Quantitative methods for linking transcriptional profiles to physiology

Abstract

Underlying the current drive towards a systemic study of biology is the tacit assumption that a quantitative relationship can be obtained between molecular markers and macro- scale physiological measurements, which can be utilized to construct predictive models of cellular behavior. This thesis explores the evidence for such quantitative relations, and then illustrates one approach for the construction of models linking phenotype and molecular measurements. Specifically, this thesis focuses on the analysis of gene expression data as generated through DNA microarrays. Through the application of dimensional reduction methods such as PCA, and interactive pattern exploration, evidence for the existence of quantitative relations between gene expression signatures and physiological markers is presented. Subsequently, a large-scale experiment is designed and conducted to provide data sufficiently rich to support the construction of predictive models. The specific system probed in this experiment is the development of insulin resistance in mice models. A bootstrap-based regression framework is then developed for the construction and evaluation of predictive models linking age of the mice and serum insulin and leptin levels to transcriptional profiles. A regression framework has the advantage of avoiding complicated and detailed assumptions regarding mechanistic behavior of the genes involved. In addition, the genes identified through the modeling often have important biological significance.

(cont.) Further, the framework is flexible, and can be readily adapted to include different sources of data, such as protein expressions and metabolic fluxes. In summary, this thesis validates the construction of predictive, quantitative models linking physiology and molecular markers, and presents in detail one specific approach for the construction of models based on these relationships.