The estimation and seeding process used in our algorithm is relatively simple and straightforward, we could for each extraction procedure view each node as a possible ‘‘seed’’ for the module to be extracted, which will progressively include its surrounding nodes to form the module during optimisation. The initial extraction with relatively few individuals would, then, act as the seed prioritizer. To ensure that the modules extracted from the biological networks are statistically significant, i.e. they are significantly different from modules that arise from random networks of an appropriate null model, we incorporated a B-score significance measure as proposed in as a quality control step for the modules. The B-score measure assumes a null model where edges within the module of PI-103 interest is held unchanged while the remaining connections in the network are randomly shuffled. Then the B-score is calculated based on the null module to quantify how often we should expect to see the module ‘‘by chance’’. The B-scoring measure has a major advantage of avoiding large amounts of resampling cycles for simulating null model results. In our later experiments, we also showed that the Bscoring measure worked well with our DiME algorithm to detect statistically significant modules. For details of the B-score calculation, the reader is referred to the original works. In order to make this paper selfcontained, we provide the full procedure for B-score computation in Section S1 in File S1. In this study all B-score calculations were based upon default parameters in the original work with 20 independent runs for each module evaluation. We then selected the a certain percentage of the top ranking gene pairs as significant co-expressions, which will be connected as a coexpression network. This percentage, called network construction threshold in our paper, will affect the edge noise level of the resulting network. For example, PLX4032 a stringent threshold will miss large numbers of true edges while a larger threshold will introduce many false-positive edges. For our glioma co-expression network analysis, we set the network construction threshold to 0.1%. In the Results section, we also present data regarding how different network construction threshold values, therefore different noise levels, affect module extraction results of the DiME algorithm. Edge widths are designed to be proportional to the number of connections between two modules, in order to illustrate strength of coordination between functional components in the disease network. Node color represents fold change of average expression level of all genes in one module compared with normal patient samples. The grade II glioma module network demonstrates a significant shift in the tumour phenotype compared with normal samples. As would be expected, there appears to be a marked down-regulation of normal neuronal function, and a significant increase in cell cycle-associated processes. It is of interest to note that the modules associated with immune response are slightly, but significantly increased in grade II tumours.