plotUMAP_split() %>% ggarrange(common.legend=TRUE) produces inconsistent colors

[micahpf]

When using the plotUMAP_split() %>% ggarrange(common.legend=TRUE) workflow to generate umaps across split object, if you group.by a metadata column labels which has different number of unique(labels), the colors will be inconsistent across umaps and the common legend will thus not correspond to the true colors.

I tried assigning the superset of levels to each sample (below), but that does not fix the problem.

common.levels <- obj.split %>%
  lapply(., function(x) x$labels %>% unique()) %>%
  Reduce(c, .) %>% unique()

obj.split <- obj.split %>%
  lapply(., function(x) {
    x$labels <- factor(x$labels, levels = common.levels)
    return(x)
  })

I think Seurat's DimPlot() is removing unused levels from the groups used to color cells, essentially ignoring the factor levels?

Maybe there's some way to revise the color scale of each DimPlot after calling Seurat::DimPlot() to include the full range of levels? Otherwise we may need to use our own version of DimPlot .