Predicting protein-ligand interactions based on bow-pharmacological space and Bayesian additive regression trees

Area of Science:

• Computational chemistry
• Pharmacology
• Machine learning in drug discovery

Background:

• Protein-ligand interactions are vital for drug discovery, target identification, and understanding drug side effects.
• Existing methods for predicting these interactions often rely on limited data subspaces.

Purpose of the Study:

• To develop a novel chemogenomics algorithm for accurate protein-ligand interaction prediction.
• To introduce a new machine learning approach and descriptor class for enhanced predictive modeling.

Main Methods:

• Application of Bayesian Additive Regression Trees (BART) machine learning algorithm.
• Development of a novel proteochemical space, the bow-pharmacological space, integrating protein, ligand, and interaction subspaces.
• Evaluation on four diverse human target datasets: enzymes, nuclear receptors, ion channels, and G-protein-coupled receptors.

Main Results:

• The algorithm achieved high prediction accuracies ranging from 94.5% to 98.4%.
• BART provided reliable probabilistic predictions of protein-ligand interactions.
• The bow-pharmacological space model demonstrated superior scope compared to classical methods.

Conclusions:

• The developed algorithm offers a powerful tool for rapid and accurate prediction of protein-ligand interactions.
• Probabilistic outputs from BART can guide assay prioritization in small molecule development.
• This approach advances chemogenomics and aids in both drug discovery and vigilance phases.