What is the proper way to identify the cell types of scRNAseq cluster?

Question

[MAPK2]

Greetings everyone,

I am currently using Seurat for scRNAseq data analysis in R. After analyzing treatment versus control, I have identified 16 distinct clusters. Now, I need to label these clusters based on differentially expressed genes and determine their corresponding cell types using marker genes. My question is regarding the optimal number of marker genes to consider. Should I focus on the top 4 genes, top 20 genes, or all significant genes? Additionally, if the markers exhibit higher expression in two different cell types in GTEx, how can we ascertain the correct cell type?

Thank you.

Answer 1

[ATpoint]

I will say that the proper way is the one that confidently identifies your celltypes.

optimal number of marker genes to consider.

That really depends. If you have well-separated clusters like in PBMCs it is often enough to look at two or three canonical markers. On the other hand, in a tight continuum you might want to look at more markers.

Should I focus on the top 4 genes, top 20 genes, or all significant genes?

I would go through the marker lists and just see what makes sense.

Additionally, if the markers exhibit higher expression in two different cell types in GTEx, how can we ascertain the correct cell type?

What does this mean? You mean ambiguous markers? Seurat marker definition is arguably poor. It just tests cells in cluster A versus all other cells, so basically some sort of average. Don't hang yourself up on that. It's a marker enrichment but they're all but specific in the way Seurat computes them.

If you have a reliable reference maybe use something like SingleR to get an idea.