Hi, I'm attempting to construct a consensus WGCNA network for 4 related expression datasets. For a signed network, two of the datasets achieve a high scale-free topology very early, but for the other two they don't reach 0.8 until a very high power.
Here, the last of four datasets achieves a scale-free topology at 28, although there appears to be a slight peak in this metric at 17.
A power of 28 seems high to me (although I may be wrong), especially when the WGCNA FAQ recommends choosing a maximum power of 18 for signed datasets when presented with a lack of scale free topology. Is there any argument to be made for 17 since the graph shows that slight peak? https://horvath.genetics.ucla.edu/html/CoexpressionNetwork/Rpackages/WGCNA/faq.html
If it makes more sense to consider our multi-expression data to lack scale-free topology, suggesting we choose a power from the table below, how should we determine our sample size from column 1?
Should we consider our total number of samples across all 4 datasets (>70), or should we choose the average, minimum, or maximum number of samples per dataset (<20)?
In a related question, does the scale-free topology look better if we use an unsigned network (please ignore the graphical numbering glitch)?
Based on these results, it seems we should choose a power of 11 since it is the lowest power where all 4 datasets have a scale-free topology >0.8.
If we were to choose an unsigned network instead, is it valid to run downstream statistics on eigengene values from an unsigned network even though modules would contain sets of genes with both positive and negative correlations? Since the eigengene values represent principal component 1 for a given module, I assume they represent the strongest expression signal within a module and would not combine signals from both positively and negatively correlated gene sets. Am I wrong in this assumption? Are the genes that contribute to PC1 in each module held consistent when producing eigengene values across all samples and datasets?
Thank you for all your help!