intended to guide research rather than establish that a particular variant directly causes a disease or biological characteristic.
Laboratory and, where appropriate, organism-level studies will still be required to confirm whether predicted genetic effects actually occur. The Atlas therefore represents a way to accelerate the process of identifying promising candidates for research, rather than a substitute for experimental evidence.
See What’s Next in Tech With the Fast Forward Newsletter
Tweets From @varindiamag
Nothing to see here - yet
When they Tweet, their Tweets will show up here.
