A Systems Biology Approach to Deciphering Oncogenic Mechanisms and Drug Activity in Cancer Cells

With the growing availability of high-throughput biological data, such as gene expression microarrays, integrative computational approaches are increasingly being applied to associate disease with specific genes or biological pathways. The goal of these approaches is to design more accurate diagnostics or "personalized" therapeutic regimens, and to generate a much more fine grained knowledge base of the progression of disease. Many methods have shown limited success at identifying diagnostic and prognostic signatures for specific diseases, such as breast cancer. In a similar context, these analyses have also been used to characterize pharmacological interventions. However, a key drawback to these approaches is that they often look at characteristics of genes in isolation, without accounting for how their activity is mediated by an underlying network of interactions with other molecules. Furthermore, it is often difficult to separate the phenotypic cause versus its downstream effects, a problem referred to as the "driver" and "passenger" question. Therefore, the computational identification of drivers and targets of these perturbations and their key downstream effectors remains an open research problem. In this dissertation, we argue that new insight into disease progression (specifically human cancer) and pharmacological intervention can be derived from a systems biology perspective. By using a comprehensive network of regulatory, physical and signaling interactions within the cell, we can identify interactions that change as a result of a perturbation and work backwards to identify candidate drivers and associated pathways. We term this approach Interactome Dysregulation Enrichment Analysis (IDEA). We first generate a network of gene interactions for human B cells using a Bayesian evidence integration framework - a B Cell Interactome (BCI). We then identify interactions that show aberrant behavior in specific phenotypes using an information-theoretic approach and a large compendium of microarray expression profiles. Finally, we use a number of enrichment approaches to predict candidate genes and pathways from this resulting data. We validated this method using a number of biological scenarios. First, we profiled a number of distinct B Cell Non-Hodgkin Lymphomas, in which there is a specific, known oncogene, and assessed how accurately IDEA predicted the known gene. Second, we analyzed Ramos (Burkitt Lymphoma) cell lines with biochemical perturbations activating the CD40 signaling pathway. Third, we generated hypotheses on a less characterized phenotype Chronic Lymphocytic Leukemia (CLL), from primary tumor samples. Lastly, we discuss ways that a network-based approach can aid in interpretation of high-throughput results. We show that such a systems perspective can be used to characterize disease (specifically Non Hodgkin Lymphomas), biochemical perturbations, and to generate new hypotheses which can be applied to further translational research in Oncology.