Identification of Drug-Side Effect Association Via Multi-View Semi-Supervised Sparse Model

The association between drugs and side effects encompasses information about approved medications and their documented adverse drug reactions. Traditional experimental approaches for studying this association tend to be time-consuming and expensive. To represent all drug-side effect associations, a bipartite network framework is employed. Consequently, numerous computational methods have been devised to tackle this problem, focusing on predicting new potential associations. However, a significant gap lies in the neglect of the Multi-View Learning (MVL) algorithm, which has the ability to integrate diverse information sources and enhance prediction accuracy. In our study, we have developed a novel predictor named Multi-View Semi-Supervised Sparse Model (Mv3SM) to address the drug side effect prediction problem. Our approach aims to explore the distinctive characteristics of various view features obtained from fully paired multi-view data and mitigate the influence of noisy data. To test the performance of Mv3SM and other computational approaches, we conducted experiments using three benchmark datasets. The obtained results clearly demonstrate that our proposed method achieves superior predictive performance compared to alternative approaches.