Question: WGCNA: Relating Modules to Categorical External Traits
gravatar for mjc5038
20 months ago by
United States
mjc503830 wrote:


I am using WGCNA to identify co-expressed gene networks from an RNAseq data set in which 24 mice were assigned to one of 4 experimental treatment groups. I have cleaned my data and there are 13861 genes that were used for the network construction using a signed network type, power = 18 (recommended from the network fit indices), deepSplit = 4, minModuleSize = 30, mergeCutHeight = 0.25. I have 19 modules ranging in size from 52 - 5486 genes. Now I would like to determine if there are any modules are related to experimental condition. Does anyone have a recommendation as to how I should compare modules across groups?

Thus far, I have run one-way ANOVAs on each module eigengene and corrected using FDR < 0.05 with the following loop. Note: datTraits[, 1] contains my grouping variable

Filename = net4_30_18 #This is the name of my network 

Unlisted = data.frame()

i = 1

while (i <= NCOL(Filename$MEs)){           
    Unlisted = rbind(Unlisted, unlist(summary(aov(Filename$MEs[ , i] ~ datTraits[ , 1]))))
  i = i+1  

names(Unlisted) = c("DF1", "DF2", "SS1", "SS2", "MS1", "MS2", "F1", "F2", "PR1", "PR2")
row.names(Unlisted) = names(Filename$MEs)

Cutoff = 0.05/NCOL(Filename$MEs)

summary(Unlisted$PR1 <= 0.1)
summary(Unlisted$PR1 <= 0.05)
summary(Unlisted$PR1 <= 0.005)
summary(Unlisted$PR1 <= Cutoff)

summary(p.adjust(Unlisted$PR1, method = 'fdr') <= 0.05)
summary(p.adjust(Unlisted$PR1, method = 'fdr') <= 0.1)

Unlisted[p.adjust(Unlisted$PR1, method = 'fdr') <= 0.05, ]

Does anyone have any recommendations on how they would approach this?





ADD COMMENTlink modified 20 months ago • written 20 months ago by mjc503830
gravatar for Peter Langfelder
20 months ago by
United States
Peter Langfelder1.2k wrote:

Well, this is really a question on how to compare 4 groups that has nothing to do with WGCNA. I think you did the WGCNA part well, now you want to test changes of a few (19, to be precise) variables between 4 treatment groups. There are multiple ways to test differences among 4 groups and you may want to consult a statistician (rather than the BioC mailing list).

ADD COMMENTlink written 20 months ago by Peter Langfelder1.2k
gravatar for mjc5038
20 months ago by
United States
mjc503830 wrote:

Thank you, Peter. Confirmation that my preliminary steps are on the right track is reassuring. 

ADD COMMENTlink written 20 months ago by mjc503830
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 266 users visited in the last hour