When we generate a UMAP plot for scRNA-seq, how detailed cluster we can go?

Question

[cabioc]

When I do a UMAP analysis for scRNA-seq, I can change "resolution" to cluster cells to different levels. Cells and cells are different. I think even I only got two cells I can still say they belong to "two" groups. So I would like to discuss how can we make a decision that how detailed or how many clusters we should go?

Answer 1

[jared.andrews07]

Nobody can answer this for you. You can go as high or low resolution as you want, through at some point in either extreme you'll either start to get cells in the same cluster with large differences or split cells based on differences that are exceedingly minor. It depends on the biology of your cells as to what setting may make the most sense and most clearly identify true differences. Try several of them. If it seems like you're getting clusters that are very similar to each other, bring it down a touch. If you have cells that you know are different (based on an obvious marker or such) being clustered together, you can bump up the resolution to try to delineate them.

Or you can just manually slap the label on them if you're just trying to identify cell types. Clustering is just a means to identify differences/similarities between closely related cells - at some point you will likely have to bring your expert knowledge back in to determine whether the differences being identified are meaningful.

[cabioc]

Thank you Jared

I should ask "how do you make decision ...... ", but your answer made sense to me.