smorabit
commented
1 week ago Ultimately it is up to you how you design your network analysis, as there are many different ways that you could do it. It is hard for me to give a recommendation one way or the other.
I am retroactively pondering what the implications of including all conditions vs. a subset would be. In our experiment we have two mouse strains, two timepoints and two treatments. To simplify it a little bit, lets just consider strain and treatment. Furthermore, I am running hdWGCNA on a subset of our data that only includes one celltype.
The overall aim was to assess the impact of treatment, but because our groups were:
...I can't entirely assess the best course of action with respect to selecting a group for
SetDatExpr(). On one hand, I took note that in figure 5 of the hdWGCNA you include multiple groups, including your controls, in your network, only subsetting for inhibitory neurons. On the other hand, the strain groups are so vastly different that it generally seems to increase softpower, possibly due to the high variance that is not explicitly specified. To that end, we ended up using Strain2 as our group forSetDatExpr(), however, in the context of already having few cells (anywhere from 2-5k cells depending on the celltype we were analyzing), I am not entirely convinced that was the right thing to do. I am also seeing fewer modules this way, typically due to turqoise module being comically massive by comparison. Nevertheless, the idea was that this would result in a network with less variance and a stronger ability to detect the effects of treatment with DMEThese are just some of my thoughts and an overview of our approach. I wonder if you have some thoughts with respect to this topic? Please let me know if you need any additional details