I notice that neighbourhood groups will make groups of different cell types if cell types are similar. Might Milo accept some per-cell labelling from a method such as SingleR to set up a boundary not to be crossed for neighbourhood merging? For example,
The function is purely for grouping neighbourhoods based on the neighbourhood adjacency matrix. It does not aim to recapitulate single-cell clustering/annotation.
I notice that neighbourhood groups will make groups of different cell types if cell types are similar. Might Milo accept some per-cell labelling from a method such as SingleR to set up a boundary not to be crossed for neighbourhood merging? For example,
Source: Cancer-associated Fibroblast Classification in Single-cell and Spatial Proteomics Data, Nature Communications
The graph of such a data set would be highly connected, meaning that heterogeneous neighbourhood groups are formed.
A split plot shows the true extent of the overlap (in two dimensions) and rampant overlplotting of dots in the overall plot.