Direct discovery · Biology

RFdiffusion applies diffusion models to de novo protein design

Experimentally validated de novo protein designs across monomers, assemblies, scaffolds, and binders.

Summary

The Nature paper introduces RFdiffusion, a generative method adapted from RoseTTAFold for designing protein backbones and protein-binding proteins. The authors report experimental validation across monomers, oligomers, symmetric assemblies, enzyme active-site scaffolds and binders.

AI role

Generated protein backbones and binders using a diffusion model adapted from RoseTTAFold.

Narrative role

RFdiffusion is important because it moves the protein story from predicting natural structures toward generating experimentally tested new molecules.

Caveat

The outputs are design candidates and demonstrations, not a general guarantee of biological function in every target setting.