WGCNA number of modules issue

Question

[1215045934]

Hi all!

I am working on RNAseq data from a 3 treatment: condition 1, condition 2, and condition 3.

There are 70k genes in my data. I would like to use WGCNA to find genes expressed correlated to conditions, in particular, strongly positive correlated to condition 1, no/low correlation to condition 2, and strongly negative correlated to condition 3. Then I would like to see what functions those genes have by kegg pathway or GO enrichment.

I applied WGCNA on the genes that have a total count more than 10 (60k genes). I got about 80 modules and the Dendrogram looks messy. I am new to WGCNA and I was wondering if this is correct. I did find some modules with the pattern described above with really low p-values.

1. Should I worry about the number of modules? If so, should I change any parameters to improve it?

2. I saw people using top 10000 most variant genes. I did try that and get 8 modules, still finding with 2k genes fit the pattern.

3. Is 2000 genes in a module too many for gene enrichment?

4. What is the difference for the module fit the pattern, and DEG upregulated in condition one in condition 1 vs 2 comparison.

Thanks a lot!

[rpolicastro]

How many samples do you have? You should also share the code you ran for each critical step.

[1215045934]

26 samples, softpower = 6 selected by the lowest number above the 0.9 line

#specify network type
softPower = 6
temp_cor <- cor       
cor <- WGCNA::cor

#-----------------Block-wise network construction and module detection
netwk = blockwiseModules(datExpr, maxBlockSize = 8000,
                         power = softPower, TOMType = "signed", minModuleSize = 30,
                         reassignThreshold = 0, mergeCutHeight = 0.30,
                         numericLabels = TRUE,
                         saveTOMs = TRUE,
                         saveTOMFileBase = "FS-M-allgenes",
                         verbose = 3)

Thanks!

[andres.firrincieli]

there is nothing wrong with that chunck of code. Therefore, the high number of modules is likely given by the strategy used to filter out low count genes/transcripts. My suggestion is to work with the top N most variable genes.

One more thing, a softpower = 6 for a signed network is actually pretty low. When you run pickSoftThreshold you must specify networkType = "signed"

[1215045934]

Thanks! Yeah I didn't specify networkType = "signed".