How to do sth like contrast for cell level DE analysis

[pariaaliour]

Hello Muscat team,
Thanks for this helpful package. As I did not have DE doing Pseudo bulk DE analysis I'm trying to run DE at cell level. In the tutorial, it does not clearly explained the steps. I am not sure how to account for some variables and also how to do a specific comparison like contrast in pseudo bulk analysis. For example, I have a variable which has 4 different levels: oculas, ocucon, medals, medcon. I want to do the analysis in two round and do two comparison: ocular vs ocucon and medals vs medcon.
I. appreciate your help on this!
Paria

[HelenaLC]

Thanks for the kind words! -- Hm, I would be cautious to use cell-level approaches under the premise that you "did not have DE doing pseudobulk". We found that, e.g., mixed models, are less performant than pseudobulk-based approaches (lower TPR, potentially higher FDR) here. This has been "confirmed" in another study here using paired bulk and single-cell data.
Perhaps you could clarify your current experimental design (e.g., sample, groups, other) / post a table of it, how your current design matrix and contrasts are specified when running pbDS etc., or post some relevant code snippets of your current analysis. At least then we could first rule out that there's nothing off with the current (lack of significant) results.

[pariaaliour]

Thanks for your swift response. Sure, I have two groups, Case and control. For each individual I have two regions (1 and 2). I would like to compare region 1 and 2 in cases. But to remove tissue specificity I integrated all samples (case and control) and do following analysis: 1. DE for region 1 in case vs control, 2. DE for region 2 in case vs control
After getting two sets of DEGs I would like to. correlate those DEGs to see how divergent those are.
It might be helpful also to share with you my code:

First contrast:

second contrast


FYI, the group_id is the combination of two interested variables (region and status): caseregion1, caseregion2, controlregion1, controlregion2.

I also appreciate if you look at my code to see if I'm doing everything correctly.
Hope it is clear.
Many thanks,
Paria