I was wondering if SoupX performs best when the clusters are biologically meaningful, or whether the method should still perform well using unsupervised clustering (such as Louvain/Leiden graph clusters). Would it improve SoupX's estimation to manually identify and merge clusters from known cell types, as compared to using the raw clustering results?
It seems this would all revolve around the "quickMarkers" step. For example, some clusters may end up having few or no good markers if they represent poor quality cells.
Thanks for your work on this tool, and looking forward to hearing any advice you might be able to offer,
Hi @constantAmateur,
I was wondering if SoupX performs best when the clusters are biologically meaningful, or whether the method should still perform well using unsupervised clustering (such as Louvain/Leiden graph clusters). Would it improve SoupX's estimation to manually identify and merge clusters from known cell types, as compared to using the raw clustering results?
It seems this would all revolve around the "quickMarkers" step. For example, some clusters may end up having few or no good markers if they represent poor quality cells.
Thanks for your work on this tool, and looking forward to hearing any advice you might be able to offer,
Patrick