Google DeepMind unveils AlphaGenome AI to predict DNA mutations and disease risk

Key facts

• Google DeepMind’s AlphaGenome AI model reads up to one million DNA letters with single-letter precision to predict how genetic changes cause diseases.
• AlphaGenome targets the 98 percent of non-coding DNA, formerly called junk DNA, that regulates gene activity and influences RNA splicing.
• In a leukaemia case study, the model compared normal and mutated DNA sequences and predicted increased activity of nearby genes linked to the cancer.
• The model is freely available to scientists for non-commercial research but is not intended for clinical use.
• Robert Goldstone, head of genomics at the Francis Crick Institute, described AlphaGenome as a breakthrough that turns the genome into a decipherable language for discovery.
• Ben Lehner of the Wellcome Sanger Institute warned that most existing biological datasets are too small and poorly standardised for optimal AI training.