An artificial intelligence model developed by Google’s DeepMind could significantly advance scientists’ understanding of human DNA and accelerate research into genetic diseases, cancer and new medicines, researchers say.
The model, known as AlphaGenome, is designed to interpret the human genome, often described as the complete biological blueprint for building and operating the human body. Scientists believe it could help explain why small genetic differences increase the risk of conditions such as high blood pressure, obesity, dementia and certain cancers.
AlphaGenome analyses DNA at an unprecedented scale, examining up to one million genetic “letters” at a time. Human DNA consists of about three billion such letters, represented by A, C, G and T. Only around 2% of this code forms genes that directly produce proteins, while the remaining 98%, known as the “dark genome,” plays a crucial regulatory role but remains poorly understood.
Researchers say AlphaGenome can help illuminate this dark genome by predicting how non-gene regions influence gene activity, including when genes are switched on or off and how they are spliced to produce different proteins. Crucially, the model can also forecast the biological impact of altering even a single letter in the DNA code.
DeepMind researcher Natasha Latysheva said the tool is intended to help scientists understand how functional elements of the genome work, potentially speeding up discoveries in basic biology and medicine. She added that AlphaGenome could help identify disease-causing mutations, guide drug target discovery and support the development of new therapies, including in synthetic biology and gene therapy.
The model was described in the journal Nature and released last year for non-commercial use. Since then, about 3,000 scientists worldwide have used it in their research.
At the University of Exeter, researchers are applying AlphaGenome to study genetic variants linked to obesity and diabetes. While large population studies have identified such variants, many are located in the dark genome, making their biological role difficult to interpret. Scientists say the AI model allows them to rapidly predict the effects of these variants and prioritise which ones to test in laboratories.
AlphaGenome is also being used in cancer research to distinguish between mutations that drive tumour growth and those that are incidental. Experts describe this capability as a major step forward in genomic AI.
Dr Robert Goldstone of the Francis Crick Institute called the model a major milestone, highlighting its ability to predict gene behaviour directly from DNA sequences. Researchers at the Wellcome Sanger Institute said large-scale testing has shown strong performance, although they stressed the system still needs refinement.
Scientists note that AlphaGenome is less accurate at predicting long-range gene regulation and variations across different tissues, areas that DeepMind aims to improve. Despite these limitations, experts say the model represents a major leap in combining genomics, biomedical research and artificial intelligence.
DeepMind, whose team won the 2024 Nobel Prize in Chemistry for its earlier AlphaFold protein-structure model, said AlphaGenome marks the beginning of a new phase in scientific discovery driven by AI-enabled breakthroughs.
With inputs from BBC