Quantitative pathway modeling and analysis in cancer

Biological pathways describe the inter-relationships of genes, proteins, and molecules within a biological system. The study of specific pathways often requires painstaking research into each possible element of the network. Consequently, existing knowledge of gene regulatory networks is limited, resulting in hypotheses that are incomplete or at various levels of refinement. Computational modeling and quantitative analytical approaches seek to fill this information gap.;This dissertation describes the development of QPACA (Quantitative Pathway Analysis in Cancer), a system for pathway visualization and analysis. QPACA addresses three aspects of the general problem: (1) representation and visualization of pathways in the context of biological data, (2) recognition of gene sets that are part of a pathway or coordinated process, and (3) augmentation of pathways by prediction of pathway membership. The pathway representation is designed to be flexible and extensible in order to enable the widest variety of pathway structures and components possible, while the analytical methods directly address the issues inherent in analysis of human systems without making limiting assumptions about the structure of pathways or discretizing data.;QPACA has been used to analyze a number of microarray data sets, employing both yeast and human samples. Four primary results are presented, each of which derives from aspects of QPACA's application to microarray data for analysis or visualization: (1) statistical analysis of differential expression patterns in the context of pathway representations supports the generation of biological hypotheses; (2) gene expression data are sufficient to support computational recognition of hypothesized pathway gene sets across a broad variety of biological processes; (3) gene expression data can be used to produce ranked lists of gene products that are enriched for proteins that interact with members of a predefined pathway; and (4) the surprising ubiquity of detectable signals in expression data that bear on human pathway structure appears to be due to largely non-annotated transcriptional programs present within established pathways.