Question about DME analysis interpretation

[thgirb-matt]

Thank you so much for this wonderful package, the tutorials have been a massive help, but I have a question as I have never done hdWGCNA or even WGCNA analysis before.

Sorry if this is a silly question but I am not 100% on what the output of the DME is telling me? If I run DME as so:

DMEs <- FindDMEs(
hdwgcna_obj,
barcodes1 = group1,
barcodes2 = group2,
test.use='wilcox',
wgcna_name='hdwgcna'
)

Does a positive log2FC reflect greater expression of the genes in the module in group1 or a greater connectivity/stronger co-regulation of the genes in group1?

I have performed the hdWGCNA using control cells in my dataset and identified 12 modules. Most of the modules seem to just reflect cell-type specific expression networks (i.e. microglia, astrocytes) (this is probably because the metacells were generated by bulking all cells from each donor, not just one specific cell type?). I wanted to see if there were differences in the modules between the control cells I used to find the modules and cells from Alzheimer's donors. I set group1 to AD cells and group2 to control cells. Now I am uncertain how to interpret the results.

Thank you for any advice.

[smorabit]

Hi,

To clarify, this is akin to differential expression analysis, so when you see a positive log2FC, that module ie expressed higher in group1 compared to group2. This doesn't explicitly tell us anything about connectivity in the networks, or regulation.

I have performed the hdWGCNA using control cells in my dataset and identified 12 modules. Most of the modules seem to just reflect cell-type specific expression networks (i.e. microglia, astrocytes) (this is probably because the metacells were generated by bulking all cells from each donor, not just one specific cell type?).

In your case, you might have more success if you perform separate analyses for each cell type of interest. At the start of the pipeline you can completely subset one cell type, and then proceed to analyze it with hdWGCNA. This is similar to what we did in Figure 6 of the hdWGCNA paper by analyzing microglia only.

[thgirb-matt]

Thank you for clarifying, that was my assumption but I just wanted to check. I have completed the analysis again using only the expression from the microglia in my dataset to generate the network and have got some good modules out. There is still one surprising result coming out of the DME analysis but a closer look may make it clear why that is.

In a similar manner I wanted to double check how the order of levels used for the ModuleTraitCorrelation influences the results. I cannot find documentation that makes it clear.

If I want to look at correlation between health and AD and the output of running ModuleTraitCorrelation says:
Trait condition is a factor with levels Control, AD. Levels will be converted to numeric IN THIS ORDER for the correlation, is this the expected order?

Does this mean it is setting Control as 0 and AD as 1 and therefore a positive correlation would mean correlated with AD status? Or is it like the differential expression where group1 (the one converted to numeric first) is the test condition and group2 the baseline? When I flip the order around (levels AD, Control) the correlations obviously flip, so I just want to confirm which direction is the correct way round to run it.

Thanks!!!