Microarray experiments to estimate heterosis

Abstract

The genetic causes for heterosis, i.e., the increased performance of a hybrid plant compared to the parental mean, may be assessed via microarrays. This thesis addresses design and analysis issues of cDNA-microarray experiments with regard to the estimation of heterosis. Standard microarray designs like the loop design or common reference design are not optimal when estimating heterosis. An optimality criterion is devised and two approaches to obtain a suitable design are shown: a rather intuitive one and an approach using simulated annealing. Data transformations are crucial before analysing microarray data. However, transformations may conceal interesting expression patterns. It is shown using a Box-Cox transformation that significance of a heterotic effect is largely influenced by the transformation parameter. Transformation of the linear predictor in a generalized linear model has a similar effect and heterotic effects may—at least partially—be removed by the transformation. For the estimation of linear contrasts between genotypes, a linear mixed model for each gene is fitted to the expression values. To improve variance estimates one may benefit from other genes’ information. Therefore, an empirical Bayes approach is developed that is capable of including more than one variance component in the model.