Cell type is lost during annotateNhoods

[carlacohen]

Cell type is lost during annotateNhoods

[MikeDMorgan]

You would need to reduce k if these cells are not well separated in the NN-graph, bearing in mind the need to balance sufficient counts in each nhood for the DA testing.

[carlacohen]

Hi @MikeDMorgan yes I was able to tweak the parameters and create neighbourhoods that represented the missing cell type. Thanks for your help

[MikeDMorgan]

Hi @carlacohen - the neighbourhood annotation is based on a majority vote of cells in each neighbourhood - therefore, if there if the missing cell types are in nhoods with other cell types, but don't reach a majority, then they won't be observed at the nhood level. You should be able to see this in the output of annotateNhood in the {variable}_col_fraction column.

You can double-check this yourself, but computing for each nhood the proportion of each cell type, and see that the missing cell type is never in the majority.

[carlacohen]

Thanks for the quick reply! That makes sense, I guess this is a caveat of having a very rare population of cells in the analysis.
Do you think reducing the size of neighbourhoods would help or will I never really be able to pull them out (the myofibroblast cluster only comprises 341 cells, and there are ~25k cells in the object)?

[MikeDMorgan]

Hi @carlacohen did this fix your issue?