https://github.com/nrbennet/dl_binder_design's ProteinMPNN-FastRelax modification to screening of AF models of binding affinity

[sungyounjoo]

Hello,

I'm a beginner with PyRosetta, and I’ve been working on peptide binder design. I would like to use the "ddg_filter & ddg_hydrophobic_filter" from this GitHub repository. There is a "RosettaFastRelaxUtil.xml" file provided, and I have adapted the following Python script, which seems to work:
RosettaFastRelaxUtil.txt

import os
import glob
import gc # For memory management
from pyrosetta import init, pose_from_pdb
from pyrosetta.rosetta.protocols.rosetta_scripts import XmlObjects
from pyrosetta.rosetta.core.scoring import ScoreFunctionFactory
from pyrosetta.rosetta.protocols.analysis import InterfaceAnalyzerMover

Initialize PyRosetta with the appropriate options

init("-ex1 -ex2aro -corrections::beta_nov16 -mute all")

Load the RosettaScripts XML file

xml_file = "RosettaFastRelaxUtil_original.xml"
xml_objects = XmlObjects.create_from_file(xml_file)

Get the FastRelax mover and ddG filters from XML

fast_relax = xml_objects.get_mover("FastRelax")
ddg_filter = xml_objects.get_filter("ddg")
ddg_hydrophobic_filter = xml_objects.get_filter("ddg_hydrophobic")

Load AlphaFold models (assuming in PDB format)

af_model_dir = "./test"
af_models = glob.glob(os.path.join(af_model_dir, "*.pdb"))

Iterate over each model, apply FastRelax, and calculate ddG

for model in af_models:
print(f"Processing {model}")

# Load the pose (structure)
pose = pose_from_pdb(model)

# Apply FastRelax
print("Applying FastRelax...")
fast_relax.apply(pose)

# Save the relaxed structure
relaxed_model = model.replace(".pdb", "_relaxed.pdb")
pose.dump_pdb(relaxed_model)

# Apply ddG filter (binding affinity calculation)
print("Calculating ddG...")
ddg_passed = ddg_filter.apply(pose)
if ddg_passed:
    # Get and print the ddG value
    ddg_value = ddg_filter.report_sm(pose)
    print(f"{model} passed the general ddG filter with strong binding affinity.")
    print(f"Calculated ddG value: {ddg_value}")
else:
    print(f"{model} did not pass the general ddG filter.")

# Apply hydrophobic ddG filter and print results
print("Calculating hydrophobic ddG...")
ddg_hydrophobic_passed = ddg_hydrophobic_filter.apply(pose)
if ddg_hydrophobic_passed:
    # Get and print the hydrophobic ddG value
    ddg_hydrophobic_value = ddg_hydrophobic_filter.report_sm(pose)
    print(f"{model} passed the hydrophobic ddG filter with strong binding affinity.")
    print(f"Calculated hydrophobic ddG value: {ddg_hydrophobic_value}")
else:
    print(f"{model} did not pass the hydrophobic ddG filter.")

# Clean up memory after each pose processing
del pose  # Delete the pose object to free up memory
gc.collect()  # Call garbage collector explicitly

print("Processing completed.")

What I would like to ask is regarding the validity of this script: does it provide similar results to the original ddG filter from the GitHub repository? (As a beginner, I couldn’t fully follow the details of the original scripts.)

Additionally, I would appreciate your opinion on whether using this script for ddG filtering is appropriate for research purposes.