Estimation of genetic effects incorporating prior information

摘要

We study estimations of the genetic effect of a marker by adjusting out covariates and incorporating the results of previous potentially heterogenous studies of the same genetic marker. Without prior information on the covariates, the procedures are based on both frequentist and Bayesian methods by simultaneously maximizing the likelihood function for the coefficients of the covariates and minimizing the loss function for the genetic effect, and hence are regarded as hybrid estimations. Although we focus on an application to case-control genetic association studies, we describe a general method for various types of traits. For the application, we show that the proposed hybrid inference based on the prospective sampling can be applied to retrospectively collected case-control data. Simulations and applications using hybrid inference are presented.