Drug selection for malignant melanoma using biomarkers generated by weighted gene co-expression network analysis (WGCNA)

Alpsoy, Semih
Chemotherapy is one of the widely applied treatment choices for cancer patients; however, it might not be effective in the majority of patients due to the inability of foreseeing which patients respond to which chemotherapeutic agents. In order to ascertain appropriate chemotherapy for patients, thus, drug biomarkers predicting the response of the patients to chemotherapy should be discovered and translated into clinical practice to decide the ideal chemotherapeutic agents in a patient-centric manner. In this way, it might be possible to tackle cancer disease more effectively, extend the life expectancy of the patients, and economize health expenditures substantially. In addition, discovering drug biomarkers might pave the way for drug target identification, drug discovery process, and eludicating drug mechanism of actions. Because of all these reasons, a systems biology based network approach known as Weighted Gene Co-Expression Network Analysis (WGCNA) is utilized in this study to discover candidate biomarkers for anti-cancer drugs profiled in two large pharmacogenomics studies, the Cancer Cell Line Encyclopedia (CCLE) and the Cancer Genome Project (CGP). In the study, malignant melanoma is selected as a model disease, and only the common anti-cancer drugs between the two pharmacogenomics studies screened against human malignant melanoma cell lines are considered. Both gene expression and drug sensitivity data available in the studies are integrated to identify candidate biomarkers for these common anti-cancer drugs. Next, support vector machine regression (SVR) machine learning algorithm is employed to assess the predictive ability of the identified candidate biomarkers both individually and in combinations. For that purpose, the CCLE expression data of the candidate biomarkers and the CCLE drug sensitivity data are trained in the first step. Predictive ability of these candidate biomarkers is tested in an independent CGP dataset later on. Thereby, in-silico validation of several candidate biomarkers could be accomplished. In conclusion, this thesis shows that the WGCNA methodology is a powerful approach for identifying gene expression-based candidate drug biomarkers for malignant melanoma. The thesis also shows that proper combinations of the candidate biomarkers generated by the WGCNA methodology improve anti-cancer drug sensitivity prediction significantly, and only a few gene combinations are sufficient to predict anti-cancer drug sensitivity powerfully.