TL;DR
This paper introduces PriorCD, a novel computational method for prioritizing existing drugs for repurposing against cancer. By integrating drug activity profiles with pathway activity derived from mRNA and microRNA expression data, PriorCD constructs a drug functional similarity network to identify promising candidates with higher accuracy than previous gene-expression-based methods.
What Problem Does This Paper Address?
Developing new cancer drugs is a notoriously slow, expensive, and high-risk endeavor, with many candidates failing during clinical trials. Drug repurposing—finding new uses for existing drugs—offers a more efficient and cost-effective alternative. However, effectively identifying which existing drugs might be effective against specific cancers remains a significant challenge. Traditional methods often focus on gene-level expression changes, potentially missing broader pathway-level effects that are crucial for drug efficacy. This paper aims to address this gap by developing a more comprehensive approach that leverages pathway-level drug activity to improve drug repurposing predictions.
Methods
The core of this work is the PriorCD (Prioritization of Candidate Drugs) approach. It involves several key steps:
- Inferring Pathway Activity Profiles: The authors use single sample gene set enrichment analysis (ssGSEA) to derive mRNA pathway activity profiles from mRNA expression data. For microRNA, they first establish microRNA-mRNA interactions and then map these to mRNA pathways to derive microRNA pathway activity profiles.
- Constructing a Drug Functional Similarity Network: They correlate the inferred pathway activity profiles with drug activity profiles (measured by GI50 values) to establish pathway-drug correlations. These correlations are then used to calculate the functional similarity between pairs of drugs, creating a drug-drug functional similarity network. This network integrates both mRNA and microRNA pathway-based similarities.
- Calculating Drug Prioritization Scores: A global network propagation algorithm, specifically the random walk with restart (RWR) algorithm, is applied to the drug functional similarity network. The "restart set" for this algorithm consists of FDA-approved drugs for a specific cancer of interest. Drugs that are more proximal and connected to this restart set in the network receive higher prioritization scores.
- Evaluation: The method's performance is assessed using leave-one-out cross-validation (LOOCV) and receiver-operating characteristic (ROC) curve analysis, with area under the ROC curve (AUROC) as the primary metric. Statistical significance is evaluated through random permutation tests.
Datasets
The study utilizes the following datasets:
- NCI-60 Cancer Cell Lines (CCLs) Data: This panel of 60 human cancer cell lines from nine tissue origins is a crucial resource.
- Chemical Compounds Anticancer Activity Data: Obtained from the CellMiner database, this includes activity data for over 20,000 compounds. After filtering, 3,645 drugs were retained.
- mRNA Expression Data: 19,794 mRNA expression data points from the NCI-60 panel.
- microRNA Expression Data: 319 microRNA expression data points from the NCI-60 panel.
- FDA-Approved Drug Lists: Datasets of FDA-approved drugs for specific cancers (breast cancer and ovarian cancer) were collected from the NCI website to serve as the "restart set" for the prioritization algorithm.
Key Insights and Ideas
- Pathway-Level Similarity is Key: The paper strongly advocates for moving beyond gene-level analysis to pathway-level understanding for drug repurposing. They argue that cellular processes are regulated by complex interactions, and pathway activity provides a more holistic view of drug effects.
- Integrating Multiple Omics: Combining both mRNA and microRNA pathway activities provides a more comprehensive picture of cellular responses and drug interactions, leading to a more robust functional similarity network.
- Network Propagation for Prioritization: The RWR algorithm effectively leverages the constructed drug similarity network to propagate information from known effective drugs to identify potential candidates, reflecting their functional relatedness.
- Cross-Cancer Applicability: The method's performance is evaluated across multiple cancer types, suggesting its potential for broad application in oncology drug repurposing.
Key Results
- High Predictive Performance: PriorCD achieved AUROC values greater than 0.82 in cross-validation tests for breast and ovarian cancer. For six different cancer types tested, AUROC values ranged from 0.82 to 0.97.
- Prioritization of Candidate Drugs: The method identified 14 candidate drugs for breast cancer and 8 for ovarian cancer, with statistically significant prioritization scores (FDR < 0.001).
- Superiority over Gene-Based Methods: When compared to classical gene-based repurposing methods (CMap and Shigemizu et al.'s approach), PriorCD demonstrated superior performance in ROC curve analysis.
- Validation through Literature: The prioritized candidates were further validated through literature searches, confirming their potential therapeutic relevance. For example, gefitinib and afatinib, known EGFR inhibitors for lung cancer, were identified as promising candidates for breast cancer.
Limitations and Future Work
- Completeness of Restart Sets: The study acknowledges that the accuracy of prioritization is dependent on the completeness of the FDA-approved drug lists used as the restart set. As more drugs receive marketing approval, the prioritized lists will become more comprehensive and accurate.
- Redundant Pathways: While the impact of highly overlapping KEGG pathways was found to be modest, the authors explored the method's robustness by removing redundant pathways, showing consistent results.
- Pathway Database Flexibility: The study demonstrated the robustness of the method by using both KEGG and Reactome pathway databases, suggesting its adaptability to different pathway annotations.
- Availability of R Package: To facilitate wider adoption, the authors have developed a freely available R-based package, "PriorCD," which includes the drug functional similarity network. This allows researchers to input their own sets of approved therapeutic drugs for a specific cancer and obtain prioritized candidate lists.
Summary updated