AI system RFdiffusion3 designs proteins at atomic level, achieves desired function in experiments

A new AI system called RFdiffusion3, capable of designing proteins at the atomic level, has been officially released. This system, developed by an international team including researchers from the University of Washington, Oxford University, and NVIDIA, was published in the journal Science on October 16. The team first open-sourced RFdiffusion3 on December 3, 2025. The system allows for precise design of proteins and nucleic acids by adjusting the three-dimensional coordinates of individual atoms. Researchers demonstrated that proteins designed with RFdiffusion3 successfully performed expected chemical reactions in experiments. David Baker, a professor at the University of Washington and recipient of the 2024 Nobel Prize in Chemistry, led the research. RFdiffusion3 outperformed its predecessor, RFdiffusion2, in 37 out of 41 tests involving enzyme active sites. The system is now expected to be widely used in various biological molecule design tasks.