|
Abstract:
|
Identifying genetic variants associated with a complex disease has benefited from recent advances in set-based and multi-trait testing methods. Jointly testing sets of variants (e.g., those corresponding to gene sets or pathways) for association with a set of phenotypic disease traits can be more powerful than testing individually with single variants or single traits. A common challenge in set-based testing is that sample size may be small and vastly exceeded by the number of variants. To overcome this, we develop an adaptive kernel-based test that uses a supervised method to filter out noncausal variants and reduce dimension. Our test does not rely on assuming a particular functional form for the association or a particular distribution for the traits, and allows the genetic effect on each trait to be captured using a different kernel function, employing a fast kernel selection method based on asymptotic results under a high-dimensional setting. We demonstrate the speed of our method written in C++ and implemented in an R package, and we compare its effectiveness against other kernel-based tests in both simulations and applications to real-world data.
|
