RF Diffusion

Generates protein backbones using RFdiffusion with support for both unconditional generation and conditional motif scaffolding. Allows fine-grained control via symmetry, contig mapping, denoiser, and guiding potentials configurations. Supports full diffusion or partial diffusion (starting from a specified timestep) and can optionally output diffusion trajectories.

Usage

Use this node to design protein structures. For unconditional generation, provide only a length range in contigs (e.g., 100-100) and leave input_pdb empty. For conditional motif scaffolding or partial diffusion, provide an input PDB and a contigs string that specifies fixed and flexible regions. Optionally connect configuration nodes for symmetry, contig map, denoiser, and potentials to refine behavior. Enable write_trajectory if you need intermediate trajectory outputs.

Inputs

Field	Required	Type	Description	Example
checkpoint	True	STRING	Model checkpoint to use. 'Auto' chooses an appropriate checkpoint automatically; other options are advanced and should match your task.	Auto
write_trajectory	True	BOOLEAN	Whether to include trajectory outputs. Disable for faster runs.	True
model_runner	True	STRING	Sampler to run the model with.	SelfConditioning
align_motif	True	BOOLEAN	Align the model’s predicted motif to the input motif during conditional runs.	True
final_step	True	INT	Reverse diffusion will stop at this step (1–50). Higher values speed up inference.	1
contigs	True	STRING	Defines sequence layout. Unconditional: length range only (e.g., 100-100). Conditional: fixed motifs (A10-25), flexible gaps (5-15), and optional chain breaks (/0). For partial diffusion, must match the full layout of the input PDB.	5-15/A10-25/30-40/0 B1-100
hotspot_res	True	STRING	Comma-separated target residues that should contact the designed chain(s). Only for conditional runs.	A30,A33,A34
partial_T	True	INT	Start denoising from this timestep (0–50). 50 disables partial diffusion (standard full diffusion). Requires input_pdb when < 50.	50
seed	True	INT	Base seed for reproducibility.	42
mode	True	STRING	Execution mode. PROD runs the service, MOCK returns predefined data, TEST uses faster minimal parameters.	PROD
timeout	True	INT	Timeout in seconds for the diffusion process (30–12000).	600
input_pdb	False	PDB	Input PDB for conditional tasks (motif scaffolding or partial diffusion). Not used for unconditional generation.
symmetry_config	False	JSON	Symmetry settings. Generate via the RF Diffusion Symmetry Config node and connect here.	{'symmetry': 'c2', 'recenter': True, 'radius': 10, 'model_only_neighbors': False, 'symmetric_self_cond': True}
contigmap_config	False	JSON	Contigmap settings for masking/unmasking sequence or structure. Generate via the RF Diffusion Contigmap Config node.	{'inpaint_seq': 'A50-80', 'inpaint_str': 'B165-178', 'provide_seq': '172-177,200-205', 'length': '100-120'}
denoiser_config	False	JSON	Denoiser noise schedules. Generate via the RF Diffusion Denoiser Config node.	{'noise_scale_ca': 1.0, 'final_noise_scale_ca': 1.0, 'ca_noise_schedule_type': 'constant', 'noise_scale_frame': 1.0, 'final_noise_scale_frame': 1.0, 'frame_noise_schedule_type': 'constant'}
potentials_config	False	JSON	Guiding potentials to steer design (e.g., contacts, ROG, substrate contacts). Generate via the RF Diffusion Potentials Config node.	{'guiding_potentials': ['type:monomer_ROG,weight:1,min_dist:15'], 'guide_scale': 10, 'guide_decay': 'constant', 'olig_inter_all': False, 'olig_intra_all': False}

Outputs

Field	Type	Description	Example
generation.pdb	PDB	Final designed protein structure(s).
trajectory_Xt-1.pdb	PDB	Denoising trajectory at Xt-1 steps. Returned only if write_trajectory is true.
trajectory_pX0.pdb	PDB	Predicted X0 trajectory. Returned only if write_trajectory is true.

Important Notes

• Checkpoint 'Auto' is recommended; manual checkpoints must be compatible with your task.
• Unconditional mode: contigs must specify only a length range (e.g., 100-100), input_pdb must be empty, and hotspot_res must be empty.
• Conditional mode (motif scaffolding): provide input_pdb and contigs must include at least one fixed region (e.g., A10-25).
• Partial diffusion requires input_pdb and partial_T < 50; partial_T = 50 disables partial diffusion (full diffusion).
• For unconditional runs, contigmap_config must not include fields other than length; any other populated field will raise an error.
• write_trajectory increases runtime and output size; disable for faster testing.
• TEST mode adjusts parameters for speed (e.g., increases final_step, disables trajectories/cautious behavior).
• Seeds are applied per output; downstream files may carry seed metadata.
• Symmetry, contig map, denoiser, and potentials are optional but powerful for advanced control.

Troubleshooting

• Error: 'Non empty contigs must be passed' – Ensure contigs is provided. For unconditional, use a simple range; for conditional, include fixed regions.
• Error: 'For partial diffusion sampling, expected input_pdb to be provided' – Provide an input PDB when partial_T < 50.
• Error: 'For unconditional sampling, expected contigs to specify only length range' – Remove chain IDs and fixed regions; use a numeric range only.
• Error: 'For unconditional sampling, expected hotspot_res to be empty' – Clear hotspot_res when no input_pdb is used.
• Error: 'For unconditional sampling, expected to be empty' – In contigmap_config leave all fields empty except length.
• Error: 'For conditional sampling, expected contigs to specify least one fixed region' – Include at least one region like A10-25 when using input_pdb.
• Timeouts: Increase timeout if designs are large or trajectories are enabled.
• Empty or malformed contigs/hotspot formatting: Use correct formats (e.g., A10-25; hotspots like A30,A33).