RFdiffusion3 AI Designs Proteins Atom by Atom, Ten Times Faster

A generative AI system called RFdiffusion3, built by a team from the University of Washington, the University of Oxford, and NVIDIA, can design proteins and nucleic acids atom by atom. Its design details were published in the journal Science and the system has been open-sourced. David Baker, winner of the 2024 Nobel Prize in Chemistry, led the research.

RFdiffusion3 borrows the network design of the protein structure predictor AlphaFold3. It can represent a single amino acid with up to 14 atoms and builds new protein structures by directly adjusting the 3D position of each atom, precisely designing the active sites of enzymes.

In tests on 41 active sites of existing enzymes, RFdiffusion3 beat its predecessor in 37 cases and works about ten times faster. The enzymes it designed completed their expected chemical reactions in real experiments, and it will be used for many biomolecule design tasks in the future.