| Identifying different or common transcriptional regulatory interactions between diseased and non diseased cells is an integral step in providing precise diagnostic information that could potentially lead to better therapeutics and personalized medicine. Reverse engineering techniques aim to infer these regulatory interactions from the gene expression measurements contained in large temporal datasets. The aim of this work is to capture the mechanism of action of Interferon Beta (IFNbeta), a drug that has been partially successful in reducing the symptoms observed in subjects inflicted with Multiple Sclerosis (MS). These IFNbeta related interactions are captured, here, by applying novel reverse engineering techniques to a large scale gene expression dataset.; In order to capture valid interactions, prior to being analyzed, dataset dimensions must be greatly reduced from thousands of individual gene profiles, to a smaller group of biologically relevant variables. In this thesis, an unsupervised and a supervised dimension reduction technique that work in a coordinated manner are introduced. Initially, the unsupervised scheme removes dispensable genes based on their expression profiles and forms a smaller set of representative metagenes. The supervised scheme then employs a priori biological information to identify a small set of metagenes to be reverse engineered.; In order to infer gene regulatory interactions and validate the results, also introduced in this work is a comprehensive neural network based modeling and validation framework. Two reverse engineering techniques, Gene Set Stochastic Sampling and Sensitivity Analysis, both aimed at identifying minimal gene sets that are predictive of target gene expression profiles are developed and validated through their application to an artificial dataset as well as a previously studied dataset. Following this, a thorough computational methodology to test the validity, causality and robustness of the inferred regulations, through novel assemblies of relevant testing datasets, is introduced.; Finally, the methods discussed above, are applied to a large gene expression dataset developed to identify the effects of IFNbeta in subjects inflicted with MS. The application of the unsupervised and supervised dimension reduction techniques to this dataset are effective in identifying a set of Interferon Beta related variables to be reverse engineered with the newly developed inference techniques. The subsequent application of these modeling approaches results in a set of inferred interactions that follow current knowledge and seem to summarize the mechanism of action of Interferon Beta. |
