What should be control group for differential expression in single cell? either distant normal samples or primary tumors when analysing metastatic samples.

Question

[gorizwango]

I'm just stuck with my differential expression in my single-cell project. My dataset has 58 samples, and about 10 of them are distant normal while remaining constitute primary tumors, metastatic brain, metastatic lymph, etc. In many studies, people perform differential expression of one cluster against other clusters (or average expression of all remaining clusters). As in my case, samples are very heterogenous, what strategy should be applied for the differential expression analysis, either I plot clusters of all the samples at once and do differential expression cluster by cluster, or should I design my study in parallel runs for normal vs tumor, normal vs brain metastatic, normal vs lymph metastatic.....??

[ATpoint]

I say this with all due respect but you and only you can answer this as this is the very core of the scientific question you're working on. People here can help with the how but not with the what/why. You must have a scientific question based on which this dataset was created and which is why you're doing the analysis. If you cannot answer this then you have to talk to your PI/committee to define your project better.

[gorizwango]

Thanks for taking your time to answer my query. I just need a little explaination of different scenarios where a particular strategy will be employed for differential expression, for example in what conditions do I have to go with cluster vs cluster, cluster vs sample, or sample vs sample?

[ATpoint]

Again, this depends on the question. Nobody here knows which celltypes sre in there and whether these are relevant for your project and whether that comparison holds potential to find novel biology. Try to discuss with your peers what the overall goal is and whether clusters hold relevant information that merits a pairwise comparison.