Identifying and modeling perturbed networks in cancer through statistical and constraint-based analysis

In my Ph.D. research, I have focused largely on modeling biological networks---using both statistical and mechanistic modeling approaches---with applications to medicine, energy, and genomic science.;The more coarse-grained of these approaches use statistical analysis of high-throughput expression data to identify molecular signatures of disease phenotypes; such signatures are indicative of aberrant function of genes or pathways. As a primary focus of my Ph.D. work, I have generated and applied tools to harness high-throughput transcriptomic data to identify molecular signatures---particularly a special class of signatures that quantifies the relative expression levels among multiple genes---and to help elucidate underlying mechanism of complex biological processes and properties. I have used both gene- and network-based signatures to pioneer approaches for predicting clinical phenotypes and outcomes of patient samples. Most notably, I led the development and implementation of a computational method for identifying biomolecular pathways (e.g., metabolic or signaling) that are perturbed in disease data. I used this method (described in Chapter 4) to identify differentially regulated and variably expressed pathways in a number of human diseases, and later investigated behavior-associated modules in a transcriptional regulatory network of the honeybee brain.;As a more functionally grounded complement to statistical approaches, stoichiometric models of biochemical reaction networks can be used to simulate disease systems in mechanistic detail. I have developed an extensive background in constraint-based analysis of biochemical reaction networks, beginning with my contribution to the genome-scale reconstruction of the pathogen Leishmania major, while working as an undergraduate researcher. Building off of the metabolic and modeling expertise I gained through this project---as well as the construction of major signaling pathways in yeast---I was able to help substantially in the metabolic reconstruction of the butanol producing bacteria Clostridium beijerinckii. This model is now being used to identify genetic modifications that may lead to increased butanol production for industrial bioenergy applications. Most recently, I have helped to pioneer and automate novel network reconstruction and modeling approaches in human tissues and cell types. In particular, I have applied these methods to reconstruct and validate the first genome-scale metabolic model for the deadly brain cancer, glioblastoma multiforme. This model and others like it will be transformative for mapping the genome-to-phenotype relationship in complex diseases to elucidate underlying mechanisms and pave the way for new therapeutic strategies. Moreover, the reconstruction and analysis of this model represents a culmination of the array of mechanistic and statistical approaches used and described over the course of my Ph.D. Perturbed metabolic pathways identified by statistical tools represent robust differences between healthy and diseased states, and in turn, serve as focal points for ongoing model simulation and development. (Abstract shortened by UMI.).