A Large-Scale Gene Expression Intensity-Based Similarity Metric for Drug Repositioning

Biological systems often respond to a specific environmental or genetic perturbation without pervasive gene expression changes. Such robustness to perturbations, however, is not reflected on the current computational strategies that utilize gene expression similarity metrics for drug discovery and repositioning. Here we propose a new expression-intensity-based similarity metric that consistently achieved better performance than other state-of-the-art similarity metrics with respect to the gold-standard clustering of drugs with known mechanisms of action. The new metric directly emphasizes the genes exhibiting the greatest changes in expression in response to a perturbation. Using the new framework to systematically compare 3,332 chemical and 3,934 genetic perturbations across 10 cell types representing diverse cellular signatures, we identified thousands of recurrent and cell type-specific connections. We also experimentally validated two drugs identified by the analysis as potential topoisomerase inhibitors. The new framework is a valuable resource for hypothesis generation, functional testing, and drug repositioning.