Case-control genome-wide association consists in testing an association
between Y, a binary variable (a case-control phenotype), and a set of cat-
egorical explanatory variables, X 1 , . . . , , X p , where X i is the ith Single Nu-
cleotide Polymorphism along the genome. Associations are usually tested
in a pointwise approach where each X i is tested sequentially. Due to the
block structure of the genome, pointwise tests are correlated and a proper
handling of the dependence is needed.
In this work, we focus on SNPSet tests where a block of variables are
jointly tested in an approach similar to the global testing introduced in [1].
In our context, both the dependence pattern and the association signal can
be very different between regions of the genome. The presentation will first
show that the two extreme choices consisting in ignoring dependence or on
the contrary whitening the pointwise test statistics cannot be uniformly pow-
erful over the variety of dependence and association patterns. We therefore
introduce a new class of aggregation methods spanning the range between
ignorance of dependence and complete decorrelation. We also propose a
method minimizing a distance between the null and non-null moment gen-
erating functions of the test statistics within the former class to choose the
more appropriate handling of dependence.
References
[1] Arias-Castro, E., Candès, E.J. and Plan, Y. (2011). Global testing un-
der sparse alternatives: ANOVA, multiple comparisons and the higher
criticism. The Annals of Statistics. 39, 5, 2533-2556..