Protein-ligand binding affinity prediction based on profiles of intermolecular contacts

Area of Science:

• Computational chemistry and cheminformatics
• Structural biology
• Machine learning in drug discovery

Background:

• Structure-based drug design relies on accurate binding affinity prediction (BAP).
• Developing concise, accurate, and interpretable descriptors for BAP remains a challenge.
• Existing methods often lack interpretability or sufficient accuracy.

Purpose of the Study:

• To introduce Intermolecular Contact Profiles (IMCPs) as novel descriptors for machine learning-based BAP.
• To demonstrate the effectiveness and interpretability of IMCPs compared to other descriptors.
• To provide an accessible implementation of IMCPs for the research community.

Main Methods:

• Utilizing IMCPs, which quantify protein-ligand contact groups by count and average distance.
• Integrating IMCPs with classical machine learning models for BAP.
• Validating the performance of IMCP-based models on multiple datasets.

Main Results:

• IMCP-based models achieved superior BAP accuracy compared to models using similar descriptors.
• IMCPs provide a simple, concise, and interpretable feature set for model training.
• The developed descriptors effectively capture structural information of protein-ligand complexes.

Conclusions:

• IMCPs represent a significant advancement in developing accurate and interpretable descriptors for BAP.
• The IMCP approach enhances the efficiency and reliability of structure-based drug design.
• IMCPs are implemented in the BAP Toolkit, facilitating their adoption in drug discovery research.