DeepMind's latest AI can foresee genetic illnesses.

About a decade ago, Žiga Avsec began his journey into genomics as a physics PhD student and is now a research scientist at Google’s DeepMind. He and his team have created AlphaMissense, a machine learning model capable of analysing missense variants (single-letter genetic code changes known to cause diseases like sickle cell anemia and type 2 diabetes) and predicting their likelihood of causing disease with 90% accuracy. The tool, which works like a large language model trained on human biology, assigns a pathogenicity score to each of the 71 million possible missense variants. The score represents the likelihood of a mutation causing disease, helping researchers identify potentially harmful mutations faster. Though the tool is significantly impactful, the team insists that its predictions should complement real-world research rather than replacing it.