|
Abstract:
|
Many association tests have been proposed for rare variants. Multiple authors show that existing tests can be divided into two classes: tests based on linear composite statistics and tests based on quadratic statistics. Power calculations for these tests currently require the specification of the exact genetic architecture relating locus and trait, which requires specifying a large number of parameters (e.g. proportion of causal variants, size and direction of genetic effect). Here we propose fast and accurate methods to calculate the power of test statistics in both classes. Through theory and simulations, we demonstrate that power can be approximated using at most three parameters: the proportion of phenotypic variation explained by the locus, the number of causal variants, and, for linear statistics, the proportion of causal variants that are deleterious. Furthermore, we use the proposed methods to investigate whether the power of rare variant tests can be increased by restricting the set of tested variants to those predicted to be functional by bioinformatic annotation. We show that the power can be increased if the AUC for identifying functional variants exceeds 0.70.
|
