I have 2 scRNA-seq data sets which come from different treatment conditions (Drug vs. Control)
I have been using Seurat and running the data through their standard integration pipeline, and at the end can get a result of DEGs between each condition for each cluster of cells.
What I am exploring is if its possible to obtain a fold-change for a given gene on a per-cell basis. For example, I would like to take the expression of gene "X" in cell #1 of cluster 0 for the Drug group, and compare it to all the cells in cluster 0 of the Control group and get a quantitative measure of how the expression in this one cell of the drug group is vs. the entire population of control cells.
Is something like this possible? I feel like it must be since the algorithm is essentially computing this same thing only in batch for all the cells in the cluster right? But Im afraid I can't think of a way to approach this?
Thanks in advance
Sure, just compare the count for your cell of interest with that of the avg. count for a cluster. I do question _why_ you'd want to do this though. Why not just compare between the clusters?
I think I wanted to look and see visually what the spread looked like within that cluster. What I mean is that for a given cluster and a given gene, just how homogenous is the effect of drug treatment across all cells in that cluster.
I think I had in mind to try plotting it in a heat map and even ordering the cells highest to lowest that respond within that cluster, and then look at how those high responders are the same/different with respect to genes that correlated with the one im looking at (maybe ones downstream in the same pathway)
I'd probably just plot violin or box plots for each cluster then.
Thanks for that suggestion!
Also, I was going to use the raw counts for this, but I don't understand why it would be incorrect to use normalized counts in this particular instance? One thing I dont understand still is when its appropriate to use raw vs. normalized counts
There is nothing wrong with using normalized counts in this case. They shouldn't really look much different unless you have pretty dramatic confounding effects that you tried to regress out (like if one of your clusters is composed almost entirely of cells in S phase whereas the other one is not, and you regressed out differences due to cell cycle phase).