Enhanced Multi-Head Self-Attention Transformer-Based Method for Prediction of Drug-Drug Interactions

The usage of various drugs has increased significantly in recent years, leading to a higher possibility of drug-drug interactions (DDIs). The concurrent usage of multiple drugs can result in potentially hazardous interactions, making it crucial to foresee DDIs to prevent adverse effects and enhance patient safety. Traditional DDI prediction methods often require extensive examinations, which can be time consuming and resource intensive. As a result, automatic DDI prediction methods have gained attention in the literature, offering clinicians support for making more accurate decisions and designing effective treatment plans. Despite progress in DDI prediction studies, substantial challenges remain in the field. Our study addresses these challenges by proposing a deep learning-based model leveraging drug features. Specifically, this study introduces an enhanced multi-head self-attention transformer-based method, which incorporates pharmacological features to achieve improved performance. The proposed method consists of two primary stages: feature extraction and model design. To evaluate the efficacy of the proposed method, performance evaluation procedures -Accuracy (ACC), Precision (PRE), Recall (REC), and F -Score-are utilized. Comparative experiments are conducted with several state-of-the-art methods on a data set specifically created for this study. Out of all, the proposed method achieves mean values of ACC, PRE, REC, and F-Score as 87.49%, 87.19%, 82.76%, and 84.56 %, respectively, surpassing the performance of other methods. The results unequivocally demonstrate the effectiveness and superiority of the proposed method in predicting DDIs.