Incorporating Explicit Water Molecules and Ligand Conformation Stability in Machine-Learning Scoring Functions

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• Structure-based drug design relies on accurate molecular docking scoring functions.
• Machine learning (ML) offers potential for advancing these scoring functions.
• Current ML scoring functions require further improvement in robustness and applicability.

Purpose of the Study:

• To develop a more robust and broadly applicable machine learning-based scoring function for molecular docking.
• To enhance prediction accuracy by incorporating novel features and advanced ML algorithms.

Main Methods:

• Utilized extreme gradient boosting (XGBoost) with Δ-Vina parametrization.
• Expanded the training dataset for improved model generalization.
• Incorporated new features, including explicit mediating water molecules and ligand conformation stability.

Main Results:

• Developed the novel scoring function Δ_vinaXGB.
• Δ_vinaXGB demonstrated consistent top-tier performance against classical scoring functions on the CASF-2016 benchmark.
• Achieved significantly improved prediction accuracy in diverse structural contexts simulating real-world docking scenarios.

Conclusions:

• The Δ_vinaXGB scoring function represents a significant advancement in machine learning for molecular docking.
• This improved scoring function enhances the reliability of structure-based drug design.
• The approach offers a more accurate tool for predicting drug-target interactions.