Improving Drug-Target Interaction Prediction Using SVM-SMOTE: A Solution for Imbalanced Dataset

Drug–target interaction (DTI) prediction is a critical step in the drug discovery process, as experimental methods are often time-consuming and expensive. Machine learning techniques have emerged as effective alternatives for this task. However, DTI datasets commonly suffer from severe class imbalance, where the number of true interactions is significantly lower than negative ones—posing a serious challenge for model training. This study proposes an effective framework for DTI prediction. The model utilizes amino acid composition (AAC) and dipeptide composition (DPC) methods to extract protein features, while FP2 molecular fingerprints are used to represent drug features. To address the class imbalance problem, the SVM-SMOTE technique—an SVM-based synthetic minority oversampling method—is employed. For model training, a Linear Support Vector Machine (LSVM) algorithm is used. The proposed model was evaluated against several state-of-the-art methods using benchmark datasets, including Enzyme, GPCR, Ion Channel, and Nuclear Receptor. The results demonstrate that the proposed framework achieves superior performance. Extensive experiments were conducted at various stages of model design, using evaluation metrics such as AUC, accuracy, F1-score, and recall, all of which confirm the effectiveness of the proposed approach.