subset(downsample= X)

[chrismahony]

Can you tell me, when I use the downsample function, how does seurat exclude or choose cells?

Thanks
Chris

[yuhanH]

Hi,
You can set invert = TRUE, then it will exclude input cells.
For example

select.cell <- WhichCells(pbmc_small, idents = 0)
pbmc_small <- subset(x = pbmc_small, cells = select.cell, invert = TRUE)

[chrismahony]

Thanks for this, but I really want to understand more how the downsample function actualy works. If I have an input of 2000 cells and downsample to 500, how are te 1500 cells excluded? What pareameters are excluding these cells? Thanks

[yuhanH]

downsample is an input parameter from WhichCells

Maximum number of cells per identity class, default is Inf; downsampling will happen after all other operations, including inverting the cell selection

[bug1303]

It's a closed issue, but I stumbled across the same question as well, and went on to find the answer.

You can see the code that is actually called as such: SeuratObject:::subset.Seurat, which in turn calls SeuratObject:::WhichCells.Seurat (as @yuhanH mentioned).

It first does all the selection and potential inversion of cells, and then this is the bit concerning downsampling:

    cells <- CellsByIdentities(object = object, cells = cells)
    cells <- lapply(X = cells, FUN = function(x) {
        if (length(x = x) > downsample) {
            x <- sample(x = x, size = downsample, replace = FALSE)
        }
        return(x)
    })

So indeed, it groups it into the identity classes (e.g. clusters or whichever idents are chosen), and then for each of those groups calls sample if it contains more than the requested number of cells. So, it's just a random selection.