Opposing expression profiles in the same WGCNA module

Question

[cfourps]

I've run WGCNA using the 'signed' approach and 18 for the soft threshold power (determined using the 'Scale independence' and 'Mean connectivity' figures as shown in the tutorial). I am confused with the output modules because many of them include genes with opposing expression profiles (negatively correlated).

I've attached an example here where I have plotted the Z-scores (row-wise) of the genes in one of the modules. You can see that the genes in the top ~2/3 are up-regulated in the right-most samples, while the bottom ~1/3 are down-regulated in those same samples. I thought the 'signed' approach should prevent such clustering from happening. Am I misunderstanding something or do I just have poor clustering?

Any other suggestions for prioritizing transcription factors for downstream experiments? I was hoping to use the 'connectivity' metrics from WGCNA for this.

[andres.firrincieli]

I thought the 'signed' approach should prevent such clustering from happening. Am I misunderstanding something or do I just have poor clustering?

This is odd, can you post the code used to generate the network?

[cfourps]

Thanks for helping. Here's the code for building the network:

net = blockwiseModules(datExpr, power = 18, # 16 (8 for signed hybrid) recommended for signed network consisting of 20-30 samples; 14 for 30-40
                       TOMType = "signed", #"signed", #signed recommended in faqs
                       minModuleSize = 30, numericLabels = TRUE, pamRespectsDendro = FALSE, 
                       saveTOMs = TRUE, saveTOMFileBase = "TOM",
                       verbose = 3, #detectCutHeight = 0.95,#mergeCutHeight = 0.25, # 0.25 is used in tutorial for mergecutheight
                       corType = "pearson", #bicor recommended in faqs, but it gave warnings: bicor: zero MAD in variable 'x'. Pearson correlation was used for individual columns with zero (or missing) MAD
                       maxBlockSize=100000, nThreads=8, randomSeed=32786)

[andres.firrincieli]

Maybe you forgot to set networkType = "signed"

Let me know

[cfourps]

Oh! Sorry I must've confused the 'TOMtype' and 'networkType' parameters.

Edit: I just confirmed that setting that parameter prevents negatively correlated genes from being clustered, as I was expecting. Thank you for pointing out the error!

[andres.firrincieli]

the networkType is for the adjacency matrix. By default is networkType = "unsigned". That might be the reason why your modules includes negatively correlated genes