Abstract:
|
Genome wide association studies (GWAS) aim to identify the genetic underpinnings of complex diseases and phenotypic traits. When analyzing quantitative traits, standard practice is to regress the phenotype on genotype and covariates, using a model that assumes normally distributed residuals. However, the normal residual assumption is often violated for skewed and heavy tailed phenotypes. Naive application of standard approaches then leads to inflated type I error. To counteract departures from normality, practitioners often implement the inverse normal transformation (INT). Despite their prevalence, the operating characteristics of INT-based tests, that differ e.g. in where covariates enter the model, have not been well studied. Outstanding questions include whether various INT-based tests provide valid inference in GWAS, and if so, which optimizes power for detecting true associations. We develop a robust, INT-based, omnibus test that is both valid and well powered. Towards providing guidance on how to use the INT for GWAS, we conducted simulations assessing the size and power of common INT-based tests. We demonstrate our omnibus approach in GWAS of non-normal sleep apnea traits.
|