feature request: clustering from within cellxgene

[LisaSikkema]

Hi,

I'm working on a very large integration of many datasets, and me and my collaborators use cellxgene to explore the data. One thing we would really like, is the option to re-cluster subsets of cells from within cellxgene. This is for example useful when we're looking into specific clusters, and realise there might be certain cell subtypes in there. Clustering would be a means to objectively subdivide your cluster/cell group of interest into sub-groups, and see if your cluster can indeed be subdivided into the subtypes you had in mind. Right now, to do that, I'd have to go back to my analysis notebook, load the h5ad, cluster, store the object, and re-load cellxgene every time.

One can of course check if possible subtypes in a cluster are distinctly mapped in the umap, but in some cases that is not the most robust way to judge similarity or distance. Moreover, having the clusters allows one to add annotation to these clusters rather than to cells circled in the UMAP, and annotations based on clusters can actually be used in follow-up analyses (I would never do this with annotations based on UMAP lassos.)

Something to consider for a future update!

Thanks,

Lisa