Abstract:
|
Linear mixed models have been widely used in ecological and biological context, in particular genetic studies. Estimation of variance components is usually a necessary part of statistical inference. Common variance component estimation methods, including ANOVA and restricted maximum likelihood estimation (REML), have important limitations such as negative estimators and/or expensive computation. We propose a new estimator based on the Restricted Haseman-Elston method (REHE), which guarantees nonnegative consistent variance component estimates and efficient computation for large dataset. We further introduce subsampling technique applied to REHE to further improve computation speed. The performance of REHE is compared with ANOVA and REML estimators using simulation studies. We also illustrate the proposed method in two case studies, one on network-based pathway enrichment analysis and the other on genome-wise association mapping.
|