Theoretical and computational aspects of association tests: application in case-control genome-wide association studies
vendredi 13 juin 2014, 9h30 - 10h30
Case-control genome-wide association studies (GWAS) aims at detecting statistical and/or biological association between a binary phenotype, Y, and a large set of genotypes, X=[X1,…,Xp]. One of the main characteristics of GWAS are the following:
1) Y, the dependent variable, as well as Xi’s, the explanatory variables, are categorical and 2) GWAS falls into the class of high dimensional data with n smaller than p.Thus, from a statistical point-of-view, performing a GWAS raises several issues such as the choice of the statistical test and the relationship between statistical association and biological perspective.
My talk will first focus on single association test where the association between Y and each Xi is tested independently. For doing so, three main tests are performed in practice: Chi2 test, Lkelihood Ratio test and a OddsRatio-based test. In this part, I will discuss the choice of the « best » test with respect to the statistical power and the computational preformance.
In a second part, I will concentrate on the detection of association between Y and a pair (Xi,Xj), often called epistasis in genetics. I will describe a test, called IndOR, proposed as an alternative of traditionnal regression-based test in case of non-linear association.
In both part, I will illustrate the theoretical results on the analysis of a benchmark dataset published in 2007 by the Welcome Trust Case Control Consortium.