Figure 2.

Gene Expression Pattern Grid of genes with significant ABA patterns from a comparison of epithelial-like normal colon tissue (blue samples) to colon cancers (red samples; Alon et al, 1999). We have previously determined that 5 samples in the Alon et al. data set were epithelial-like normal using unsupervised bootstrap cluster analysis and removed the remaining muscle-like normals from this analysis. These, and many other published cancer microarray data sets are 'on-tap' in our GEDA web application. The Gene Expression Pattern Grid, which is generated for any set of differentially expressed genes with the GEDA web application, summarizes the types of differential expression in a way that is related to the PPST test. Color signifies that an individual in one group exhibits an expression value that is significantly different from the expression pattern in the other group (red = overexpression; green = underexpression). Black signifies that an individual exhibits an expression value within the specified upper and lower percentiles in the other group. Tumor samples that fall within the upper and lower 95^th %tiles of the distribution of expression values from the normal samples are labeled black, showing which genes for which a sample is are not different from normal. This representation includes information on both the population-level informativeness as well as which individuals appear to exhibit uniquely differentially expressed profiles. Samples within sample group are arranged according to their relative position in a hierarchical agglomerative clustering with pairwise distance = 1-Pearson's correlation coefficient. 'Not expressed' is a hypothesis generated in these graphs when the expression intensity value of that gene for that individual falls in the lower 95^th %tile of the entire data set. Expression pattern grids were produced online with the Gene Expression Data Analysis web application http://bioinformatics.upmc.edu/.