I am trying to interpret the output of the signedKME function. My data comprises 16 RNA-seq samples across four time points - 8 healthy, 8 crippled. My signed hybrid network (since I am only interested in the positive correlation between downregulated genes) elicits a a dendrogram with one clean, long branch that represents a module of 1217 genes, a number which isn't far off of the number of differentially expressed genes found in other programs. However, I was looking for a stronger signal and without any other trait data, I looked for genes with the highest degree of interconnectedness with the signedKME function. I interpret this as genes most central to the downregulatory cascade by way of having the least deviation from the eigenvalue of count data (expression level). In this way, connectedness is inferred by similarity of expression profile? I sorted for highest values and used orthology searches in the KEGG database and found genes related to the Krebs cycle. I am unlikely to find a smoking gun with this approach, but would the top 50 or 100 genes of this list represent what I am looking for? Is this a reasonable interpretation?

WGCNA does not necessarily give you stronger signal; in fact, if all you want is a list of differentially expressed genes, software that does just that will do usually do a better job. WGCNA often leads to modules that have cleaner biological interpretation that lists of differentially expressed genes (shameless plug: examples are provided in this paper: http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0061505).

Your goal, inasmuch as you state it, is very vague and it is difficult to add anything definite to the answer. The first question would be whether the module is related to your trait (does the module eigengene correlate strongly with the trait). If it is, then yes, the genes with the highest connectivity (highest kME) are the ones to look at, and 50-100 is a good number. If the module eigengene does not correlate with the trait, the module may be irrelevant for the trait.

Connectedness is not really inferred from similarity of gene expression profiles to the summary profile (eigengene), but it turns out that in gene co-expression networks the intramodular connectivtiy tends to be very similar to kME (which is correlation with the eigengene). See the article "Connectivity, Module-Conformity, and Significance: Understanding Gene Co-Expression Network Methods" by Jun Dong and Steve Horvath. Since kME is easier to compute and to assess significance of, we use kME much more than we do the actual intramodular connectivity.

Hope this helps,

Peter