Note that while more than 45% of extracted genes were retained under the most stringent B-score cutoff used, such robustness against statistical significance cutoffs was not observed for other algorithms such as MCODE and modularitybased community detection. Even at a less stringent B-score cutoff of 0:05, the MCODE and modularity-based modules would generally suffer from a loss of over 50% and 95% of identified genes, respectively. Therefore, we did not include the B-score significance measure for the MCODE modules in all comparative analyses. Using the Rembrandt grade II glioma data as an example, the largest module identified by the community detection method as of, consisting of 1,372 genes out of a total of 3,888, was deemed statistically non-significant under the B-score scheme. A careful inspection of this large module showed that three of the statistically significant DiME modules, with sizes of 212, 39 and 42 genes respectively, are contained or almost contained within it. It also has significant overlaps with several other non-significant DiME modules. In comparison, three MCODE modules are contained within the above mentioned large module,Rapamycin with sizes of 77, 18 and 13 genes respectively. Such an observation suggests that community detection is not appropriate for disease module identification in large biological networks, since it generates huge modules with large numbers of genes which add difficulties to validation and interpretation. An analysis of the variability of module identification results show that core modular structure of the Rembrandt coexpression networks used in the case study is well conserved under varying network construction parameters. Such conservation is consistent with the concept of ‘‘module core’’ described by the original authors of module extraction. It is worth pointing out, however, that the less conserved modules do not necessarily bear little functional significance in the network, as their fluctuations may be due to the noise in the biological data itself, rather than in the module identification algorithm. The construction of a highly robust network per se is still a highly active area of research and is not the main focus of this paper. The module connectivity networks for SAR131675 grade II glioma and GBM samples provide a high-level yet insightful understanding of brain tumour progression and the associated rewiring of cellular machinery. A common expression signature of both tumour grades is down-regulation of nervous system development and normal neuronal functions and upregulation of cell cycle related progresses, light green and red nodes). Such concomitant alterations in transcriptome are consistent with a malignant phenotype – cells that are becoming less differentiated and are proliferating more. The coordination between the two types of functional processes is remarkably strengthened in GBM compared with grade II glioma samples, a possible consequence of the significant increase in the transcription factors AR and ETS1 shared by the two processes in both grades. Core components of the two processes are also conserved across microarrays, as is shown by the expression levels of modules 2, 3, and 6 in Figure 7. Also of pathological significance is the significant increase in the activity of the angiogenesis-related module in GBM.