Abstract
An approach based on analysis of variance was applied to raw expression data on 44,760 transcripts in order to identify those with significant differential expression across ileum and colon in Crohn's disease (CD) and ulcerative colitis (UC). The design treated tissue as a block effect, thereby removing this effect statistically and increasing the power to test for effects of disease states (control, CD, and UC). A significant F-statistic for the disease effect was not correlated with the ratios CD/control or UC/control, evidently because many transcripts with high-expression ratios to the control showed inconsistent patterns across tissues. Of 1,053 transcripts showing a significant effect of disease state at the 1% level by the bootstrap test, 508 showed significant difference at the 1% level in a post hoc test for difference between the mean scores for CD and control. These included a number of genes relevant to the mechanism of pathogenesis of CD and a number of genes mapping to genomic regions that have previously shown linkage to CD in association studies.