RFdiffusion3 Enables Atomic-Level Design of Functional Proteins and Nucleic Acids

Researchers from Washington University, Oxford, and NVIDIA published details of RFdiffusion3 in Science, a generative AI system capable of designing functional proteins and nucleic acids at the single-atom level. The system generates brand-new protein structures roughly ten times faster than its predecessor and successfully performs real-world chemical reactions. This breakthrough shifts protein engineering from merely predicting natural structures to precisely creating entirely custom biological molecules with targeted real-world functions. It marks a major milestone for biotechnology, accelerating drug discovery, enzyme engineering, and synthetic biology applications. RFdiffusion3 incorporates design concepts from AlphaFold3, representing each amino acid with up to 14 atoms and directly manipulating 3D coordinates. In benchmark tests across 41 enzyme active sites, RFdiffusion3 outperformed RFdiffusion2 in 37 cases, successfully generating functional enzymes like organophosphate-degrading and peptide-cleaving enzymes.