Drug repositioning by employing information-driven disease profiles

Abstract

Drug repositioning is the application of known drugs to new diseases. It reduces time, cost and risk com-pared with traditional drug development. Because of its importance, there have been many previous ap-proaches to predict new drug repositioning candidates. In this research, based on different type of infor-mation, we developed two novel approaches to generate disease profiles that are useful to drug repositioning. In PART I, we constructed clinical attribute combination disease profiles by using National Health and Nutrition Examination Survey (NHANES) database and inferred disease associations based on the con-structed disease profiles among 26 diseases. Here, we applied association rule mining technique to generate disease profiles and cosine similarity measure to obtain disease associations quantitatively. We validated that the inferred disease associations have great potentiality to drug repositioning when compared with existing drug-repositioned diseases. We also verified that the results of this study perform better than existing disease networks in terms of drug repositioning. Furthermore, we suggested candidate disease pairs for drug reposi-tioning, such as a pair of gout and heart disease and a pair of cataract and heart disease. In part II, we generated essential gene profile for three cancer types based on connectivity map and na-tional cancer institute (NCI) database and predicted therapeutic targets based on those profiles. The predict-ed therapeutic targets can be used to drug repositioning in efficient ways. Instead of RNA interference ap-proach, which are the most widely used approach identifying essential genes, we predicted essential genes by computing correlations between compound-treated gene expressions and cell viabilities with linear regression model. We verified the results of this study showed better performance to predict therapeutic targets than RNA interference approaches in several cell lines. In addition, we proved that the essential genes by com-pound and RNA interference approaches are complementary to each other in predicting therapeutic targets and proto-oncogenes.