Dear

Using WGCNA you can select modules significantly correlated to a specific trait (binary). Does it makes sense to use only the genes within these modules as input for the randomGLM package to build a prediciton model for that specific trait of interest (rather than using all the genes in your initial dataset)? Additionally, if you do so, do you have to select only the genes within these modules which seem differentially expressed (calculated by unpaired t-test) between the 2 groups of your binary trait or should you use all the genes within your module?

Thans for your ideas, remarks, feedback,...

randomGLM performs its own variable selection, so no, you don't have to pre-select genes from modules that relate to the trait. It is possible but by no means guaranteed that a selection based on network analysis could help. In some cases it could also hurt.

Peter