I see where you're coming from. But that doesn't seem to me to address why not to do multiple UMAPs. To make my train of thought more explicit:
- Like you said, because of biological and technical variability, UMAP clusters can get a little loose
- If you include cells with different treatments in the same UMAP, there's more variability, and your clusters will be looser and thus less clean
- Therefore, if you made a UMAP for each treatment, you'd decrease the variability within each UMAP, get cleaner clusters with fewer cells misclassified, and as a result your downstream analysis would be more accurate
I know nobody does this so I'm sure there's some flaw in my above logical train, I just want to know what it is
"If you're expecting cell types to transcriptionally change in response to treatment, wouldn't it make more sense to run UMAP on just cells that share the same treatment?"
No, please don't do this. If a population of Th2 cells changes in expression between immunotherapy treatment and control, why would you put them on different plots? How will you see that they're different?
You could put them on different plots because that would increase the purity of each cluster, because mixing Th2 cells that received different treatments would give you larger, more heterogeneous clusters that are more likely to include non-Th2s by mistake. Once you have Th2 clusters from the different UMAP plots, you could do, say, differential expression analysis between those two groups of cells to see any differences.
Because nobody does this, I'm sure there's a reason not to do it, I just want to know what that reason is.