Efficient experimental designs for two colour microarray systems measuring relative gene expression generally require a mixed model approach to data analysis. Although variance components due to, for example, arrays, animals and residuals are heterogeneous across genes, combining information across genes using shrinkage or empirical Bayes (EB) estimation should jointly improve control of Type I and Type II errors. Shrinkage estimation procedures have been previously developed for several other simpler designs with one notable exception based on REML in a procedure we denote as EB-REML. We propose an alternative approach based on the shrinkage of ANOVA estimates of variance components that facilitates the determination of posterior sums of squares and posterior degrees of freedom leading to posterior F test statistics having well defined null distributions. We conducted a simulation study based on the commonly used loop (incomplete block) design. We compared our method (EB-ANOVA) to EB-REML and mixed model inference based on gene-specific ANOVA and REML estimates of variance components for each of 5,000 genes. The EB-ANOVA method was shown to have the greatest power to detect differentially expressed genes without increasing the false positive rate relative to the other methods.

Robert J Tempelman, L. Xiao

Proceedings of the World Congress on Genetics Applied to Livestock Production, Volume , , 23.22, 2006
Download Full PDF BibTEX Citation Endnote Citation Search the Proceedings

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.