RFdiffusion

RFdiffusion is the Baker lab's diffusion model for de novo protein design: a RoseTTAFold-derived network that generates novel protein backbones directly from noise, conditioned on functional constraints. One unified framework handles unconditional generation, motif scaffolding, protein binder design, symmetric oligomers, and enzyme active-site scaffolding — the dominant backbone-generation engine in modern protein design, typically paired with ProteinMPNN for sequence design and AlphaFold for validation.

Published in Nature (Watson et al., 2023) and made free for both non-profit and for-profit use in March 2023 under a governed license, RFdiffusion's designs have been experimentally validated in the lab across binders, enzymes, and symmetric assemblies. It is backbone-only — sequences still come from a downstream inverse-folding step — and compute-heavy, but it remains the standard starting point for generative protein design pipelines.