homology-modeling

Homology Modeling — Protein Structure Prediction

MODELLER 10.x · ColabFold · ESMFold · Biopython · pdbfixer · ProDy. For building protein 3D models from sequence when no experimental structure is available.

When to Use This Skill

• No X-ray/CryoEM structure for your target protein (or coverage is partial)
• Building a receptor model for docking or MD when AlphaFold DB lacks your variant/mutant
• Constructing chimeric or engineered proteins not in existing databases
• Validating or improving an AI-predicted structure with experimental template data
• Generating a starting conformation for free-energy calculations (→ force-fields skill)

Decision Tree — Which Method to Use

Target sequence available?
  NO → retrieve from UniProt / NCBI first

Do you have a homologous template (sequence identity > 25%)?
  YES + identity > 50%  → MODELLER (comparative, references/modeller-basics.md)
  YES + identity 25–50% → MODELLER multi-template or AlphaFold2 with template
  NO / < 25%             → AlphaFold2 / ColabFold (references/alphafold-esm.md)

Throughput?
  Single target           → ColabFold interactive / MODELLER script
  Batch (>10 proteins)    → ColabFold batch CLI or ESMFold API
  No MSA / fast screen    → ESMFold (references/alphafold-esm.md)

After modeling:
  → Validate model        → references/structure-quality.md
  → Prepare for MD        → references/structure-prep.md
  → Prepare for docking   → references/structure-prep.md + docking skill

Quick Start

# --- Option A: MODELLER comparative modeling (single template) ---
from modeller import Environ
from modeller.automodel import AutoModel

env = Environ()
env.io.atom_files_directory = ['.', '../templates']

a = AutoModel(env,
              alnfile  = 'alignment.pir',   # PIR format — see modeller-basics.md
              knowns   = '5HT2A_template',  # template code (PDB ID, no extension)
              sequence = 'TARGET_SEQ')      # target sequence ID in .pir file

a.starting_model = 1
a.ending_model   = 5    # generate 5 models, pick best by DOPE score

a.make()

# Select best model
results = [(m.molpdf, m.name) for m in a.outputs
           if m['failure'] is None]
results.sort()
print(f"Best model: {results[0][1]}  DOPE: {results[0][0]:.1f}")