ISCMF: Integrated similarity-constrained matrix factorization for drug–drug interaction prediction

Drug–drug interaction (DDI) prediction prepares substantial information for drug discovery. As the exact prediction of DDIs can reduce human health risk, the development of an accurate method to solve this problem is quite significant. Despite numerous studies in the field, a considerable number of DDIs are not yet identified. In the current study, we used Integrated Similarity-constrained matrix factorization (ISCMF) to predict DDIs. Eight similarities were calculated based on the drug substructure, targets, side effects, off-label side effects, pathways, transporters, enzymes, and indication data as well as Gaussian interaction profile for the drug pairs. Subsequently, a non-linear similarity fusion method was used to integrate multiple similarities and make them more informative. Finally, we employed ISCMF, which projects drugs in the interaction space into a low-rank space to obtain new insights into DDIs. However, all parts of ISCMF have been proposed in previous studies, but our novelty is applying them in DDI prediction context and combining them. We compared ISCMF with several state-of-the-art methods. The results show that It achieved more appropriate results in five-fold cross-validation. It improves AUPR, and F-measure to 10% and 18%, respectively. For further validation, we performed case studies on numerous interactions predicted by ISCMF with high probability, most of which were validated by reliable databases. Our results provide support for the notion that ISCMF might be used unequivocally as a powerful method for predicting the unknown DDIs. The data and implementation of ISCMF are available at https://github.com/nrohani/ISCMF .