Google DeepMind maps 9 billion genetic changes with new AI tool

Google DeepMind has launched an AI-powered database that maps the potential biological effects of about 9 billion possible single-letter changes in human DNA, giving scientists a new way to sift through genetic variations at a scale that would be difficult to study experimentally one by one.Called AlphaGenome Atlas, the platform uses predictions generated by DeepMind's AlphaGenome model to estimate how individual DNA changes could affect biological processes across different cell types and tissues.The company describes the Atlas as its most comprehensive catalogue yet of the predicted molecular effects of genetic variants. Rather than testing every possible DNA change in a laboratory, researchers can use the database to identify variants that appear most likely to have a biological effect and then focus experiments on those candidates. A massive genetic mapDeepMind has made AlphaGenome Atlas available for non-commercial use through a web portal, with access also offered through the AlphaGenome application programming interface, or API. Commercial access through Google Cloud is planned, the company said.The database is about 1 petabyte in size, making it more than 30 times larger than the AlphaFold Database, another major DeepMind resource that catalogues predicted protein structures.The Atlas is designed to be accessible through a regular web browser, allowing researchers to search genetic variants without writing code. Academic researchers can use it for free.The broader goal is to help scientists identify genetic changes that could be important in disease, understand how DNA influences biological processes and decide which variants warrant further experimental work.Looking for answers in rare diseasesOne early application involves rare diseases whose genetic causes remain unclear.Researchers at the Broad Institute, working with the GREGoR Consortium, used a score generated by AlphaGenome to assess the potential impact of genetic variants. The analysis highlighted a previously overlooked change affecting the DNM1 gene, which has been linked to epileptic encephalopathy, according to DeepMind.AlphaGenome predicted that the variant could create an incorrect splice site—a change in the way genetic instructions are processed before a protein is made. The resulting protein could therefore be abnormally extended.Laboratory experiments subsequently supported the prediction and found that nearby genetic variants could produce similar effects, DeepMind said.The example illustrates the role DeepMind sees for the Atlas: not replacing laboratory research, but helping scientists decide where to look first.Finding genetic signals that might otherwise be missedResearchers have also used the Atlas to study genetic variation in large populations.In one analysis involving more than 54,000 participants from the UK Biobank, Gareth Hawkes, a Medical Research Council fellow at the University of Exeter, grouped rare genetic variants according to their predicted effects on molecular processes.DeepMind said the approach uncovered 22% more associations involving non-coding regions of DNA than would otherwise have been identified.Non-coding DNA does not directly provide instructions for making proteins, but it can influence when and where genes are switched on or off. Some of the variants identified in the analysis were linked to proteins such as PLA2G7, which has been associated with ageing, and EGLN1, a cellular oxygen sensor.Hawkes also used AlphaGenome to examine hundreds of millions of non-coding variants for possible links to body mass index. By concentrating on the 1% of variants that AlphaGenome predicted would have the strongest effects, the analysis identified 19 genetic regions for further study.DeepMind said the results show how AI-based predictions could help researchers make sense of the enormous number of genetic differences found in human DNA.The company cautions that AlphaGenome's predictions are a tool for prioritising research, rather than proof that a particular genetic variant causes a disease or biological trait. Experimental validation remains necessary to establish whether a predicted effect actually occurs in living cells or organisms. Also Read: Google DeepMind unveils next generation of drug discovery AI model