Examining connectivity of specific genes in WGCNA

Question

[nancydong20]

Hello!

I am currently learning about WGCNA as preparation for an up-coming project, studying a potential novel transcription factor. I am interested in seeing which genes are expressed in a correlated manner with my transcription factor, can WGCNA be used for this?

Thank you very much for shedding some light on my very basic question!

[Lluís R.]

Yes! WGCNA creates groups of genes working coordinated. But you'll need more than 20 samples per condition.

[nancydong20]

Thank you very much!

Answer 1

[ildem]

yes, definitely.

Keep in mind: 1. sample purity: if you are using tissues, differential composition of samples can drive artificial connections 2. Need to use RNA-seq from a similar batch (i.e. sequenced around the same time on the same machine, prepared by the same experimenter etc.) Otherwise the differences can null/dull down your data. 3. I would say you need at least 35 samples to have good p-values for connectivity, assuming each sample has a different condition

good luck!

[nancydong20]

Thank you very much for the suggestions!

Regarding your suggestion #3, if I have three experimental groups such as control, "stimulated" and knockdown+"stimulated", do I need 35 samples per group?

As a side question, is WGCNA a suitable approach if I this is what I'm planning:

1. First characterize the overall "transcriptional landscape" after "stimulation" and see what the potential role of my transcription factor of interest plays in it. (In the context of WGCNA would be to see which module it is in and do functional enrichment of the other genes in that module?)
2. Then knockdown the transcription factor and see how the "transcriptional landscape" is perturbed. I maybe can do this using differential WGCNA, comparing the wild type "stimulated" WGCNA with the knockdown "stimulated" WGCNA?

[Lluís R.]

Depend what kind of co-expression relation do you want and how pure is. The better it would be to have enough samples per group otherwise the differences between the groups might hide real connections between genes.

1. Yes. It is not unusual to do so. In fact WGCNA has functions to do so (but I would recommend more on purpose software like topGO, clusterProfiler, GOseq, or others.
   1. In the webpage of WGCNA you have a tutorial to compare two "transcriptional landscapes", to see which modules are conserved in the two conditions and so on.

[nancydong20]

Fantastic, I will go look at that. Thank you very much!