To generate the differential expression signatures firstly each gene was classified into a starting signature based on its pattern of significant differential expression over the comparisons: <*mde_comparisons*>. For example in an experiment with three groups (a, b and c) and two comparisons (a vs b, and a vs c) there would be ten starting signatures (a vs b up + a vs c up, a vs b up + a vs c down, a vs b up + a vs c unchanged, a vs b down + a vs c up, a vs b down + a vs c down, a vs b down + a vs c unchanged and , a vs b unchanged + a vs c unchanged. Genes that were unchanged in all differential comparisons were excluded. Next the expression (for all samples in the sample groups (<*sample_groups*>) values for each gene was converted into an expression z-score. For each signature a metagene expression value was created, i.e the mean expression z-score across all genes in that signature was calculated for each sample. To determine signatures with a similar expression profile, for each pairwise combination of signatures the two expression metagenes were correlated to each other using a Spearman Correlation Coefficient. The two signatures with the highest correlation were merged (provided the correlation was above <*signatures_SCC*>. This process was iteratively repeated from the generation of the metagene step until there were no correlations above <*signatures_SCC*>. The resultant merged signatures were used.

