MIXED EFFECT LINEAR MODEL FOR THE ANALYSIS OF GENE EXPRESSION DATA

- ii-Microarray technology enables researchers to simultaneously monitor the expression levels of thousands of genes. This thesis addresses the fundamental question of identifying differentially expressed genes in replicated measurements of biological tissue samples in two different conditions. A statistical model of gene intensity observations is developed. We use a mixed-effect linear model which explicitly accounts for both patient to patient variability and residual sources of error in the gene intensity observations. Applications of this model to the analysis of 3 data sets collected using different types of microarray technology are presented. Two of these data sets contrast “end stage human heart failure ” tissue to “normal ” human heart tissue, and the other investigates variability of gene expression in a large population of yeast cell cultures [11]. Our procedure has proven successful in the analysis of these 3 data sets. The assumptions underlying the model were verified, and sets of differentially regulated genes could be identified based on the calculation of statistical significance level