Computational drug repositioning using collaborative filtering via multi-source fusion

Drug repositioning contributes to a remarkable reduction in time and cost in traditional de novo drug discovery. In this study, we propose a multi-source-based drug repositioning method by using collaborative filtering to discover new indications of drugs. First, we integrate multiple data sources which are drug chemical structures, drug target proteins, and drug-disease associations to extract similarity matrices of drugs and diseases, respectively. Based on different similarity matrices, collaborative filtering is utilized to predict the drug-disease incidence matrix. Then an optimization objective function is designed to adjust the weight of each data source, and informative sources are noticed with the larger weights. Finally, experimental results on benchmark data sets reveal that the proposed algorithm is helpful to improve the prediction performance, by taking Alzheimer's disease and stroke as two examples, it is confirmed that the proposed algorithm can produce credible repositioning drugs in the treatment for these two diseases.