Leveraging nonstructural data to predict structures and affinities of protein-ligand complexes

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Computational methods are crucial for predicting ligand-target interactions in rational drug design.
• Existing methods include physics-based (using 3D target structure) and ligand-based (using experimental data for similar ligands) approaches.
• A gap exists in effectively combining these diverse data sources for improved prediction accuracy.

Purpose of the Study:

• To develop a rigorous statistical framework for integrating physics-based and ligand-based computational methods.
• To create a novel method for predicting ligand pose using readily available information on related ligands.
• To enhance virtual screening of drug candidates by combining different modeling strategies.

Main Methods:

• Developed a statistical framework to combine physics-based and ligand-based computational modeling.
• Created a ligand pose prediction method leveraging known binding ligands without 3D structures.
• Applied the framework to develop an improved virtual screening method for drug candidates.

Main Results:

• The combined modeling approach significantly improves ligand pose prediction accuracy across major drug target families.
• The novel virtual screening method outperforms traditional physics-based and ligand-based methods.
• Demonstrated the efficacy of integrating diverse data sources for enhanced predictive capabilities.

Conclusions:

• Combining physics-based and ligand-based computational methods offers a powerful strategy for improving ligand property prediction.
• The developed framework provides a robust platform for enhancing drug design and discovery pipelines.
• Customized machine learning approaches integrating diverse data show broad potential for future applications in computational chemistry and pharmacology.