Abstract:
|
Traditional association test usually treats a complex trait as the response variable and genotype as the predictor. The general idea of 'reverse' regression has been proposed (e.g. MultiPhen, O'Reilly et al. 2012) but its implementation via ordinal logistic regression is limiting. In this work, we propose a new 'reverse' regression framework that adjusts for complex pedigree structure. We show that the score test statistic under this new framework takes identical form to the traditional formulation, but requires no distributional assumptions on the phenotype traits, and thus can handle extreme-sampling phenotypes. When no phenotypic trait is included, the model can be further extended to estimate the allele frequency in complex pedigree while adjusting for other covariates. We prove that this estimator is the best linear unbiased estimator (BLUE, McPeek et al. 2004). We support our analytical findings using evidence from both simulation and application studies.
|