Treatment VS Control in Single Cell RNAseq analysis

Question

[grayapply2009]

I have scRNA-seq data for 4 groups Treatment 1, Control 1, Treatment 2, Control 2.

My goal is to compare gene expression in each cell type between Treatment 1 and Control 1, between Treatment 2 and Control 2, and between Treatment 1 and Treatment 2.

Should I perform QC, Normalization, and clustering on the combined 4 scRNA-seq datasets and then do DE analysis in each identified cluster?

Any suggestions?

Answer 1

[ATpoint]

I suggest you go through https://osca.bioconductor.org/ extensively before diving into analysis. Read all paragraphs, it answers 99.9% of all standard questions.

• QC, initial normalization and selection of informative genes typically is done per dataset separately.
• Then an integration step to merge all datasets, e.g. with fastMNN or the Seurat anchoring framework.
• clustering based on the integrated dataset
• DE analysis between clusters based on the unintegrated dataset with the clusters defined with the integrated dataset

For an extensive discussion see the linked workflow. See also C: Batch correction in scRNA-seq data via fastMNN | pro/contra

[grayapply2009]

Thank you for your prompt reply, ATpoint. What do you mean by "DE analysis between clusters based on the unintegrated dataset with the clusters defined with the integrated dataset"? I'm interested in the DE analysis between treatment groups and control groups in the same cluster. I don't care about the comparison between different clusters except when I'm identifying cluster marker genes.

[ATpoint]

Sorry, misread the question. The point is the same though. You have to use the unintegrated values for DE analysis, not the ones that the integration produces, see https://osca.bioconductor.org/multi-sample-comparisons.html#sacrificing-differences

[grayapply2009]

Thank you, my friend. That was helpful. I'm reading it now.