singleR annotation shows overlapping clusters

Question

[paria]

Hello everyone, I used singleR to annotate my clustered Seurat dataset. Before annotating using singleR I did manual annotation using marker genes. Below is the result of my manual annotation

Then, I used singleR and below is the annotation results.

The result shows that I can trust my manual annotation. However, I see overlapping clusters in the last annotation. For examples, I have some oligodendrocytes clustering with microglia coloured in blue and some are clustered with Astrocyte. I was wondering what it means and what is the best approach here if I want to use the second annotation umap plot. Should I get rid of those oligodendrocytes clustering with other cell types? or any other recommendation? I appreciate any comment on this.

Answer 1

[jared.andrews07]

Hard to say for certain, as it's going to depend quite a bit on which reference(s) you are using, how you're running SingleR, etc. Initial inclination is to say that either your dataset has some doublets or the reference dataset isn't annotated with super high granularity/also has doublets. SingleR is only as good as the reference dataset(s).

Use of scDblFinder or scrublet or similar tools will spit out doublet predictions in your data (which should be done before QC filtering as it's based on an assumed doublet rate - 1% per 1000 cells in 10X data). Overlaying those predictions with your "mixed" populations may be illuminating, as might be looking at astrocyte/oligo markers in them.

I've sometimes found it helpful to run SingleR in cluster mode while heavily overclustering. This helps to reduce some of the messiness/mixing while still providing decent granularity, though it does still make the assumption that each cluster is only a single cell type (which is not necessarily true). At some point, you have to decide what tradeoffs between interpretability and heterogeneity you're willing to make.