Simultaneous confidence intervals for differential gene expressions

In this article, we describe an environment for designing and analyzing gene expression data from 2-channel microarrays, from design of the arrays, to exact multiplicity-adjusted inference for differential gene expressions.There are two possible formulations to the comparison of gene expression levels: tests of equality and confidence intervals. The testing formulation has been more popular, but confidence intervals are more informative because they not only infer the existence of differential expressions, but also bound the magnitude of such differentials. We provide simultaneous confidence intervals.Before gene expression levels can be compared, the data need to be pre-processed to adjust for variations in the arrays, and unequal affinity of genes to dyes. Processed data are then normalized with respect to control genes (which may be all the target genes or a set of housekeeping genes).The approach we take is to model the data. Pre-processing to adjust for array and dye differences amounts to fitting array and dye effects and gene×dyegene×dye interaction in the model. A set of control genes, chosen to be normalized, specifies a set of estimable functions as the normalized expression differentials. The correlation structure of estimates of these functions can then be derived based on the model, allowing accurate multiplicity adjustment to obtain simultaneous confidence intervals. This is illustrated with the analysis of a real data set.