Metabolomics in Obesity and Methods for Analyzing Multiple Phenotypes

Abstract

In this dissertation, we first assess the associations between circulating metabolites and body mass index in two U.S. prospective cohorts. Then, we propose several methods to analyze multiple phenotypes in genetic association studies based on summary statistics.

In the Chapter 1, we investigate the associations between circulating metabolites and repeatedly measured body mass index (BMI) among Caucasian men and women. We employed linear mixed models with a random intercept term to account for the within-subject correlation and adjust for potential baseline confounders including baseline obesity status, smoking, physical activity, alcohol intake, total caloric intake, age at blood draw and follow-up years. We didn't find the associations were heterogeneous between Caucasian men and women, and therefore we used fixed effect meta-analysis to combine evidences of associations from these two cohorts. We found that valine, leucine and isoleucine were positively associated with BMI, while acetylglycine was negatively associated with BMI.

In Chapter 2, we present linear mixed model based score tests for the detection of pleotropic genetic variants that are associated with multiple correlated traits based on summary statistics. Our tests are robust to effect heterogeneity and correlation structures among multiple traits. We conducted simulation studies to compare the proposed methods with existing methods. We also applied our methods to a global lipids GWAS summary statistics data set and identified hundreds of novel genetic variants.

In Chapter 3, we propose a geometric perspective on the powers of principal component association tests based on GWAS summary statistics. Utilizing eigen-analysis of the correlation matrix and asymptotic power analysis, we investigate when PCA is powerful and when it is not to detect the genetic signals. We further apply our methods to a global lipids level genome-wide association study data set and identify hundreds of novel genetic variants that were missed by conventional single-trait analysis approaches. Our results can help guide researchers to choose powerful PCAT methods in multiple phenotype association studies and also better interpret the association results.