Google DeepMind rises above the AI scrum with genome atlas

Google's DeepMind has unveiled AlphaGenome Atlas, a vast database containing a petabyte of data predicting the effects of nine billion possible nucleotide variations in the human genome. This platform, now publicly available to researchers, aims to address the challenge of pinpointing which genetic variations are responsible for specific traits or diseases.

The database builds upon DeepMind's previous AlphaGenome model, which could predict the impact of genetic variants on biological processes. To make the model more accessible, DeepMind precomputed AlphaGenome's predictions at scale and created an easily searchable resource. Each predicted variation is assigned an AlphaGenome Variant Impact (AVI) score, which helps researchers rank genetic variants by their potential impact, allowing them to focus on the most significant ones.

In an experiment with the GREGoR Consortium, researchers used AVI scores from the database to identify a genetic variant affecting DNM1, a gene linked to epileptic encephalopathy. The AVI scores also revealed the mechanism by which the genetic variant caused an abnormal extension of the resulting protein. While the content of the AlphaGenome Atlas is still based on predictions, DeepMind notes that as the AlphaGenome models improve, these predictions should become more accurate as well.