How to label clusters from a UMAP

Question

[pairedttest]

So this is a broad question that I was hoping someone could shed some light on.

I have a Visium tissue data set which I've clustered using scanpy, and for each tissue slice I usually get 5 or 6 resulting clusters. I can also see which genes in each cluster serve as the markers. My PI wants me to figure out which of the 7 cell types in the tissue we are working with corresponds to each cluster and assign it to each of the clusters so that we can do downstream analyses. However, I don't know how this could be done. It seems entirely subjective to assign a cell type to each cluster has each cluster probably contains multitudes of different cells, and looking at marker genes seems entirely subjective as well and not at all clear. I've already ran other analyses such as Cell2Location that predict the proportion of each cell type in the spots. So I'm not really sure what I'm being asked to do here.

Answer 1

[benformatics]

Seems like a poor premise to begin with (as you say). You are correct in that the spots in spatial transcriptomics generally correspond to a mix of cell types. Unless a certain region is specifically only 1 cell-type than this doesn't really make sense...

I guess you could try using the "majority" for a given cluster to assign it to a cell type. You would need some reference marker genes for this or could try using an existing "reference" from Azimuth (hopefully your tissue of interest + species is already there - otherwise you can create your own reference).

You can run Azimuth via their website or locally following this guide: https://satijalab.github.io/azimuth/articles/run_azimuth_tutorial.html