An Artificial Intelligence model developed by Google’s Deepmind, called Alphagenome, is being hailed by researchers as a potentially transformative tool for understanding how the human genome works and how genetic variation contributes to disease. The model can analyse dna, described as the complete recipe for building and running the human body, and is designed to predict how both genes and the vast non-coding regions of the genome influence biological functions. Experts quoted in the article say the system could accelerate discoveries in genetic disease, cancer research and drug development, while its creators acknowledge that the technology is not yet perfect.
The human genome is described as being made up of three billion letters of dna code represented by A, C, G and T, with around 2% forming genes that code for proteins and the remaining 98% referred to as the “dark genome”. The article explains that this dark genome plays a crucial role in organising how genes are used and harbours many disease-linked mutations, but remains poorly understood. Alphagenome can analyse one million letters of code at a time and can predict where genes are located, how the dark genome affects gene expression and gene splicing, and the impact of changing even a single letter in the genetic code. Deepmind research engineer Natasha Latysheva says they see Alphagenome as a tool to understand functional elements in the genome, which they hope will accelerate fundamental understanding of the code of life and help pinpoint mutations that cause rare genetic diseases.
The model was described in the journal Nature and was released for non-commercial use last year, and 3,000 scientists have since used the tool in their research. Dr Gareth Hawkes at the University of Exeter is using Alphagenome to investigate how mutations in the dark genome might alter the risk of obesity and diabetes, noting that large studies sequencing the entire genetic code of tens of thousands of people have found relevant variants in these poorly understood regions. Alphagenome has also been used to predict which mutations are driving cancer and could be targeted by therapies. Dr Robert Goldstone at the Francis Crick Institute calls it a “major milestone in the field of genomic Artificial Intelligence” and an “incredible technical feat” for its ability to predict gene expression from dna sequence alone, while Prof Ben Lehner at the Wellcome Sanger Institute says the model has been tested in more than half a million experiments and is performing well but is still “far from perfect”. The article notes that Alphagenome is a “sequence-to-function” model trained on human and mouse cell data, that it struggles with long-distance gene regulation over more than 100,000 letters of code and tissue-specific accuracy, and that its development builds on Deepmind’s earlier Alphafold work, which won a Nobel prize for chemistry in 2024.
