use AverageExpression to stratify samples

Question

[ndra1456]

I have a group of cancer patient samples for scRNA datasets. I need to stratify my samples based on the expression level of one gene (low vs high) in the tumor cells and examine their T cell components.

Is it reasonable to extract the tumor cell cluster and rank the gene expression level using AverageExpression function? I will choose the 1st and 4th quartile or the top and bottom half for following comparisons.

Not sure it is a bias method and it seems hard to find literatures with similar question.

[ATpoint]

You can do that, but more information would be necessary to explain your open-ended question more. Why do you think it biases analysis? As usual, scRNA-seq and bioinformatics in general is all but standardized, so you almost never find the one paper of vignette that 100% justifies what you want to do.

[ndra1456]

It concerns me because the AggregrateExpression function is also known for pseudobulk analysis and I am not sure which is the best way.

Also, considering the way of finding differentially expressed gene using FindMarkers, I am curious if I can rank the samples based on the percent expression of the gene in each sample, like they are clusters. That will give me another result too.