DeepMind's AlphaGenome Lets AI Scan a Full Million-Letter Genome at Once

DeepMind's AlphaGenome, described in Nature, can analyze up to about 1 megabase of DNA at a time—roughly a million letters—outperforming prior AI models in 25 of 26 tests and predicting thousands of functional genomic tracks along with nearly 6,000 human genetic signals. Trained on both human and mouse genomes, it aims to illuminate how coding and non-coding regions interact and regulate genes. While praised as a milestone, experts caution that data quality and standardization remain limiting factors; the model could help diagnose rare genetic diseases, identify cancer mutations, and reveal drug targets, but it isn’t a complete solution yet.