An Integrative Computational Framework for Defining Asthma Endotypes

The rapid pace of drug development in recent years has led to the recognition that new pharmacotherapies do not have the same effect on all patients. This is particularly true in the case of complex common diseases such as hypertension, diabetes and asthma, where a diversity of pathogenetic factors may interact to produce the same disease, resulting in a large degree of heterogeneity in the response to medical therapy. For this reason, the ability to differentiate between different disease endotypes is of increasing importance to clinical medicine.\ud \ud In the case of asthma, initial studies have hinted at the presence of multiple disease endotypes with different clinical characteristics. Additional studies have identified novel genetic risk factors and differences in gene expression among asthmatic patients with different disease endotypes. Despite the presence of large-scale clinical and molecular datasets from asthmatic patients, limited efforts have been made to integrate these different formats to develop a systems-level understanding of disease mechanism.\ud \ud In this thesis, we develop a computational framework for addressing the problem of disease heterogeneity by integrating data from multiple sources, including the genome, phenome and transcriptome in order to define clinically-relevant disease subtypes, and we demonstrate its application in a cohort of asthmatic children. First we perform a cluster analysis of clinical phenotypic data and detect the presence of multiple disease endotypes in a cohort of children with mild-to-moderate asthma. We evaluate the clinical significance of these endotypes by demonstrating their longtudinal stability and association with differential response to pharmacotherapy. Next, we develop a transcriptional network from the gene expression profiles of these patients and identify the relationship between discrete patterns of expression and asthma endotypes. Finally, we combine longitudinally-derived clinical phenotypes with genetic data to uncover novel genetic associations corresponding to changes in gene expression and the expression of longitudinal clinical traits.\ud \ud \ud \u