RFdiffusion

RFdiffusion fine-tunes the RoseTTAFold structure prediction network on protein structure denoising tasks to generate protein backbones for diverse functional protein design from simple molecular specifications.

Key Features:

• Unconditional protein monomer design: Generates protein structures without conditional information.
• Topology-constrained protein monomer design: Generates protein structures with specified topological constraints.
• Protein binder design: Generates proteins that bind to specific targets, including de novo binder design.
• Symmetric oligomer design: Generates higher-order symmetric protein architectures.
• Enzyme active site scaffolding: Generates protein scaffolds that support enzyme active sites.
• Symmetric motif scaffolding: Generates symmetric scaffolds for therapeutic and metal-binding protein design.
• Conditional structure generation: Generates structures with conditional information including motifs and targets.

Scientific Applications:

• Symmetric assembly design: Enables experimental characterization of hundreds of designed symmetric assemblies.
• Metal-binding protein design: Enables design of functional metal-binding proteins.
• Protein binder design for viral targets: Enables design of binders to targets such as influenza haemagglutinin, with accuracy confirmed by cryogenic electron microscopy.
• Therapeutic protein design: Enables symmetric motif scaffolding for therapeutic protein applications.

Methodology:

RFdiffusion fine-tunes the RoseTTAFold structure prediction network on protein structure denoising tasks to obtain a generative model of protein backbones using a diffusion model framework.