A cluster expresses two types of cells' markers in scRNA-seq

Question

[feather-W]

Hi, I met a question when I analyzed scRNA-seq data. I found a cluster expresses the markers of myeloid cells and B cells. And in this cluster, cells expressing myeloid marker did not overlap with cells expressing B cell marker. The images as follows:

The markers of myeloid cells:

The markers of B cells:

And I think the cluster 10 is not doublet because its nCount_RNA and nFeature_RNA are normal.

So why does this happen, and what should I do to annotate this cluster.

Can anyone help me or give me some advice?

Thanks

[ATpoint]

I see no cluster borders. Is 10 the whole plot? Myeloid markers seem to be more in this wing to the left and lymphoid markers rather go in upward direction. Run a dedicated doublet detector to check this possibility.

[feather-W]

Hi ATpoint, thanks for your help!

The whole plot as follows:

And I think cluster10 is not doublet, because the nCount_RNA and nFeature_RNA as follows:

As you can see, both nCount_RNA and nFeature_RNA of cluster 10 are low. So it is most likely not doublet.

And now I don't how can I annotate cluster 10.

Answer 1

[bk11]

As ATpoint already mentioned, you should run a dedicated doublet finder to check the possibility of any doublets in your data. Can you tell what type of tissue is it? It looks like you are trying markers based annotation to call cell types in your data. I would recommend you to use CellTypist for annotation of your data if you have not tried it before.

https://www.celltypist.org/#header

[feather-W]

Hi, bk11, thanks for your help! And I will try it.

Answer 2

[jared.andrews07]

Without any additional context as to what the sample actually is, my assumption is that those are monocytes interacting with B cells. I'd third the recommendation to run a doublet detector that will simulate multiplets from your various clusters (like scDblFinder).

While genes detected tends to scale with doublets, that's not always the case, and it's worth a quick check using orthogonal approaches.

It's also important to note that clustering is a mildly useful lie most of the time - it can and will vary widely depending on the parameters used, clusters don't actually guarantee biological differences, etc. If you just want those two populations to be clustered separately, just re-run your clustering with increased resolution.

[feather-W]

Hi, jared.andrews07. Thanks for your advice! And I will try it.