Detecting Genetic Variants Based on a Random-Effects Model

In our recently published paper in the Statistics in Medicine (2014), we propose a new approach, based on a random-effects model, to detecting association signals for a group of variants. Our method does not compare genetic information directly between the case and control groups. Instead, we test whether the fraction of causal variants equals zero. We demonstrate that this approach is more powerful than most of the well-known rare-variant association tests and more robust to the inclusion of non-causal variants or high frequency but low effect variants, and to the presence of missing genotypes. In addition, the same approach can be modified for detecting gene-disease association signals in case-parents studies. Unfortunately, the former method does not allow for the inclusion of covariates in the analysis. On the other hand, the latter method cannot be applied for detecting gene x gene or gene-environment interactions. In this three-year project, I propose to address these issues. Specifically, I propose to study methods for 1. detecting association signals in genetic studies with covariates based on a random-effects model, including the study of the population stratification effect and correction; 2. simultaneously detecting main and/or interaction effects of a single or group of SNPs in population-based case-control studies; 3. simultaneously detecting main and/or interaction effects of a single or group of SNPs in family studies. We shall compare the performance of the new methods with the well-known methods in the literature and make suggestions to the users.