Abstract:
|
Kernel machine (KM) models are powerful tools for exploring associations between sets of genetic variants and complex traits. While KM analyses involving a single kernel are useful for assessing the marginal effect of a variable set, multi-kernel models are becoming increasingly popular for studying more complex problems such as studying gene-gene or gene-environment interaction, or assessing conditional effects of a variable set. The KM framework is robust and provides efficient dimension reduction for multi-factor analyses, but requires estimation of high dimensional nuisance parameters. Traditional estimation techniques like regularization and the EM algorithm are computationally expensive and are not scalable to larger sample sizes. Under the context of gene-environment interaction, we propose a computationally efficient fastKM algorithm for multi-kernel analysis. Based on a low-rank approximation to the nuisance effect kernel matrices, fastKM is just as powerful as EM-based KM methods for quantitative traits but is much faster. Additionally, fastKM is applicable to many trait types, and can be implemented using any existing single-kernel analysis software.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.