Understand this story
An international joint research team published the RFdiffusion3 design scheme in Science. The team consists of institutions including the University of Washington, the University of Oxford, and NVIDIA; the system can model individual atoms and is used to design proteins and nucleic acids. RFdiffusion3 references the AlphaFold3 network architecture, uses up to 14 atoms to represent a single amino acid, and directly adjusts the three-dimensional coordinates of each atom. In comparative tests on 41 existing enzyme active sites, it outperformed RFdiffusion2 in 37 groups, and its generation speed was about ten times that of the latter; the paraoxon-degrading enzyme and peptide-cleaving enzyme it designed completed the expected chemical reactions in real experiments. The report also states that the team open-sourced RFdiffusion3 on December 3, 2025. There are currently no subsequent developments.
AI-generated from public reports
Latest development · Report
AI system RFdiffusion3 can design proteins at the atomic level, paper published in ScienceDevelopments
1 confirmed developments
Research team publishes RFdiffusion3 design scheme in Science
Source:IT Home · AI
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 19
IT Home · Artificial Intelligence
AI system RFdiffusion3 can design proteins at the atomic level, paper published in Science
An international team comprising institutions such as the University of Washington, the University of Oxford, and NVIDIA published the RFdiffusion3 design scheme in Science; the system can model individual atoms and is used to design proteins and nucleic acids. RFdiffusion3 references the AlphaFold3 network architecture, representing a single amino acid with up to 14 atoms and directly adjusting the three-dimensional coordinates of each atom. In comparative tests on 41 existing enzyme active sites, it outperformed RFdiffusion2 in 37 cases, with a generation speed about ten times faster; the paraoxon-degrading enzyme and peptide-cleaving enzyme it designed completed the expected chemical reactions in real experiments. The team open-sourced RFdiffusion3 on December 3, 2025.