A Gene-Trait Similarity Regression for Genetic Main and Interaction Effects in Genomewide Association Haplotype Analysis

Abstract

Genomewide association studies of complex traits demand statistical tools that are powerful, flexible, and computationally efficient. These tools must be able to detect small-effect genes, model complex interaction effects, and have convincing speed performance. Among the many different multimarker approaches that aim to improve single-marker analysis, haplotype-similarity methods provide one simple and effective way to aggregate information across different loci. However, classic similarity methods tend to share a number of impediments: appropriateness to only binary traits, inability to incorporate covariate information, restriction to genetic main-effect analysis, and/or requirement to permutations when complex model is used. Here we introduce a regression framework for similarity-based approaches that is computationally efficient, able to account for covariates, has the capacity to model both main and interaction effects, and applicable to various trait types. The model regresses trait similarities for pairs of unrelated individuals on their genetic similarities, and detects the significance using a score test for which the limiting distribution is derived. We show that the gene-trait similarity regression is equivalent to random effects haplotype analysis, which enables the derivation of analytical properties of the similarity statistics even when the direct derivation under similarity regression model becomes challenging. Finally, we show that the gene-trait similarity regression does not require phase information and explicitly models the non-additive effects across markers, which makes it an ideal tool for evaluating association between phenotype and marker sets defined by haplotypes, genes or pathway.