Inconsistent results when supplying cellTypes as factor vs character to toSF()

[Liuy12]

Thanks for developing such a useful tool! I encouter some issues when analyzing my curio-seeker data.

• I see significant differences in downstream results (makeShuffledCells / findTrends) depending on whether I pass a factor or a character vector to toSF().
• Expected: identical analysis when labels are the same (factor vs character with same values).
• Actual: results differ. Suspected cause: character → factor conversion inside toSF() creates alphabetical levels (factor ordering) which changes processing order when generating the shuffled list.

Reproduction

# factor input (preserve Seurat factor levels)
slide_f <- crawdad:::toSF(pos = coord,
                          cellTypes = seurat_obj$first_type)

# character input (explicitly converted to character first)
slide_c <- crawdad:::toSF(pos = coord,
                          cellTypes = as.character(seurat_obj$first_type))

scales <- c(50, seq(100, 1000, by = 100))
ncores <- 10
seed <- 1
perms <- 3

shuffle_f <- crawdad:::makeShuffledCells(slide_f, scales = scales, perms = perms, ncores = ncores, seed = seed)

shuffle_c <- crawdad:::makeShuffledCells(slide_c, scales = scales, perms = perms, ncores = ncores, seed = seed)

res_f <- crawdad::findTrends(slide_f, neighDist = 80, shuffleList = shuffle_f, ncores = ncores, verbose = FALSE, returnMeans = FALSE)
res_c <- crawdad::findTrends(slide_c, neighDist = 80, shuffleList = shuffle_c, ncores = ncores, verbose = FALSE, returnMeans = FALSE)

I also see similar issues when randomly re-level the cell type variable supplied to toSF(). I can share my data if needed. Please let me know your thoughts.