Experiments in genomics, proteomics, and metabolomics generate measurements on large sets of features (genes, proteins, compounds). To identify the genes which are differentially expressed in a microarray study, it is necessary to employ hypotheses testing on a large scale. From a statistical point of view, when a large number of hypotheses are being tested simultaneously, false discoveries become inevitable. Adjustment for multiple comparisons and/or false discovery rate estimation is necessary to target the genes most likely to show differential expression in follow-up experiments. In addition, interest lies not just in whether a gene is differentially expression but in how much differential expression is present, and this question is subject a similar problem: in a large set of genes, some will appear highly differentially expressed just by chance. In this research, we present statistical methods which, by sharing information across genes and taking the variability of the data into account, estimate the probability of equal expression and the expression ratio in a manner that accounts for the possibility of false discoveries. Although the explanation is phrased in terms of differential gene expression, the methods are general and can apply to differential comparisons in proteomics, metabolomics, and other settings with large numbers of features.

We have conducted gene expression profiling of yeast fermentation cultures using custom microarrays designed for the Agilent 8X15K platform. We will present statistical analyses of expression profiles of wild-type progenitor and xylose-utilizing genetically engineered strains grown in the presence of glucose or xylose.