AI system RFdiffusion3 can design proteins at the atomic level, paper published in Science

AI system RFdiffusion3 can design proteins at the atomic level and can achieve the intended functions in experiments

2026-10-19 07:22 IT Home - Yuanyang

IT Home, October 19 news: The detailed design plan of RFdiffusion3, a generative artificial intelligence system that can design proteins and nucleic acids by assembling them atom by atom, has now been officially released. Proteins newly designed with RFdiffusion3 can achieve the intended functions in real chemical reactions.

An international joint research team composed of institutions including the University of Washington in the United States, the University of Oxford in the United Kingdom, and NVIDIA has built this system, which can model individual atoms and is used to design proteins and nucleic acids, and published the related research results in the international journal Science on the 16th local time.

IT Home notes that the research team had already open-sourced RFdiffusion3 on December 3, 2025. The paper published in Science explains in detail the design principles of RFdiffusion3.

AI-based protein research has progressed from predicting the three-dimensional structures of existing proteins to designing entirely new proteins with target structures and functions.

David Baker, a professor at the University of Washington, is the corresponding author of this RFdiffusion3 study. He won the 2024 Nobel Prize in Chemistry for his contributions to using computers to predict protein structures. One of Baker's signature achievements is the development of RosettaFold, an artificial intelligence system for protein structure prediction.

RosettaFold learns the spatial relationships between the amino acids that make up proteins and uses this information to predict the three-dimensional structures of proteins. RFdiffusion1, released in 2023, and RFdiffusion2, released in 2025, were both developed based on the structure prediction network used by RosettaFold.

RFdiffusion uses diffusion models to determine the positions of amino acids, thereby completing protein design. Diffusion models are a class of artificial intelligence models that learn the distribution patterns of existing data and then generate new data.

RFdiffusion3 was developed with reference to the network architecture of AlphaFold3, another protein structure prediction program. RFdiffusion3 can represent a single amino acid with up to 14 atoms and designs entirely new protein structures by directly adjusting the three-dimensional coordinates of each atom.

With this system, the research team achieved precise design of enzyme active sites; enzymes can catalyze various chemical reactions. Enzymes are proteins and are biological catalysts that can accelerate the rates of various chemical reactions in living organisms.

When the researchers not only specified the positions of the atoms required for the reaction but also set the conditions for hydrogen bond formation and the distribution ranges of specific atoms within the protein, RFdiffusion3 was able to generate protein structures that satisfied all the constraints.

The researchers selected 41 active sites from existing enzymes to compare design performance. The results showed that in 37 of the test groups, RFdiffusion3 performed better than RFdiffusion2. Its speed in generating protein structures was about ten times that of RFdiffusion2.

The paraoxon-degrading enzyme and peptide-cleaving enzyme designed by RFdiffusion3 both successfully completed the intended chemical reactions in real experiments. The research team stated: "In the future, RFdiffusion3 will be widely used in various biomolecular design tasks and related application fields."

Reference materials:

Advertising statement: The external jump links contained in the article (including but not limited to hyperlinks, QR codes, passwords, etc.) are used to convey more information and save selection time. The results are for reference only. All articles on IT Home that contain external links include this statement.