Integrating Competing but Complementary Association Tests with Applications to Rare Variants Analyzes and Interaction Studies

• Activity Number: 210
• Type: Invited
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lei Sun*+
• Companies: University of Toronto
• Keywords: Statistical Genetics ; Robust Statistics ; Rare Variants ; Interaction ; Fisher's Method ; Power

Abstract

In many studies of genetic data, different statistical methods are proposed with competing claims about power depending on the alternatives. For example, in association studies of complex human traits it is often assumed that the risk allele changes the trait mean level, hence the popular use of various location-tests (T1). However, complex genetic etiologies including GxG and GxE can result in trait variance differences between genotypes, and if so the class of scale-tests (T2) would be more powerful. In rare variants analyses, two classes of tests have been proposed (Derkach et al. Statistical Science), where the linear class (T1, e.g. CAST) is more powerful if the majority of the variants are truly associated with the same direction of effect, while the quadratic class (T2, e.g. SKAT) is better for other cases. To achieve robustness it is natural to combine information from the two (or more) complementary tests. To do so we first show that in both examples T1 and T2 are asymptotically independent under the null. We then discuss ways (e.g. Fisher's method) to aggregate information. We show through simulations and applications the utilities of this robust approach.