nFeature histogram 2 peaks both higher than 1000

[AnaOliviaFernandes]

Hello everyone! I am currently performing individual sample QC of snRNAseq data and I have a question related to filtering. I saw multiple discussions addressing the problem with having 2 peaks in the nFeature plot, but normally the first peak always corresponds to <500 nfeatures and it is evident that should be excluded due to clear indication of low quality cells. What about when I have 2 peaks at nFeature>1000?

Here are the vlnplots of my sample:

And here is the histogram of nFeature:

hist(sample2.23@meta.data$nFeature_RNA)
abline(v=500, col="red")
abline(v=7500, col="red")

I am not sure how to filter this sample. Is the second peak correspondent to doublets and should be excluded?
I already run DoubletFinder and it does not exclude the majority of the cells belonging to the second peak...

I would appreciate any feedback the community can give me.

Many thanks

[YoukaiFromAccounting]

Hello @AnaOliviaFernandes,

At first glance I don't think the peak around 3k features corresponds to doublets, it could just be the experimental data itself. Some additional info would help get to the bottom of this issue:

What tissue or system are these cells from?

Is this sample expected to be one fairly homogeneous cell population, or a mix of different cell types?

If you run a basic clustering and look at UMAP or tSNE, do the cells in the higher nFeature_RNA peak form their own clusters, or are they mixed in with the rest?

It might also help to replot nFeature_RNA on a log scale, for example:
VlnPlot(sample2.23, features = "nFeature_RNA", log = TRUE)
or
hist(log10(sample2.23@meta.data$nFeature_RNA))

This sometimes makes it easier to see whether the peaks are really separate populations or part of a continuous distribution.

[AnaOliviaFernandes]

Hello @YoukaiFromAccounting

First of all thank you for your fast response.
We are talking about brain tissue here! If i understood correctly you suggest that the second peak is just a different cell type? And in that case i should not exclude those cells.

In any case, this is how the Umaps look like:

High nfeature_RNA peak corresponds to one cluster, but that cluster is also the one where we can find most doublets... Any thoughts on this?

[YoukaiFromAccounting]

Hello @AnaOliviaFernandes,

If your sample consists of brain tissue, I'm thinking the multiple nFeature peaks are a result of tissue heterogeneity (ie: neurons vs glia). Based on your nFeature_RNA UMAP and the fact that your doublets are not clustering out of real cells, its possible that the "doublets" in your dataset are mislabeled and are actually singlets, which DoubletFinder tools are classifying as doublets due to the presence of many cells with low RNA counts. I think you should be safe to proceed with clustering and annotation keeping your current cutoffs.

Hope this helps!

[AnaOliviaFernandes]

It was extremally helpful! I think I will proceed with this cutoff.
Thank you so much for your input!