Causality analysis in biological networks

Abstract

Systems biology is a rapidly emerging field, shaped in the last two decades or so, which promises understanding and curing several complex diseases such as cancer. In order to get an insight about the system – specifically the molecular network in the cell – we need to work on following four fundamental aspects: experimental and computational methods to gather knowledge about the system, mathematical models for representing the knowledge, analysis methods for answering questions on the model, and software tools for working on these. In this thesis, we propose new approaches related to all these aspects. In this thesis, we define new terms and concepts that helps us to analyze cellular processes, such as positive and negative paths, upstream and downstream relations, and distance in process graphs. We propose algorithms that will search for functional relations between molecules and will answer several biologically interesting questions related to the network, such as neighborhoods, paths of interest, and common targets or regulators of molecules. In addition, we introduce ChiBE, a pathway editor for visualizing and analyzing BioPAX networks. The tool converts BioPAX graphs to drawable process diagrams and provides the mentioned novel analysis algorithms. Users can query pathways in Pathway Commons database and create sub-networks that focus on specific relations of interest. We also describe a microarray data analysis component, PATIKAmad, built into ChiBE and PATIKAweb, which integrates expression experiment data with networks. PATIKAmad helps those tools to represent experiment values on network elements and to search for causal relations in the network that potentially explain dependent expressions. Causative path search depends on the presence of transcriptional relations in the model, which however is underrepresented in most of the databases. This is mainly due to insufficient knowledge in the literature. We finally propose a method for identifying and classifying modulators of transcription factors, to help complete the missing transcriptional relations in the pathway databases. The method works with large amount of expression data, and looks for evidence of modulation for triplets of genes, i.e. modulator - factor - target. Modulator candidates are chosen among the interacting proteins of transcription factors. We expect to observe that expression of the target gene depends on the interaction between factor and modulator. According to the observed dependency type, we further classify the modulation. When tested, our method finds modulators of Androgen Receptor; our top-scoring result modulators are supported by other evidence in the literature. We also observe that the modulation event and modulation type highly depend on the specific target gene. This finding contradicts with expectations of molecular biology community who often assume a modulator has one type of effect regardless of the target gene.