STATISTICAL METHODS FOR ANALYSIS OF GENOME-WIDE ASSOCIATION STUDIES ACROSS MULTIPLE TRAITS
QI-DISSERTATION-2020.pdf (31.74Mb) (embargoed until: 2024-05-01)
MetadataShow full item record
Pleiotropy is the phenomenon that one genetic variant has effects on multiple phenotypes. Genome-wide association studies found widespread pleiotropy across complex traits and diseases, which has transformed the interpretation of GWAS results and understanding of genetic architecture. The discovery of pleiotropy has provided major opportunities for novel statistical analysis of GWAS. In this thesis, I first describe a method to aggregate information across multiple traits to improve power of genetic association tests. Application of the method identified novel loci associated with blood lipids, psychiatric diseases and social science traits. Second, I describe a method for robust Mendelian randomization analysis using mixture models, in order to identify causal relationships between risk factors and diseases even when the instrumental variable assumptions are violated. Application of the method identified a protective effect of later menarche on the risk of breast cancer, and no causal effect of HDL cholesterol and triglycerides on the risk of coronary artery disease. In addition, I present a comprehensive evaluation of Mendelian randomization methods using realistic simulation studies informed by recent studies on heritability and genetic effect size distribution. Comparison of the methods in real data analysis to study the causal effect of blood and urine biomarkers on type 2 diabetes revealed major heterogeneity in estimated causal effects among some biomarkers. In conclusion, novel statistical methods for pleiotropic analysis have led to new insights into the genetics of complex traits and the causal role of risk factors in diseases.