Homology Modeling

Homology Modeling

• a computational technique to predict a protein’s secondary and tertiary structure based on sequence similarity to a known structure (template)

Homologous Proteins

• homology: common evolutionary ancestry

• we infer homology when two sequences share more similarity than expected by chance (excess similarity)

• however homologous protein do not always share significant sequence similarity

• homology is used for structural modeling

Key Assumption

• proteins w/ similar sequences have similar structures

• from structural studies, protein structures is more highly conserved throughout evolution than the protein’s sequence

  • in some protein families, fewer than 5% identical residues are present, while 50% of the related protein’s structure is highly conserved

• below a certain threshold, may or may not be homologous

  • need to have at least 30% sequence identity to find a homolog

  • if we can’t, forget about homolog

Homology Modeling Pipeline

Homology Modeling Pipeline: Template Identifcation

• tools: BLAST, PSI-BLAST

• databases: protein data bank (PDB)

• selection criteria:

  • sequence identity >30% for good results

  • structural resolution of template

Homology Modeling Pipeline: Alignment

• tools: Clustal Omega, MUSCLE, MAFFT

• challenges:

  • correctly aligning gaps and insertions

  • handling low sequence similiarity

• Importance: errors in alignment propagate to the final structure

Homology Modeling Pipeline: Model Building

• rigid body modeling: copying template backbone

  • transplant helices from template (no side chains, just backbone) and put it into the model

• specifically, identifying sequence conserved regions and copying the coordinates of the backbone atoms

Homology Modeling Pipeline: Loop modeling

• loop modeling: predicting missing regions

• identifying sequence variable regions and testing different loop templates/models (database search)

• loop regions may differ b/w homologs

Homology Modeling Pipeline: Side-Chain Modeling + Model Optimization

• side chain placement, trying rotamers (Rosetta)

  • modeling in the side chains; in conserved sequences just take from the template

  • side chains can only adopt 3 rotamers and how they fit into the structure to build the model

  • the first model might have lots of clashes, so then we put it into a force field to minimize clashes (force field automatically separates atoms)

    • goal is to optimize structure

• energy minimization: corrects geometry (e.g bond angles, clashes0

• MD simulations: improves local flexibility and structural accuracy

• Iterative refinement: combines multiple round of modeling and energy minimization

Validation of Predicted Models

• Ramachandran Plot: evaluates backbone dihedral angles

  • want to know if phi, psi bonds are ok

• correct bond lengths and angles

  • model might accidentally push or pull bond angles and interactions, so make sure that the model is consistent w/ chemistry

• energies of models (Force field, DOPE Score used by MODELLER)

• burying of hydrophobic residues and exposure of polar residues

• combining these features: QMEAN in the SWISS-MODEL server