How to remove center population from seurat cluster

Question

[synat.keam]

Dear all,

apologize for many posts. I have done my clustering, but there is weird tidy population (in red circle) that is not stay close to its assigned cluster (kindly see attach). In the attached imaged, that cluster has been assigned to cluster 15. I am wondering whether there is any tool that I could omit/take out that population (one in red circle) from the UMAP? Really appreciate your input.

Kind Regards,
Synat,

[ATpoint]

UMAP and graph-based clustering do not always agree. I would just ignore this and continue. The clustering is what it is, don't introduce bias because you don't like the look of it.

[jv]

To add a bit more: it's a common misconception that the UMAP and cluster results are directly related but they are not.

If OP hasn't already, they can do some optimization of their UMAP reduction using scDEED https://doi.org/10.1101/2023.04.21.537839

[synat.keam]

thank you, ATpoint. really great insight!

[Hamtaro]

First of all, you must identify the cells in the outlying population. You can do this by looking at the coordinates of the cells within the UMAP plot and determining which cells fall within the red-circled area.

Once you have the identities of these cells, you can create a new Seurat object that excludes them using the subset function.

In some point should be like:

# Assuming `seurat_obj` is your existing Seurat object and `cells_to_remove` is a vector of cell names to remove

# Remove the cells
seurat_obj <- subset(seurat_obj, cells = !colnames(seurat_obj) %in% cells_to_remove)

# Recalculate the UMAP with the cells removed
seurat_obj <- RunUMAP(seurat_obj, dims = 1:20)

# Plot the new UMAP
DimPlot(seurat_obj, reduction = "umap")

[fracarb8]

You can easily select cells_to_remove by running CellSelector(DimPlot(seurat_obj)) and drag the mouse over the cluster of cells you want to remove.

[ATpoint]

Sure you can but you shouldn't. It's not a reproducible criterium to say "there is a little splash on the UMAP I don't like". Plus, unless you use the UMAP for something other than visualization it really does not matter. Play with UMAP parameters, maybe the splash just moves when you alter spread and min_dist parameters.

[fracarb8]

agree, but if the user is confident and wants to do it, that's a way. Another possibility is to increase the resolution of the clustering and see if/when it clusters on its own.

[synat.keam]

Thanks, I agreed with you all. I will discuss with my PI before I make a decision to remove or not. Thank you all brothers.

[synat.keam]

Dear Fracarb, thank for your help. Really great tool. I have tested it and after I dragged around cells I wanted to remove and clicked done, I did not get a new seurat objec, but a value instead. Do you know after clicking done, how could I proceed to get a new seurat obj? sorry for asking again.

[synat.keam]

thanks, i have just read seurat tutorial and found it. thanks you