In males, 13 of the 35 modules identified were strongly preserved in females but seven modules showed very weak preservation (i.e. unique to males) (Additional file 5).
The most highly-connected and most closely correlated gene to the Eigen gene for each module showed very similar expression patterns for all modules and we have therefore only plotted the most highly connected genes.
For example, two hub genes of modules, PDGFRL and CD163, also show very weak expression similarities across all seven breast cancer datasets: the average Pearson correlation between these two genes is only 0.24.
Three modules, however, showed very weak preservation (i.e. unique to females) in males (Additional file 5: Figure S1A).
They show very little drift.
Genes in module M10 showed very similar temporal down-regulation across brain regions and included several canonical markers of early neural progenitors including DCX, SOX4 and SOX11 mentioned earlier, and showed GO enrichment for cell adhesion and neurotrophin signaling.
Finally, the last two non-core modules show opposite patterns to the core modules.
Because tumor specimens with the same grade annotation showed very different expression pattern of the modules (see Additional file 1), we explored whether expression pattern of the modules could be used to classify tumors more accurately with regard to their likelihood of recurrence.
MCL has been shown very robust to find relevant modules in protein interaction networks [33].
As the CCA network showed very clear modular structures, two network modules (File S6) were manually identified without using other tools.
