Prioritization of candidate cancer drugs based on a drug functional similarity network constructed by integrating pathway activities and drug activities

Abstract

Due to the speed, efficiency, relative risk, and lower costs compared to traditional drug discovery, the prioritization of candidate drugs for repurposing against cancers of interest has attracted the attention of experts in recent years. Herein, we present a powerful computational approach, termed prioritization of candidate drugs (PriorCD), for the prioritization of candidate cancer drugs based on a global network propagation algorithm and a drug-drug functional similarity network constructed by integrating pathway activity profiles and drug activity profiles. This provides a new approach to drug repurposing by first considering the drug functional similarities at the pathway level. The performance of PriorCD in drug repurposing was evaluated by using drug datasets of breast cancer and ovarian cancer. Cross-validation tests on the drugs approved for the treatment of these cancers indicated that our approach can achieve area under receiver-operating characteristic curve (AUROC) values greater than 0.82. Furthermore, literature searches validated our results, and comparison with other classical gene-based repurposing methods indicated that our pathway-level PriorCD is comparatively more effective at prioritizing candidate drugs with similar therapeutic effects. We hope that our study will be of benefit to the field of drug discovery. In order to expand the usage of PriorCD, a freely available R-based package, PriorCD, has been developed to prioritize candidate anticancer drugs for drug repurposing.

Keywords: drug activities; drug functional similarity network; drug repurposing; pathway activities.

Figure 1

Workflow of PriorCD. (A) Data preparation. Drug–disease relationships were collected from the FDA; mRNA and microRNA expression data and drug activity profiles in NCI‐60 cell lines were obtained from CellMiner. (B) Both mRNA and microRNA expression data were enriched into mRNA and microRNA pathway activity profiles, respectively, and then correlated with drug activity profiles to calculate mRNA‐ and microRNA‐based pathway–drug correlations across NCI‐60 cell lines. Based on these correlations, the functional similarity between each pair of drugs was calculated, and a drug‐drug functional similarity network was then generated. Through mapping of known cancer therapeutic drugs to the network, a global network propagation algorithm was subsequently applied to the network to achieve a prioritized list of drugs, which was validated by ROC curve analysis.

Figure 2

Cell–cell and tissue‐of‐origin correlation. Pearson correlation coefficient (PCC) of 227 mRNA and 124 microRNA pathway activity profiles, respectively, presented at the levels of NCI‐60 cell line and tissue of origin. (A) Heatmap of cell–cell correlation coefficient for mRNA pathway. (B) Mean tissue of origin correlation coefficient for mRNA pathway. (C) Heatmap of cell–cell correlation coefficient for microRNA pathway. (D) Mean tissue of origin correlation coefficient for microRNA pathway.

Figure 3

Clustered image of (A) 227 mRNA and (B) 124 microRNA pathway activity levels in NCI‐60 cell lines, where red indicates high activity level and blue indicates low activity level.

Figure 4

Chemical structures of camptothecin and its derivatives. (A) Camptothecin (NSC94600). (B) Topotecan (NSC609699), FDA‐approved drug for ovarian cancer. (C) Camptothecin derivative NSC629971. (D) Camptothecin derivative NSC610457. (E) Camptothecin derivative NSC681644. Structures in red represent their common structure.

Figure 5

Cross‐validation and comparison results. (A) ROC curves for 6 different cancer drug sets were generated. The AUROC values for each cancer drug set were calculated and were displayed in the brackets, respectively. (B) Comparison between PriorCD and two other methods. We applied PriorCD on three different drug datasets to compare its performance with Lamb et al. and Shigemizu et al. The TPR and FPR were calculated, and then, AUC values behind the color bar were used to measure their performance. UCDB: drugs that can down‐regulate up‐regulated cancer genes. DCUB: drugs that can up‐regulate down‐regulated cancer genes.