| Cancer is a leading cause of death worldwide, accounting for over 7.9 million deaths annually. In this dissertation, computational methodologies utilizing whole-genome gene expression data were used to identify preventive and therapeutic opportunities to combat this disease. Three important aspects of cancer progression were studied: host response to carcinogens, early stages of tumorigenesis, and deregulated pathways in the primary tumor. First, host response to cigarette smoke in the bronchial airway of healthy current and never smokers was studied to elucidate what may be key regulatory relationships in the metabolism of carcinogens. MicroRNAs, which are short, non-coding RNAs involved in post-transcriptional gene regulation, were found to be primarily down-regulated in the airway of smokers. By integrating microRNA and mRNA airway expression data, mir-218 was identified as putatively inhibiting the transcription factor MAFG, which is predicted to regulate genes involved in the response to tobacco exposure. Second, early events in the development of lung cancer were studied using computational modeling approaches that incorporate gene expression signatures defined by in vitro perturbation of specific oncogenic pathways. When analyzing the cytologically normal bronchial airway of smokers with lung cancer and high-risk smokers with dysplastic airway lesions, the PI3K pathway had heightened activity throughout the respiratory tract prior to oncogenesis. This has significant implications regarding preventive opportunities, as we found that a chemopreventative agent for lung cancer, myo-inositol, previously shown to cause regression of dysplasia, inhibits PI3K in vitro and in vivo. Finally, personalized treatment of primary breast cancer was explored by training models on gene expression data from in vitro drug response experiments to predict the responsiveness of a tumor to a drug. Ten compounds were studied, and predicted drug responsiveness was significantly linked to survival rates, highlighting their biological/clinical relevance. Some compounds showed synergy with conventional breast cancer subtypes, while others had autonomous patterns of sensitivity. Drug sensitivity predictions were validated in two mouse xenograft models, suggesting the computational methodology is pertinent and accurate. In summary, clinically relevant therapeutic information regarding deregulated pathways can be uncovered in gene expression data and used to improve our understanding of tumorigenesis and guide treatment of cancer. |
