Title: Testing statistical hypotheses for genotype—phenotype association
Abstract: The aim of this presentation is to discuss selected statistical challenges in single and multiple hypothesis testing for the analysis of data from genome-wide association studies (GWAS). The statistical basis will be the score test in generalized linear models (GLM), and challenges and results will be illustrated on real data.
In GWAS we scan genetic markers across the genome with the aim to identify genetic variants that are more common in individuals with a particular trait or disease than without. For many diseases this has impoved our understanding of the disease mechanism.
By construction a GWAS is a fishing expedition, where we perform a single hypothesis test for each of in the order of million genetic markers. P-values from single hypothesis tests need to be constructed to be valid. In GWA studies the standard is to control the familywise error rate (FWER), and the high multiple testing burden has resulted in the use of a p-value cut-off of 5*10^-8.
The score test in the GLM setting will be presented, and I will in particular address the role of non-genetic covariates, methods for calculating p-values and a method for efficient FWER control. No previous knowledge of GWA studies is needed.