DeepMind publishes AlphaGenome study

Scientists at Google DeepMind in London have published a detailed study of AlphaGenome, an artificial intelligence system that predicts how changes in DNA affect the way genes are switched on and off.

The study was published on 28 January, describing a model the company first made available to researchers last year.

The dark matter of the genome

The human genome contains more than three billion letters of DNA code. Only a small fraction of it directly contains the instructions for making proteins. Much of the rest helps to control when, where and how strongly genes are used, but scientists have struggled to understand exactly how.

That part of the genome, sometimes called its dark matter, is important for medicine, because many genetic variants linked to disease lie outside genes.

What AlphaGenome does

AlphaGenome can read stretches of DNA up to a million letters long and make thousands of predictions about what the sequence does, including how active nearby genes will be, how the DNA is packaged and how genetic messages are spliced together.

Crucially, it can predict the effect of changing a single letter in that long sequence. That could help scientists work out which genetic variants cause disease and which are harmless.

The model can make predictions across 11 different types of biological measurement at once. In tests, it outperformed the best existing models in most of the comparisons made.

Limitations

The model was trained on data from humans and mice, and has not been tested on other species. It also finds it harder to predict the effects of changes that lie very far from the genes they affect.

The Alpha family

AlphaGenome is the latest in DeepMind's family of scientific AI tools. Its best known is AlphaFold, which predicts the shapes of proteins and won a share of the 2024 Nobel Prize in Chemistry for Demis Hassabis and John Jumper.

DeepMind was founded in London in 2010 and is now part of Google. Its headquarters are in King's Cross.

Free for research

The model can be used free of charge by researchers for non-commercial work, through an online interface.

Scientists say tools like AlphaGenome could speed up research into rare genetic diseases and cancer, and help doctors interpret the results of genetic tests.

The study adds to growing evidence that AI is changing the way biology is done, allowing researchers to explore questions that would take years in the laboratory.