WGCNA-consensus network analysis using RNA-seq data

Question

[pixie@bioinfo]

Hello, I have raw read counts from various stress conditions (4 types) from the same RNA-seq platform and tissue. I had previously determined the DEGs from the individual stresses separately and also extracted their normalized expression matrix for each of the stress conditions .

If I wish to perform a consensus module analysis using the WGCNA frame-work, do I need to perform 'vst' normalization in DeSeq2 across all the conditions ? Any leads to go about it will be very useful. Thanks

Answer 1

[andrew.j.skelton73]

For use in WGCNA, your data should be log2-like, so vst or rlog transformations are appropriate. Alternatively, you could use Limma Voom too for RNAseq data.

[pixie@bioinfo]

Thanks, my only concern was that by merging all the samples of all the conditions in one matrix, am I diluting the any information. But after going through their tutorial, they have possibly done that (i.e. normalized across male and female data)

[andrew.j.skelton73]

If these are from different experiments, then that's a whole different box of questions. Normalising cross experiment is far from trivial, and can only be done in some cases. The big caveat for WGCNA is that you need a decent number (>20) of samples to get interpretable output.

If these are cross experiment, then I'd recommend that you do WGCNA per condition agnostic of one another, then compare / contrast after you've generated modules.

If you're worried about covariates, or there's a strong effect that you want to account for, you could always take the residuals from a model fit (check out the removeBatchEffects function in Limma). Word of warning though, you're then going down the road of a lot of statistical caveats, make sure you truly understand the consequences of each step.

[pixie@bioinfo]

Thanks so much for your time and detailed explanation. I will give a proper thought to this before I jump into network analysis.