comparisons with other experiments, expression levels were scaled such that the mean expression of all arrays was log2(1000). Because the primary objective was identifying genes that could be subjected to further bioinformatic analysis, all probesets currently annotated by Affymetrix as “expressed sequence tags” or whose gene names contained the words “riken”, “predicted”, or “similar to” were filtered out. Next, probe sets with a very low likelihood of actual expression in our samples were removed, with this accomplished by the Bioconductor package “genefilter.” Probe sets that did not have at least 25% of samples with normalized scaled expression greater than 64 were filtered out as well. Linear modeling to calculate gene-wise p-values for the contrasts of the CA versus W group and the MSA versus W group was performed using the package Limma (Smyth, 2004); probe sets were considered to be statistically significant at p < 0.01, with a false discovery rate (FDR) less than 0.15. An FDR of 0.15 was used as a cutoff because this allowed a significant number of genes to be included in the Gene Ontology (GO) and Ingenuity® Pathways Analysis to help identify networks of genes that changed. This FDR value is a reasonably stringent cutoff
