RF Diffusion Scaffold Guided Config

Builds the scaffold-guided (fold-conditioning) configuration for RF Diffusion. It connects pre-computed secondary structure and block adjacency tensors for scaffolds (and optionally a target protein) to guide design toward desired topologies. Supports both GCS folder references and inline .pt file payloads, with options for loop masking, residue insertions at loops/termini, and selective scaffold usage.

Usage

Use this node when you want to condition generation on known fold topologies or perform binder design against a target protein with fold constraints. Typical workflow: 1) Load scaffold .pt files (e.g., via a loader node providing SCAFFOLD_PT_FILES), 2) Optionally provide a target PDB and its .pt files for PPI or target fold conditioning, 3) Optionally restrict to a list of scaffold names and tune sampling/loop settings, 4) Connect the output to the 'scaffoldguided_config' input of the RF Diffusion node.

Inputs

Field	Required	Type	Description	Example
scaffold_pt_files	True	SCAFFOLD_PT_FILES	Pre-computed scaffold .pt files containing secondary structure (*_ss.pt) and block adjacency (*_adj.pt) tensors. Supports either a GCS folder reference or inline payloads.	{"type":"gcs_path","gcs_folder_path":""} or {"type":"inline","pt_files":{"scaffoldA_ss.pt":"","scaffoldA_adj.pt":""}}
target_pdb	False	PDB	Optional target protein PDB for conditioned binder (PPI) design. Enables hotspot residue constraints at the RF Diffusion node.	{"filename":"target.pdb","content":""}
target_pt_files	False	SCAFFOLD_PT_FILES	Optional target .pt files for target fold conditioning. Same supported formats as scaffold_pt_files.	{"type":"gcs_path","gcs_folder_path":""} or {"type":"inline","pt_files":{"target_ss.pt":"","target_adj.pt":""}}
scaffold_list	False	STRING	One scaffold name per line to restrict which scaffolds are used. Leave empty to use all provided scaffolds. Lines starting with # are ignored.	scaffold_1 scaffold_2 # comment scaffold_3
sampled_insertion	False	INT	Max residues to insert into loop regions per run. Enables flexible loop lengths while preserving secondary structure.	10
sampled_N	False	INT	Max additional residues to sample at the N-terminus.	5
sampled_C	False	INT	Max additional residues to sample at the C-terminus.	5
ss_mask	False	INT	Residues to mask at the end of each secondary structure block, allowing flexibility at helix/strand boundaries.	2
systematic	False	BOOLEAN	If true, iterate through scaffolds in a fixed order instead of random sampling.	true
mask_loops	False	BOOLEAN	If true, loop regions are masked to encourage flexible redesign. Set false for PPI when loops should be preserved.	false
contig_crop	False	STRING	Crop a specific region from the scaffold using chain and residue range format. Leave empty for full scaffold.	A10-50

Outputs

Field	Type	Description	Example
scaffoldguided_config	RFDIFFUSION_SCAFFOLDGUIDED_CONFIG	Configuration object to drive scaffold-guided/fold-conditioning behavior in RF Diffusion.	{"scaffoldguided":true,"scaffold_secstruc_adj":{"type":"gcs_path","gcs_folder_path":""},"scaffold_list":["scaffold_1","scaffold_2"],"sampled_insertion":10}

Important Notes

• Scaffold .pt inputs support two modes: gcs_path (service loads from cloud storage) and inline (base64-encoded pt files). A legacy inline dict of pt files is also accepted.
• In scaffold-guided mode, sequence contigs can be auto-derived at runtime; you don't need to define contigs for the main generation node.
• If you plan to use hotspot residues in the generation node (for PPI), you must provide a target_pdb here; otherwise hotspot_res will be rejected.
• mask_loops defaults to true for flexible loop redesign. For binder design that must preserve target-facing loops, set mask_loops to false.
• scaffold_list is parsed line-by-line; empty lines and lines starting with # are ignored.
• contig_crop must follow chain/range formatting (e.g., A10-50). Invalid formats will cause validation errors downstream.

Troubleshooting

• Error: 'scaffold_pt_files is required' — Provide a non-empty SCAFFOLD_PT_FILES input, either as a GCS folder reference or inline pt_files.
• Hotspot residues were specified but no target protein is configured — Add a target_pdb here or clear hotspot_res in the generation node.
• Invalid contig_crop format — Use chain and inclusive residue range like 'A10-50'.
• No scaffolds used after providing scaffold_list — Ensure the names match those present in the .pt set and that lines are not commented out.
• Inline mode failing due to payload — Ensure pt_files includes both _ss.pt and _adj.pt entries for each scaffold, base64-encoded.
• GCS path not accessible — Verify the gcs_folder_path and permissions for the GPU service to read the folder.