Improving homology models for protein-ligand binding sites

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Drug Discovery

Background:

• Homology modeling is crucial for predicting protein structures.
• Accurate prediction of protein-ligand binding sites is essential for drug discovery.
• Current homology modeling methods can be limited by sequence and structural diversity between target and template proteins.

Purpose of the Study:

• To improve the accuracy of protein-ligand binding site prediction using homology modeling.
• To integrate knowledge of binding residues into the homology modeling framework.
• To develop a sequence-based prediction method for identifying binding residues.

Main Methods:

• Residues were classified as binding or nonbinding using true and predicted labels.
• Sequence-based predictions utilized a support vector machine with a window-based kernel.
• Binding labels guided a sensitive sequence alignment method for target-template alignment.
• Optimal alignment parameters were identified through systematic searching.

Main Results:

• Incorporating a priori knowledge of binding residues significantly improves homology models of protein-ligand binding sites.
• The sequence-based prediction method provided sufficient information to enhance modeling for low sequence identity and high structural diversity target-template pairs.
• The developed framework demonstrated improved accuracy in predicting binding sites.

Conclusions:

• A priori knowledge of binding residues is beneficial for enhancing homology models of protein-ligand binding sites.
• The sequence-based prediction approach is effective, particularly for challenging homology modeling scenarios.
• This method offers a valuable tool for improving protein-ligand interaction predictions in computational drug discovery.