Using mb and mbc normalization versions Bin Zhang has built comethylation networks (his code; Pearson correlation). The left one is for mb, the right one is for mbc.
The networks were built for 511 patients (9 patients out of the original 520 were unmatched normal). I did some investigation of the modules in the comethylation network. First two graphs show boxplots of the first principal component of each module in the mb (left) and the mbc (right) networks:
Next two plots show percent variance explained by the first principal component of the first principal component (mb - left, mbc - right):
I think we got pretty good networks and this is very encouraging! Also, for comparison of preprocessing approaches, Bruce has built comethylation networks with TCGA Level2 (beta value) using Bin's and UCLA code. In these networks we could see a gigantic turquoise module that was composed of almost half of the CpG. I think this is something worth thinking about. Does it mean that this module represent all of the methylation or unmethylated CpGs reflecting binary status of DNA methylation (not quite binary because there is also 50% of methylation besides 0% (unmethylated) and 100%(completely methylated, both loci)). Something to think about.