rfdiffusion-viz

name	rfdiffusion-viz
description	Use when inspecting RFdiffusion/RFdiffusion2/RFdiffusion3 outputs, viewing diffusion trajectories, tiling scaffold candidates, or performing visual QC on generated backbones through PyMOL.
version	0.1.0

RFdiffusion Output Visualization

Workflows for inspecting and QC'ing RFdiffusion backbone outputs in PyMOL. Composes with @rfd3 (config building) and @proteinmpnn-viz (sequence design step).

Send all cmd.* code via: ~/.pymol-agent-bridge/bin/pymol-agent-bridge exec "..." (or heredoc for multi-line). See @pymol-fundamentals for details.

Loading Outputs

Single Design

cmd.load("/path/to/output/design_0.pdb", "design_0")
cmd.remove("resn HOH+WAT")
cmd.show("cartoon")
cmd.orient()

Batch Loading

import glob, os
output_dir = "/path/to/rfdiffusion/output"
for f in sorted(glob.glob(os.path.join(output_dir, "*.pdb"))):
    name = os.path.splitext(os.path.basename(f))[0]
    cmd.load(f, name)

Trajectory Visualization

RFdiffusion outputs trajectory files in a /traj/ subdirectory as multi-state PDBs showing the diffusion process from noise to structure.

Load and Play Trajectory

cmd.load("/path/to/output/traj/design_0_pX0_traj.pdb", "trajectory")
cmd.show("cartoon", "trajectory")
cmd.spectrum("count", "blue_white_red", "trajectory")

# Play through states
cmd.mset("1 -%d" % cmd.count_states("trajectory"))
cmd.mplay()

View Specific Steps

# First frame (noise)
cmd.frame(1)

# Last frame (final structure)
cmd.frame(cmd.count_states("trajectory"))

# Midpoint
n = cmd.count_states("trajectory")
cmd.frame(n // 2)

Grid Tiling for Batch QC

Tile all designs in a grid for rapid visual screening.

# Load all outputs first, then:
cmd.set("grid_mode", 1)
cmd.show("cartoon")

# Optional: color each by chain
for obj in cmd.get_object_list():
    cmd.do("util.cbc " + obj)

Exit Grid Mode

cmd.set("grid_mode", 0)

Contig Region Coloring

Color designed vs. fixed regions using the standard design color scheme (see @rfd3 for full scheme).

# Color scheme: cyan=fixed/motif, magenta=designed, yellow=ligand
# Assuming the contig was: [5-25/A10-30/30-50]
# Residues 1-25 are designed, 26-46 are from motif (A10-30), 47-71 are designed

cmd.color("magenta", "design_0 and resi 1-25")   # designed
cmd.color("cyan", "design_0 and resi 26-46")      # fixed motif
cmd.color("magenta", "design_0 and resi 47-71")   # designed

cmd.show("sticks", "design_0 and resi 26-46")     # show motif as sticks

Automated Region Identification

# If you know the motif residues from the input PDB:
cmd.select("motif_region", "design_0 and resi 26-46")
cmd.select("designed_region", "design_0 and not motif_region")

cmd.color("cyan", "motif_region")
cmd.color("magenta", "designed_region")
cmd.show("sticks", "motif_region")
cmd.show("cartoon", "designed_region")

Self-Consistency Check

Align the RFdiffusion output to the AlphaFold2 prediction of the designed sequence to validate designability. See @alphafold-validation for the AF2 side.

# Load both
cmd.load("rfdiffusion_output.pdb", "design")
cmd.load("af2_prediction.pdb", "af2_pred")

# Align on CA atoms
rms = cmd.align("af2_pred and name CA", "design and name CA")
print("Self-consistency RMSD: %.2f A" % rms[0])

# Color by deviation
cmd.color("green", "af2_pred")
cmd.color("cyan", "design")

Pass/Fail Criteria

Metric	Pass	Marginal	Fail
CA-RMSD	< 1.5 A	1.5-2.5 A	> 2.5 A
pLDDT (AF2)	> 80	60-80	< 60

Motif RMSD (Scaffolding QC)

For motif scaffolding designs, verify the motif geometry is preserved.

# Align only the motif residues
# Suppose motif is residues 26-46 in the design, 10-30 in the original
cmd.align("design and resi 26-46 and name CA", "original and chain A and resi 10-30 and name CA")

# Per-atom RMSD for motif
rms = cmd.rms_cur("design and resi 26-46 and name CA", "original and chain A and resi 10-30 and name CA")
print("Motif RMSD: %.2f A" % rms)

Motif Fidelity Criteria

RMSD	Interpretation
< 0.5 A	Excellent motif preservation
0.5-1.0 A	Good — minor backbone shifts
1.0-2.0 A	Marginal — check contacts
> 2.0 A	Poor — motif geometry distorted

Binder Design Outputs

For binder design runs (target + designed binder), visualize the complex.

cmd.load("binder_design.pdb", "complex")

# Identify chains
chains = cmd.get_chains("complex")
# Typically: target=chain A, binder=chain B

cmd.color("gray80", "chain A")       # target
cmd.color("marine", "chain B")       # designed binder
cmd.show("surface", "chain A")
cmd.set("transparency", 0.7, "chain A")
cmd.show("cartoon", "chain B")

Hotspot Contacts

# Show designed binder residues contacting the target
cmd.select("interface_binder", "byres (chain B within 4 of chain A)")
cmd.select("interface_target", "byres (chain A within 4 of chain B)")
cmd.show("sticks", "interface_binder")
cmd.show("sticks", "interface_target")

# H-bonds across the interface
cmd.distance("interface_hbonds", "chain A", "chain B", mode=2)
cmd.set("dash_color", "yellow", "interface_hbonds")

Symmetric Assembly Outputs

cmd.load("symmetric_design.pdb", "assembly")
cmd.do("util.cbc assembly")  # color by chain
cmd.show("cartoon")

# Show inter-subunit contacts
chains = cmd.get_chains("assembly")
if len(chains) >= 2:
    cmd.select("interface_AB", "byres (chain %s within 4 of chain %s)" % (chains[0], chains[1]))
    cmd.show("sticks", "interface_AB")

Tips

• Always remove waters before inspecting: cmd.remove("resn HOH+WAT")
• Grid mode (grid_mode=1) is essential for screening >5 designs
• Trajectory files can be large — load one at a time for smooth playback
• Self-consistency RMSD < 1.5 A is the standard designability threshold
• Color scheme matches @rfd3: cyan=fixed, magenta=designed, yellow=ligand