M3S-GRPred: a novel ensemble learning approach for the interpretable prediction of glucocorticoid receptor antagonists using a multi-step stacking strategy

Accelerating the discovery of drugs for glucocorticoid receptor (GR)-related disorders, particularly through the use of innovative machine learning (ML)-based methods, offers significant potential for advancing therapeutic development, improving treatment effectiveness, and reducing negative side effects. While laboratory experiments can accurately pinpoint GR antagonists, their cost-effectiveness for large-scale drug discovery is often limited. Consequently, computational methods that utilize SMILES information for precise in silico identification of GR antagonists are essential, enabling drug discovery processes that are both efficient and scalable.

In this study, a novel ensemble learning method employing a multi-step stacking strategy (M3S), named M3S-GRPred, was developed with the goal of rapidly and accurately identifying new GR antagonists. To the best of our knowledge, M3S-GRPred represents the first SMILES-based prediction tool specifically designed to identify GR antagonists without relying on three-dimensional structural information. In the development of M3S-GRPred, different balanced subsets of data were initially created using an under-sampling technique. These balanced subsets were then used to explore and evaluate a diverse range of base-classifiers.

These classifiers were trained using various SMILES-based feature descriptors in combination with widely used ML algorithms. Finally, M3S-GRPred was constructed by integrating probabilistic features from a selection of these base-classifiers, a selection process guided by a two-step feature selection technique. Comparative experiments conducted in this study demonstrate that M3S-GRPred is capable of precisely identifying GR antagonists and effectively handling the challenges posed by imbalanced datasets. When compared to traditional ML classifiers, M3S-GRPred achieved superior performance on both the training data and an independent test dataset.

Furthermore, M3S-GRPred was applied to screen FDA-approved drugs for potential GR antagonist activity. Promising candidates identified through this screening were then further validated using molecular docking techniques, followed by detailed molecular dynamics (MD) simulation studies to explore their potential for drug repurposing in the context of Cushing’s syndrome. It is anticipated that M3S-GRPred will serve as a valuable and efficient screening tool for discovering novel GR antagonists from extensive collections of uncharacterized compounds in a cost-effective manner.