Statistical Design Issues for Microarray Analysis

Abstract

Microarray experiments generate a high dimensional response from typically a small number of experimental units, the microarray slides. Designs should allocate the small number of microarray slides to treatment conditions in an optimal fashion. In addition, this presentation considers the optimal number of replicate DNA spots per slide in an experiment on Arabidopsis plants, exposed to varying durations of ozone stress. The response is the expression level of some 1200 specific genes. The choice of experimental designs is discussed for efficient detection of the potential regulatory genes. Split plot and incomplete block designs are natural for these experiments. Since two-color microarray slides are subject to dye biases, orthogonal (dye swapped) designs are recommended and described to eliminate the dye effects.