Residual-based Test of Conditional Association between Continuous and Ordinal Variables with Application to Genome-wide Association Studies

Abstract

The discovery of genes linked with a large array of diseases has been accelerated by genome-wide association studies (GWAS), in which genetic variants in different individuals are examined for relationship with a specified phenotype. Most GWAS analyses require modeling the association between single nucleotide polymorphisms (SNPs) and the outcome of interest as additive, dominant, or recessive. In general, this relationship is not known. The genotypes of a marker can be regarded as ordered categorical. An additive model assumes linearity, and approaches that categorize the data ignore order information, resulting in loss of power. Therefore, a method that only assumes a monotonic relationship between SNPs and the outcome of interest may be more robust and powerful than standard approaches. In this thesis, we explore the use of such a method using pharmacogenomics data from a clinical trial that randomized 1858 HIV-infected patients to one of four antiretroviral regimen combinations (tenofovir+efavirenz, tenofovir+atazanavir, abacavir+efaverinz, and abacavir+atazanavir). We are specifically interested in detecting SNPs that are associated with tenofovir clearance and creatinine clearance. We assess the performance of the new method versus the additive, dominant, and recessive models via simulation studies and real data analyses, and compare and contrast findings.