Statistical Methods for Modeling Heterogeneous Effects in Genetic Association Studies

[摘要] Effect-size heterogeneity is a commonly observed phenomenon when aggregating studies from different ancestries to conduct trans-ethnic meta-analysis. Irrespective of the sources of heterogeneity, traditional meta-analysis approaches cannot appropriately account for the expected between-study heterogeneity. Therefore, to bridge the methodological gap, in the first two projects, I develop statistical methods for modeling the heterogeneous effects in trans-ethnic meta-analysis for genome-wide association studies (GWAS). In the third project, I extend the methods in trans-ethnic GWAS meta-analysis to a general statistical framework for modeling heterogeneity in biomedical studies. In the first project, I develop a score test for the common variant GWAS trans-ethnic meta-analysis. To account for the expected genetic effect heterogeneity across diverse populations, I adopt a modified random effects model from the kernel regression framework, and use the adaptive variance component test to achieve robust power regardless of the degree of genetic effect heterogeneity. From extensive simulation studies, I demonstrate that the proposed method has well-calibrated type I error rates at very stringent significance levels and can improve power over traditional meta-analysis methods. In the second project, I extend the common variant meta-analysis approach to the gene-based rare variant trans-ethnic meta-analysis. I develop a unified score test which is capable of incorporating different levels of heterogeneous genetic effects across multiple ancestry groups. I employ a resampling-based copula method to estimate the asymptotic distribution of the proposed test, which enables efficient estimation of p-values. I conduct simulation studies to demonstrate that the proposed approach is well-calibrated at stringent significance levels and improves power over current approaches under the existence of genetic effect heterogeneity. As a real data application, I further apply the proposed method to the Type 2 Diabetes Genetic Exploration by Next-generation sequencing in multi-Ethnic Samples (T2D-GENES) consortia data to explore rare variant associations with several traits. In the third project, I develop a supremum score test for jointly testing the fixed and random effects in a generalized linear mixed model (GLMM). The joint testing framework has many applications in biomedical studies. One example is to use such tests for ascertaining associations under the existence of heterogeneity in GWAS meta-analysis; another example is the nonparametric test of spline curves. The supremum score test first re-parameterizes the fixed effects terms as a product of a scale parameter and a vector of nuisance parameters. With such re-parameterization, the joint test is equivalent to testing whether the scale parameter is zero. Since the nuisance parameters are unidentifiable under the null hypothesis, I propose using the supremum of score test statistics over the nuisance parameters. I employ a resampling-based copula method to approximate the asymptotic null distribution of the proposed score test statistic. I first investigate the performance of the method through simulation studies. Using the Michigan Genomics Initiative (MGI) data, I then demonstrate its application by assessing whether the genetics effects to Low Density Lipoprotein Cholesterol (LDL-C) can be modified by age.

[发布机构] University of Michigan

[效力级别] Trans-ethnic meta-analysis

[关键词] Effect-size heterogeneity;Trans-ethnic meta-analysis;GWAS;Statistics and Numeric Data;Science;Biostatistics