RFdiffusion
Generative diffusion for de novo protein backbone design
What's new
RFdiffusion3 (open-sourced 2025) is an all-atom foundation model ~10× faster than RFdiffusion2, now distributed via the Rosetta Foundry.
RFdiffusion brought denoising diffusion to protein design, generating novel backbones for binders, symmetric oligomers, motif scaffolds and enzyme active sites. Designs are typically paired with ProteinMPNN for sequence design and a structure predictor for validation.
The lineage
- RFdiffusion — backbone generation conditioned on motifs, symmetry and target hotspots.
- RFdiffusion2 — improved enzyme and all-atom design.
- RFdiffusion3 — a unified all-atom foundation model, ~10× faster, covering protein–protein, protein–DNA, protein–small-molecule binding and enzyme design.
Related tools
Open source
AI Models
ProteinMPNN
Fast, robust inverse folding — sequences for a target backbone
Python libraryCommand lineModel weights
Open source
AI Models
BindCraft
One-shot de novo binder design with AlphaFold2 backpropagation
Python libraryCommand line
Open source
AI Models
BoltzGen
Toward universal binder design across modalities
Python libraryCommand lineModel weights