DStruBTarget: Integrating Binding Affinity with Structure Similarity for Ligand-Binding Protein Prediction

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Identifying ligand-binding proteins is crucial for drug development.
• Current methods relying solely on ligand structure similarity have limitations in predicting binding quality.
• There is a need for improved computational approaches to accurately identify target proteins for drug candidates.

Purpose of the Study:

• To develop and evaluate a novel computational method, DStruBTarget, for identifying ligand-binding proteins.
• To integrate 2D and 3D ligand structure similarity with binding affinity for enhanced prediction accuracy.
• To assess the performance of DStruBTarget against existing methods and its potential for broader application.

Main Methods:

• Developed DStruBTarget by combining 2D and 3D ligand structure similarity with known ligand-protein binding affinity (BA).
• Evaluated DStruBTarget using 10-fold cross-validation on the DBD dataset (9197 ligands, 1111 proteins).
• Performed independent testing on a SwissTargetPrediction dataset and compared with other methods, including integration with protein BLAST.

Main Results:

• DStruBTarget achieved a 77% hit rate in top 1 prediction on the DBD dataset, outperforming methods using only 2D similarity or 2D+3D similarity.
• On an independent dataset, DStruBTarget reached a 44.02% top 1 hit rate, surpassing SwissTargetPrediction and other comparative methods.
• DStruBTarget with BLAST showed improved performance for ligands binding to multiple proteins, suggesting utility in predicting novel protein targets.

Conclusions:

• DStruBTarget offers a significant advancement in identifying ligand-binding proteins by effectively integrating structural and affinity data.
• The method demonstrates superior performance compared to existing approaches, providing a valuable tool for drug discovery.
• Integrating DStruBTarget with BLAST expands its applicability for predicting novel protein targets, particularly for ligands with diverse binding profiles.