Methods for comparative analysis of gene co-expression networks
MetadataVis full innførsel
Real-life biological systems consist of ensembles of interconnected components which may affect each other in complex ways. While the study of these components at the individual level provides important insight, a complete understanding of biological behavior necessarily relies on also understanding their interactions at a larger scale. Due to the complexity and quantity of interactions that may be of importance for a given biological system, it is of great importance to develop suitable tools and methods to properly identify them. In this dissertation, I present the work I have performed over the course of my doctoral studies in systems biology, with a focus on two particular fields: differential gene co-expression network analysis, and flux balance analysis. While differential gene co-expression network analysis has already been subject to extensive research, with a multitude of powerful tools already made available for the study of genetic interactions based on expression data, these methods do not attempt to classify interactions according to important qualitative differences between co-expression patterns. In order to address this, we have developed a new method, named the CSD method, which classifies interacting genes according their type of differential co-expression. This dissertation presents the principles and details of this method (Paper 1), as well as its application to two different case studies: one in which we compare two different areas of the human brain (cortex and basal ganglia, Paper 1), and one in which we compare the dynamics of HIV and Mycobacterium tuberculosis co-infection (Paper 3). In both cases, the method identifies groups of genes whose function provides additional insight to the workings of major lethal diseases. On the more general topic of co-expression analysis, we also evaluate the capabilities of the weighted topological overlap measure (wTO) in identifying genetic interactions for low-quality expression data (Paper 2). While wTO is a method with established use in the field, I am not aware of any previous efforts to quantify its advantages over simpler correlation measures. We show that wTO indeed confers some substantial advantages over simple correlation scores, in particular with terms of robustness of results to low sample sizes. Lastly, we have developed a new model for in silico simulations of the behavior of Mycobacterium tuberculosis, with particular focus on improving the representation of iron uptake mechanisms (Paper 4), which proves to be an incremental improvement on pre-existing models. Applying this method with gene expression data from normal and low-iron conditions, we identify potential key drug targets with respect to iron uptake.