Enhancing Drug-Drug Interaction Prediction: An Unified Similarity-Based Neural Network Approach

Drug-drug interactions (DDIs) pose significant challenges in the field of pharmaceutical research and clinical practice. Accurate prediction of potential DDIs is crucial to ensure patient safety and optimize treatment outcomes. In this study, an improved approach to enhance DDI prediction using a neural network framework (named as DDI-USNN) integrated with drug similarity measures has been developed. The proposed approach leverages comprehensive datasets containing drug pairs and their interaction labels to develop a robust predictive model. By incorporating drug similarity into the neural network architecture, the proposed approach captures subtle yet vital patterns underlying DDIs. The performance measures have been evaluated in terms of precision, recall and F-measure. Traditional classifiers commonly used for drug-drug prediction, including support vector machines (SVM), neural network based (NDD), adaptive boosting (AB), and gradient boosting decision tree (GBDT), were considered to compare the performance of proposed method by considering various datasets. Through meticulous experimentation and rigorous cross-validation, the simulation results unequivocally established the superior performance of the proposed integrated similarity-based neural network when compared to conventional methods and standalone neural networks. Notably, the DDI-USNN model achieved a substantial improvement in precision, outperforming AB by 32.7%, GBDT by 26.6%, NDD by 7%, and SVM by 5.5%. Furthermore, the recall values of each method were compared with the proposed DDI-USNN, revealing a marginal decrement of 1.6% in the recall value for DDI-USNN when compared to NDD. Additionally, the mean F-measure for the proposed DDI-USNN was computed and compared with existing methods, further underscoring its effectiveness in drug-drug interaction prediction.