Google DeepMind launches DNA predicting AI model, AlphaGenome

Google DeepMind has launched a new AI model aimed at analysing DNA and supporting the development of future health treatments.

The AI, called AlphaGenome, analyzes up to one million DNA letters to create a better understanding of patients with rare disorders, speed up drug discovery, and assist predictions for gene editing and other health technologies of the future.

Deepmind, in a statement, explained that the new AI tool outputs thousands of molecular property predictions such as gene start sites, splicing events and chromatin accessibility across hundreds of cell types and tissues.

AlphaGenome also scores the effects of single-nucleotide variants by comparing predictions on mutated versus unmutated sequences, the Google company added.

AlphaGenome is built on training data from large public consortia including ENCODE, GTEx, 4D Nucleome and FANTOM5.

Google said that AlphaGenome demonstrates state-of-the-art performance, outperforming prior models on 22 out of 24 sequence prediction tasks and matching or exceeding benchmarks on 24 out of 26 variant-effect evaluations.

It has launched an API preview for non-commercial researchers, with the full model slated for release in the future.