DeepMind Unveils AlphaGenome to Map Genome's Regulatory Code
AlphaGenome analyzes up to 1 million DNA base pairs to predict how mutations alter gene regulation, researchers say.
Overview
Google DeepMind released AlphaGenome, an AI model that analyzes up to 1 million DNA base pairs at once and predicts how mutations affect gene regulation, according to a paper published in Nature and a DeepMind press briefing.
The tool targets the non-coding 'dark matter' that comprises about 98% of the roughly 3 billion–base human genome and could narrow candidate mutations linked to heart disease, autoimmune disorders and cancer, DeepMind researchers said.
Pushmeet Kohli, DeepMind's vice president for science, said in a press briefing that AlphaGenome helps 'decipher the complex regulatory code,' and Marc Mansour, a clinical professor at UCL, called it a 'step change' for finding genetic cancer drivers.
DeepMind said AlphaGenome was trained on public human and mouse datasets, predicts nearly 6,000 human functional genomic tracks and matched or outperformed other models in 25 of 26 benchmark tests, according to the Nature paper.
DeepMind made a research version of AlphaGenome freely available to other scientists but said the tool is not for clinical use and that researchers, including Žiga Avsec, must improve uncertainty reporting and expand training beyond humans and mice.
Analysis
Center-leaning sources frame the story optimistically, using evaluative language ("revolutionary," "most comprehensive," "marks a true milestone") and emphasizing DeepMind's performance metrics and applications. Editorial choices prioritize company claims and potential benefits while relegating skepticism to a single expert caveat, shaping a narrative of technological breakthrough over balanced uncertainty.