Multivariate methods for identifying differentially expressed genes

Motivation: Univariate testing procedures remain the most common way to identify differentially expressed genes (DEGs). Univariate techniques suffer from the multiple comparison problem and reduced power, because they fail to account for gene interaction. Motivated by these issues, we adopt a multivariate procedure. Namely, we utilize the sup-norm test, which was specifically developed for high dimensional, low sample size problems. We propose an algorithm which repeatedly applies the sup-norm test to screen for DEGs. Results: We evaluate our methodology with both simulated and experimental data. Our simulation studies establish the validity of the sup-norm statistic in terms of Type I error and power. With simulated data sets, the screening algorithm retains the majority of DEGs under a variety of experimental conditions. We also used our methodology to analyze the publicly available ApoAI knockout data set. Our algorithm identified the biologically significant genes, as discussed by other authors (Callow et al., 2000; Smyth, 2004). Availability: A set of Matlab functions used to implement the proposed methodology is available at Bioinformatics online. Contact: [email protected]