Skip to content

RF Diffusion

Runs inference of a RF Diffusion model (conditional or unconditional) for protein structure generation. RF Diffusion

Quick Start

  1. Select the appropriate checkpoint or use "Auto" for automatic selection.
  2. Specify the generation mode by setting contigs (length range for unconditional, motif/gap layout for conditional).
  3. (Optional) Provide an input PDB for conditional generation.
  4. Adjust other parameters as needed (e.g., write_trajectory, final_step).
  5. Run the node to generate protein structures.

Setup Guide

1. Unconditional Generation

  1. Set contigs to a length range (e.g., 100-100).
  2. Leave input_pdb empty.
  3. Configure other parameters as desired.

2. Conditional Generation (Motif Scaffolding)

  1. Set contigs to specify fixed motifs and flexible gaps (e.g., 5-15/A10-25/30-40/0 B1-100).
  2. Provide an input_pdb containing the motif.
  3. (Optional) Connect additional configuration nodes for symmetry, contigmap, denoiser, or potentials.

Basic Usage

Unconditional Generation

  • Generate novel protein backbones of specified length.
  • Use for de novo protein design.

Motif Scaffolding

  • Scaffold a fixed motif within a generated backbone.
  • Specify complex layouts with flexible regions and chain breaks.

Configuration

Required Inputs

Field Description Type Example
checkpoint Checkpoint name or "Auto" for automatic selection. COMBO Auto
write_trajectory Whether to include the protein sample trajectory. BOOLEAN True
model_runner Sampler name. COMBO SelfConditioning
align_motif Align the model's prediction of the motif to the input motif. BOOLEAN True
final_step Step to stop reverse diffusion (higher = faster, less accurate). INT 1
contigs Sequence layout: length range (unconditional) or motif/gap layout (conditional). STRING 100-100 or 5-15/A10-25
hotspot_res Residues on the target protein to involve in inference (comma-separated). STRING A30,A33,A34
seed Base seed for randomness. INT 42
mode Node execution mode: MOCK, PROD, or TEST. COMBO PROD

Optional Inputs

Field Description Type Example
input_pdb Initial PDB to start generation from (for conditional generation). PDB
symmetry_config Additional symmetry configuration (from RFDiffusionSymmetryConfig node). JSON
contigmap_config Additional contigmap configuration (from RFDiffusionContigmapConfig node). JSON
denoiser_config Additional denoiser configuration (from RFDiffusionDenoiserConfig node). JSON
potentials_config Additional potentials configuration (from RFDiffusionPotentialsConfig node). JSON

Outputs

Field Description Type Example
generation.pdb Generation result. PDB
trajectory_Xt-1.pdb Xt-1 trajectory PDB (if write_trajectory is True). PDB
trajectory_pX0.pdb pX0 trajectory PDB (if write_trajectory is True). PDB

Best Practices

Input Preparation

  • For unconditional generation, specify only a length range in contigs and leave input_pdb empty.
  • For motif scaffolding, provide both a motif-containing input_pdb and a detailed contigs string.

Advanced Configuration

  • Use configuration nodes for symmetry, contigmap, denoiser, or potentials to fine-tune generation.
  • Start with MOCK or TEST mode to validate setup before running full PROD jobs.

Troubleshooting

Common Issues

  • Error: Non-empty contigs must be passed: Ensure you specify a valid length range or motif/gap layout.
  • Error: For unconditional sampling, expected contigs to specify only length range: Remove motifs or chain breaks from contigs.
  • Error: For conditional sampling, expected contigs to specify at least one fixed region: Add a motif region (e.g., A10-25) to contigs.

Need Help?