DeepMind’s New AlphaGenome Atlas Maps How Single-Letter DNA Changes Affect the Genome

Google DeepMind has released AlphaGenome Atlas, a database that shows the likely effects of every possible change to one letter of human DNA.

Human DNA is made up of a vast number of base pairs. Only 2% of it codes for proteins, and researchers understand that part fairly well. The other 98% has stayed largely unknown, even though Google DeepMind’s AlphaGenome AI model has already shown that a change in these regions that do not code for proteins can trouble processes like protein production.

What the Atlas Contains

AlphaGenome Atlas uses the AlphaGenome AI model to work out, in advance, the effect of every possible change to a single letter across the genome. The result is a dataset of enormous size. To make that amount of information easy to work with, the Atlas introduces the AlphaGenome Variant Impact score, or AVI, which brings together predictions for coding regions and for regions that do not code for proteins, into one figure. Researchers can use the AVI score to decide which variants deserve closer study, instead of sorting through thousands of individual data points by hand.

A Rare Case Solved

At the Broad Institute, Laura Covill and her team used the AVI score to rank variants in a case of rare illness that had not yet been solved. The score pointed to a critical variant in the DNM1 gene, with the AlphaGenome model showing that it likely caused an incorrect joining point in the gene’s message. This provided supporting evidence to solve the case.

Tracing Complex Traits

Finding rare variants in regions that do not code for proteins, and linking them to complex traits, is difficult because of noise in the data. Dr. Gareth Hawkes applied AlphaGenome Atlas to data from more than 54,000 people in the UK Biobank, grouping variants by their likely molecular effects. That approach uncovered 22% more genetic links in regions that do not code for proteins than before. Looking only at the top 1% of the strongest variants, Hawkes found 19 genetic regions linked to body mass index, giving his team a clearer direction for further study.

Reaching Researchers Everywhere

Google DeepMind is offering AlphaGenome Atlas through a website portal that requires no coding skills. That design choice is meant to open the resource to clinical researchers and those who study living things, wherever they are, who might not otherwise have the training to query a dataset this size. Google DeepMind describes the release as part of its aim to move genomic discovery and science forward for all.

Here’s a quick look at how the Atlas project has developed so far, based on details in the announcement:

Every possible change to a single letter, worked out across the whole genome and gathered into one enormous dataset, is the scale of the data behind this release, and it is what sets it apart.

An enormous number of possible changes, a vast body of results, and a single score meant to make sense of it all: Google DeepMind is betting that this combination will help researchers move faster on questions that have sat open for years. The DNM1 case and the UK Biobank findings offer early signs of what that faster pace might look like in practice, though the wider payoff will depend on how many researchers pick up the resource now that it is public.

Source: blog.google

AI research · AlphaGenome · DeepMind · genomics · UK Biobank