Using Protein Homology Models for Structure-Based Studies: Approaches to Model Refinement

Visvaldas Kairys, Michael K. Gilson, Miguel X. FernandesThe Scientific World JOURNAL 2006 · DOI

TL;DR

This review explores how computational models of protein structures, built using "homology modeling" (when experimental data is scarce), can be effectively refined and utilized for drug discovery. It highlights that while these models aren't perfect, various refinement strategies exist, and their success is highly dependent on the specific protein system, with more complex methods not always yielding superior results.

What Problem Does This Paper Address?

The vast explosion of protein sequence data from genomics has outpaced our ability to experimentally determine their 3D structures. This creates a significant bottleneck for structure-based drug design, which relies on knowing a protein's shape to design molecules that can interact with it. Homology modeling offers a solution by building a 3D model of a "target" protein using the known structure of a similar "template" protein. However, these initial models can contain inaccuracies that hinder their use in drug discovery. This paper addresses the challenge of refining these homology models to make them more reliable for identifying potential drug candidates.

Methods

The paper reviews various approaches to refining homology models, categorized as follows:

  • No further optimization or simple energy minimization: This is the most basic approach, where the initial model is used directly or subjected to a quick energy minimization to resolve clashes and improve basic geometry.
  • Using an ensemble of protein structures: Instead of a single model, multiple variations of the homology model are generated and used for docking. This can capture some of the inherent flexibility of the protein.
  • Ligand-directed optimization of the binding pocket: This approach uses information about known ligands that bind to the target protein to guide the refinement of the binding site. This can involve incorporating mutagenesis data or NMR interaction signals.
  • Full optimization using Molecular Dynamics (MD) or related procedures: This is the most advanced method, where the protein and its complex with a ligand are simulated over time using MD. This allows for more realistic exploration of conformational changes and interactions.

The paper also discusses methods for validating these models, primarily through enrichment analysis (how well the models can identify known binders from a larger set of molecules) and structural analysis (comparing the refined models to experimentally determined structures).

Datasets

Not discussed in the paper. The paper focuses on reviewing existing studies and methodologies rather than presenting new experimental data or benchmarks.

Key Insights and Ideas

  • Homology models are surprisingly useful for drug discovery: Despite their inherent inaccuracies, homology models can significantly enrich virtual screening results, meaning they can effectively identify known drug binders from large compound libraries.
  • Refinement is crucial, but not always complex: While advanced methods like MD can be powerful, simpler approaches can also be effective, especially when the sequence identity between the target and template is high. The choice of refinement strategy is highly system-dependent.
  • Template choice matters: The quality of the initial homology model, and thus its usefulness, is heavily influenced by the choice of the template protein.
  • Ligand information is gold: Incorporating knowledge about known ligands (e.g., from mutagenesis studies or existing binding data) can greatly improve the accuracy and reliability of homology models for drug design.
  • Flexibility is key: Proteins are not rigid structures. Methods that account for protein flexibility, such as using ensembles of structures or MD simulations, tend to yield better results.

Key Results

The paper presents qualitative and quantitative findings from various studies:

  • Enrichment: Studies show that homology models can achieve enrichment factors comparable to or even better than experimental structures in some cases. For example, one study found that when sequence identity was >50%, homology models identified active compounds ~5 times more than random selection.
  • Structural Accuracy: RMSD values between homology models and experimental structures are reported, often in the range of 1-3 Å for backbone atoms, with binding pocket RMSDs being even lower (e.g., 0.18 Å).
  • Correlation: Several studies demonstrate good correlations (e.g., R=0.69 to 0.83) between predicted binding affinities from homology models and experimental data or actual protein structures.
  • Drug Discovery Success: Numerous examples are cited where homology models, often with refinement, have led to the discovery of novel inhibitors and agonists for various therapeutic targets.

Limitations and Future Work

  • Limited Template Availability: The primary limitation is the ongoing scarcity of experimentally determined 3D structures, which restricts the availability of suitable templates for homology modeling, especially for proteins with low sequence identity to known structures.
  • System Dependence: The effectiveness of any refinement method is highly dependent on the specific protein system being studied. There's no one-size-fits-all solution.
  • Future Directions: The authors anticipate a future where protein folding, ligand docking, and homology model refinement (including MD) are seamlessly integrated into automated, efficient procedures. They also foresee the increasing incorporation of data from mutagenesis, ligand binding, thermodynamics, and functional studies to supplement structural data and build reliable models for proteins with very low sequence identity.

Summary updated

Abstract

Homology modeling is a computational methodology to assign a 3-D structure to a target protein when experimental data are not available. The methodology uses another protein with a known structure that shares some sequence identity with the target as a template. The crudest approach is to thread the target protein backbone atoms over the backbone atoms of the template protein, but necessary refinement methods are needed to produce realistic models. In this mini-review anchored within the scope of drug design, we show the validity of using homology models of proteins in the discovery of binders for potential therapeutic targets. We also report several different approaches to homology model refinement, going from very simple to the most elaborate. Results show that refinement approaches are system dependent and that more elaborate methodologies do not always correlate with better performances from built homology models.
BibTeX
@article{kairys2006using,
  author = {Visvaldas Kairys and Michael K. Gilson and Miguel X. Fernandes},
  title = {{Using Protein Homology Models for Structure-Based Studies: Approaches to Model Refinement}},
  journal = {The Scientific World JOURNAL},
  year = {2006},
  doi = {10.1100/tsw.2006.250},
}
Figure