Chapter 3: Classical Homology Modeling & Template Refinement

Chapter 3: Classical Homology Modeling & Template Refinement

Before deep learning revolutionized structure prediction, comparative (homology) modeling was the primary methodology for generating 3D models. It rests on an evolutionary principle: protein structural conformation is far more conserved during evolution than primary amino acid sequence.

1. Automated Comparative Modeling with MODELLER

MODELLER generates models using satisfaction of spatial restraints: bond distances, dihedral angles, and non-bonded atom pairs derived from homologous template structures.

from modeller import *
from modeller.automodel import *

# Initialize Modeller environment
env = Environ()
env.io.atom_files_directory = ['.', './templates']

# Define modeling job
class MyModel(AutoModel):
    def select_atoms(self):
        # Select all atoms for spatial restraint satisfaction
        return Selection(self)

a = MyModel(env,
            alnfile='target_template_align.ali', # PIR alignment file
            knowns='2v6sA',                      # Template PDB ID
            sequence='kinase_target')            # Target sequence name

a.starting_model = 1
a.ending_model = 5                              # Generate 5 candidate models
a.md_level = refine.slow                        # Thorough molecular dynamics refinement
a.make()                                        # Compute models and DOPE scores

2. Structural Quality Validation

Models must be validated before functional interpretation:

1. Ramachandran Validation (PROCHECK / MolProbity): >90% of non-glycine residues should reside in core favored regions.
2. DOPE Score (Discrete Optimized Protein Energy): Lower DOPE scores indicate more thermodynamically favorable folds.
3. ERRAT / Verify3D: Assesses atomic non-bonded interaction statistics and 3D-1D compatibility.