Quantitative Structure-activity Relationship (QSAR) Models for Docking Score Correction.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Protein-drug docking simulations are crucial for predicting binding affinity.
• Existing docking scores often require correction for improved accuracy in drug discovery.
• Quantitative structure-activity relationship (QSAR) models can refine these predictions.

Purpose of the Study:

• To develop and validate structure-based QSAR models for correcting protein-drug docking scores.
• To enhance the accuracy of binding affinity predictions using public affinity data.
• To explore different regression techniques for improved QSAR model performance.

Main Methods:

• Developed structure-based QSAR models using protein-drug docking simulations against approximately 600 proteins.
• Utilized descriptor-based regression, with compound descriptors including docking scores against multiple targets and non-targets.
• Approximated binding free energy using weighted averages of docking scores and applied linear, weighted linear, and polynomial regression models.
• Combined regression models for specific datasets like IC50, Ki, and %inhibition.

Main Results:

• The weighted linear regression model demonstrated higher accuracy compared to simple linear regression.
• Cross-validation confirmed the improved predictive performance of the developed QSAR models.
• QSAR approaches integrating public affinity data effectively refined docking score predictions.

Conclusions:

• Structure-based QSAR models utilizing public affinity data can significantly improve protein-drug docking score accuracy.
• Weighted linear regression offers a more accurate approach for QSAR modeling in this context.
• These refined docking scores hold promise for advancing drug discovery efforts.