This work is underway, and function and input names are all subject to potentially change until merged into devel. But as this also seems to be a major hole in the field, anyone wishing to use/test these functionality early can install from this branch with remotes::install_github("dtm2451/dittoSeq@neighbor-diversity"), then restart your R if you'd already had dittoSeq loaded.

One thing to note, and I'll aim to work this into the documentation, is that the algorithm I'm basing the calcNeighborMetadataDiversity() calculation off of made use of a NearestNeighbors calculation run specifically for this purpose that used k=sqrt(ncells). This number of neighbors may be overkill for gigantic datasets, but the important thing is that using only the 20-or-so neighbors regularly recorded after a "standard" Seurat algorithm run 1) cannot yield quite as accurate of results, and 2) negates some of the utility of using a quantile cutoff on neighbor distances due to the relative inflation of such distances.

So:

• adjust documentation to recommend running neighbors calculation with higher number of neighbors retained
• adjust defaulting to 1) warn and 2) not use the distance quantile trimming unless the number of neighbors in neighbors is at least 30.