I'm just getting into using WGCNA and had a question about my last run through. I'm looking at differences between two species in brain gene expression using RNAseq data. WGCNA produced 4 modules including the set of unassigned genes (labeled module 0 or the grey module).
The grey module was highly and significantly correlated with species identity, whereas none of the other modules had any apparent relationship.
It makes intuitive sense to me that any genes related to species differences would have a lot of noise and, thus, be difficult to assign to a module when grouping together both species.
My question is if this interpretation is correct and, if so, if it makes more sense to use a consensus module approach to compare species?
I have not seen your data and analysis so it's difficult to make definitive statements, but I would be suspicious if the genes different between species were (mostly) in the grey module. I would expect the that inter-species differences are the biggest source of variation in the data and hundreds if not thousands of genes should have strong differences between the species, although this obviously depends on what species are being compared. If the differences are large, they should show up as large modules.
Maybe you could tell us a bit more about what the species are and how did you run the analysis, especially how you transformed the counts before running WGCNA.
Thanks very much for your reply! It's fantastic that you respond to support questions...
The two species are very closely related songbird species (< 1my since last common ancestor). Sequencing was done on a smallish brain region on 5 individuals from each species (and 3 hybrids). The counts were transformed into transcripts per million reads. Then, for WGCNA, I transformed them using a log(tpm +1) transformation and removed genes with very low values.
Only 55 genes were found to be differentially expressed between species.
As far as the analysis goes, I followed the tutorial more or less, running a single analysis on all individuals together, using a soft threshold of 6.
Then, I considered species as a "treatment." The eigengenes for the grey module were all positive for one species, all negative for the other, and were close to 0 for the hybrids. This is why I originally found the result potentially interesting.
I've since tried running both separate WGCNA on each species and/or a consensus approach and am now figuring out how to compare module consistency, etc... which I am under the impression might be a better approach?
I am surprised that you only see 55 genes DE between the two species, but maybe that's the biology. I would double-check that there were no batch effects or other technical effects. You could make the module identification more relaxed, e.g. lower the minimum module size and/or increase deepSplit; it seems that there is a module (or modules) of genes DE between the species but the module is not very distinct or is too small.
Ah, fantastic! Thanks for the help. The few DE genes is something that's been found a few times independently in these species, so we think it does represent the biology.
Increasing deepSplit seems to have worked, producing three modules (between 80 to 160 genes) whose eigengenes are significantly associated with species identity. Two of those also associate hybrids in between pure species.
Thanks very much for your reply! It's fantastic that you respond to support questions...
The two species are very closely related songbird species (< 1my since last common ancestor). Sequencing was done on a smallish brain region on 5 individuals from each species (and 3 hybrids). The counts were transformed into transcripts per million reads. Then, for WGCNA, I transformed them using a log(tpm +1) transformation and removed genes with very low values.
Only 55 genes were found to be differentially expressed between species.
As far as the analysis goes, I followed the tutorial more or less, running a single analysis on all individuals together, using a soft threshold of 6.
Then, I considered species as a "treatment." The eigengenes for the grey module were all positive for one species, all negative for the other, and were close to 0 for the hybrids. This is why I originally found the result potentially interesting.
I've since tried running both separate WGCNA on each species and/or a consensus approach and am now figuring out how to compare module consistency, etc... which I am under the impression might be a better approach?
I am surprised that you only see 55 genes DE between the two species, but maybe that's the biology. I would double-check that there were no batch effects or other technical effects. You could make the module identification more relaxed, e.g. lower the minimum module size and/or increase deepSplit; it seems that there is a module (or modules) of genes DE between the species but the module is not very distinct or is too small.
Peter
Ah, fantastic! Thanks for the help. The few DE genes is something that's been found a few times independently in these species, so we think it does represent the biology.
Increasing deepSplit seems to have worked, producing three modules (between 80 to 160 genes) whose eigengenes are significantly associated with species identity. Two of those also associate hybrids in between pure species.
Thanks again, very much, for the help!
David