Drug-Target Interactions Prediction Using Stacking Ensemble Learning Approach

The process of drug discovery, particularly in the domain of drug-target interactions (DTI), is often time-consuming and costly, requiring extensive experimentation and validation before global approval can be obtained. To streamline this process and reduce associated costs, computational methods such as machine learning and deep learning are increasingly employed. However, these approaches often face challenges, including the need for large datasets, significant computational resources, and a tendency towards overfitting. Addressing these limitations, this study explores the use of Stacking Ensemble Learning (SEL) as a promising solution for DTI prediction. The proposed SEL model integrates multiple base learners, including Adaptive Boosting, Gradient Boosting, K-Nearest Neighbor, Random Forest, and Support Vector Machine, and enhances their performance through careful parameter tuning. To address data imbalances, the Synthetic Minority Oversampling Technique (SMOTE) is employed, ensuring more reliable predictions. The model was rigorously evaluated and demonstrated remarkable efficacy, achieving an accuracy of 99.047%. This research underscores the potential of the SEL approach in advancing computational drug discovery by enhancing prediction accuracy and robustness, offering a viable pathway for more efficient and cost-effective drug development processes.