The Empirical Hierarchical Bayes Approach for Pathway Integration and Gene-Environment Interactions in Genome-Wide Association Studies

Abstract

Complex diseases such as cancer result from a complicated interplay of multiple genetic and environmental factors. To unveil their genetic component, the simple analysis of single-nucleotide polymorphisms (SNP) as done in genome-wide association studies (GWAS) is not sufficient. Complementary approaches considering the complexity of diseases, such as the incorporation of biological pathway information or detection of gene-environment interaction, are necessary. In this thesis we focus on an empirical hierarchical Bayes model proposed for the integration of external information into genome-wide association studies. We use the approach to incorporate biological pathway information and provide a new test for gene environment interaction (GxE) by adapting the method for that purpose. In an application, we furthermore integrate pathway information with GxE interaction effects by two different strategies. The integration of pathway information by the empirical hierarchical Bayes approach in a GWAS for Rheumatoid Arthritis, characterized by an extremely high number of strong association signals, supports the simple single SNP results, while in an application to four lung cancer GWAS it leads to higher consistency of results between the different studies. In both cases, the strength of the pathway integration approach is supported by the biological plausibility of the results. In comprehensive simulation studies, our new empirical hierarchical Bayes approach for GxE interaction outperforms other GxE methods, having high power to identify GxE interactions in the presence of population-based gene-environment associations (G-E). In our real data application to the lung cancer GWAS with smoking as environmental factor, the method leads to similar results as the powerful case-only test, with the latter in general accompanied by false positive results given population-based G-E associations. However, in the particular data sets, no strong G-E association effects with smoking are observed. By additional integration of pathway information with GxE effects, the consistency of results between the studies increases. Due to the biological plausibility of the results, good candidates for further investigations are identified.