|
Abstract:
|
Set-based tests have become popular for the identification of rare genetic variants that are associated with disease traits in sequencing studies. Most of the existing variants-set association tests are prospective analyses which may have in?ated type I error when the trait model is misspecified. Here, we propose LSRAT (Longitudinal variant-Set Retrospective Association Tests) and RSMMAT (Retrospective variant-Set Mixed Model Association Tests), two groups of retrospective set-level tests for longitudinal phenotypes that are constructed based on the genotype model given the phenotype and covariates. Within each model, we propose three set-level retrospective association tests, corresponding to the burden, variance component, and Cauchy combination tests, respectively. Simulation studies showed that our proposed tests are robust to the trait model misspecification and gain power compared to some of existing methods. We illustrated our method in a longitudinal cohort to evaluate the association of rare variants with cocaine use trajectory.
|
