Predicted and Explained: Transforming drug discovery with AI for high-precision receptor-ligand interaction modeling...

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Pharmaceutical drug development faces challenges in predicting receptor-ligand interactions, impacting efficacy and safety.
• Traditional methods for predicting these interactions are often slow and costly.
• Efficient computational approaches are crucial for accelerating the drug discovery pipeline.

Purpose of the Study:

• To develop and validate a machine learning framework for accurate prediction of docking scores.
• To integrate diverse molecular representations for enhanced predictive performance.
• To improve the interpretability of predictions and understand binding dynamics.

Main Methods:

• Developed a machine learning framework integrating Lipinski descriptors, fingerprints, and graph-based molecular representations.
• Implemented and compared early fusion (feature-level) and late fusion (decision-level) strategies.
• Applied Local Interpretable Model-agnostic Explanations (LIME) for feature importance analysis.

Main Results:

• The early fusion model demonstrated superior predictive accuracy and robustness compared to other methods.
• Identified critical physicochemical and structural features influencing docking scores through LIME.
• Validated the framework's reliability and biological plausibility using bioinformatics tools and 3D visualizations.

Conclusions:

• Integrating multi-scale molecular representations with machine learning enhances prediction of receptor-ligand interactions.
• The developed framework accelerates therapeutic development by enabling data-driven prioritization of drug candidates.
• The approach provides valuable insights into ligand-receptor binding mechanisms, aiding in the design of effective treatments for complex diseases.