Abstract:
Protein-ligand interactions are fundamental in regulating protein activity within cellular environments, playing a pivotal role in numerous biochemical processes. These interaction studies are important for understanding the mechanisms of biological regulation, and they provide a theoretical basis for the design and discovery of new drug targets. Traditionally, till date, these interactions are uncovered based on simple binding energy calculations. Yet, the lacunae of no functional information arising from these interaction studies remains the biggest challenge. The complexity of these interactions increases when ligands can function as either agonists or antagonists, contingent on specific conditions. Understanding these functional dynamics is crucial for advancing drug discovery and development. This project aims to harness machine learning (ML) techniques to classify agonist and antagonist molecules that relate the biological response of the complexes, based on their mode of action, to their structural features, including hydrophobic, electronic, steric, and topological parameters. The main goal of this study is to make use of limited information to create more general models that cover a wider variety of protein-ligand interactions. This is different from traditional methods that usually focus on specific targets or ligands. In this study, we developed a robust computational tool capable of classifying any proteinligand complex responses as either agonist or antagonist with a test accuracy of 0.87 and precision of 0.91. We also conducted extensive validation of the model on diverse datasets to ensure its accuracy and reliability in predicting protein-ligand interactions. This novel tool demonstrates the ability to interpret and generalize effectively, ensuring precise classification of the mode of action of any specific ligand-protein pair.
