GausSKAT : A Data‐Adaptive Weighted Gaussian Kernel Procedure for Rare‐Variant Association Testing Within the SKAT Framework
- 18 November 2026
- journal article
- Published by Wiley in Statistics in Medicine
- Vol. 45 (25-27)
Abstract
Genome‐wide association studies have limited power to detect rare‐variant associations. The sequence kernel association test (SKAT) enhances power by jointly testing variants within a region, but its power depends on how the kernel represents genetic similarity. The weighted‐linear kernel predominates in rare‐variant testing, whereas the weighted Gaussian kernel remains less widely used. Varying the Gaussian bandwidth hyperparameter changes the representation of genetic similarity from localized and nonlinear to increasingly linear‐like. However, very small hyperparameter values can cause similarities between nonidentical weighted genotype profiles to collapse toward zero, whereas very large values cause the weighted Gaussian kernel to behave increasingly like weighted‐linear SKAT and therefore provide little distinct information. Because the underlying genetic architecture is unknown, no single hyperparameter value can be expected to work best across regions. We propose GausSKAT, which evaluates weighted Gaussian SKAT over a phenotype‐independent, data‐adaptive hyperparameter grid and combines the resulting dependent values using the aggregated Cauchy association test (ACAT). The grid endpoints are informed by weighted genetic distances and the large‐hyperparameter behavior of the covariate‐adjusted kernel. In simulations, GausSKAT generally achieved higher power than weighted‐linear and weighted identity‐by‐state SKAT under the genetic architectures considered. The best‐performing hyperparameter value varied across settings, supporting aggregation across multiple values. In the Dallas Heart Study analysis of triglyceride levels, the relative performance of GausSKAT and weighted‐linear SKAT varied among ANGPTL gene‐family variant sets. These findings support GausSKAT as a complementary extension that broadens the similarity structures represented within SKAT.This publication has 20 references indexed in Scilit:
- Convex combination sequence kernel association test for rare‐variant studiesGenetic Epidemiology, 2020
- Rare variant association testing for multicategory phenotypeGenetic Epidemiology, 2019
- Small Sample Kernel Association Tests for Human Genetic and Microbiome Association StudiesGenetic Epidemiology, 2015
- Sequence Kernel Association Tests for the Combined Effect of Rare and Common VariantsThe American Journal of Human Genetics, 2013
- Kernel Machine SNP‐Set Testing Under Multiple Candidate KernelsGenetic Epidemiology, 2013
- Kernel machine SNP-set analysis for censored survival outcomes in genome-wide association studiesGenetic Epidemiology, 2011
- Rare-Variant Association Testing for Sequencing Data with the Sequence Kernel Association TestThe American Journal of Human Genetics, 2011
- Powerful SNP-Set Analysis for Case-Control Genome-wide Association StudiesThe American Journal of Human Genetics, 2010
- Methods for Detecting Associations with Rare Variants for Common Diseases: Application to Analysis of Sequence DataThe American Journal of Human Genetics, 2008
- A Powerful and Flexible Multilocus Association Test for Quantitative TraitsThe American Journal of Human Genetics, 2008