Default random seed in WhichCells prevents random sampling with []

[nicolaromano]

Problem

Wanting to randomly subset a Seurat object, one could do

library(Seurat)

set.seed(12345)
for (i in 1:5)
{
    ten_cells <- pbmc_small[, sample(Cells(pbmc_small), 5)]

    print(Cells(ten_cells))
}

This prints

[1] "CATGAGACACGGGA" "CGTAGCCTGTATGC" "ACTCGCACGAAAGT" "CTAGGTGATGGTTG" "TTACGTACGTTCAG"
[1] "GATAGAGATCACGA" "GGCATATGCTTATC" "ATGCCAGAACGACT" "AGATATACCCGTAA" "TACAATGATGCTAG"
[1] "GATAGAGATCACGA" "GGCATATGCTTATC" "ATGCCAGAACGACT" "AGATATACCCGTAA" "TACAATGATGCTAG"
[1] "GATAGAGATCACGA" "GGCATATGCTTATC" "ATGCCAGAACGACT" "AGATATACCCGTAA" "TACAATGATGCTAG"
[1] "GATAGAGATCACGA" "GGCATATGCTTATC" "ATGCCAGAACGACT" "AGATATACCCGTAA" "TACAATGATGCTAG"

Expected behaviour

The five rounds should all return different cells, while it always (but for the first round) returns the same cells.

Issue

Credits to wurli on Stackoverflow for figuring out the problem.

Seurat overloads [] to call subset.Seurat() which in turn calls WhichCells(), which has a default seed parameter of 1.
The first time our seed 12345 works (hence the first sample being different), but for the following rounds it is overwritten to 1.

Workaround

library(Seurat)

set.seed(12345)
for (i in 1:5)
{
    ten_cells <- pbmc_small[, sample(Cells(pbmc_small), 5), seed=NULL]

    print(Cells(ten_cells))
}

This prints

[1] "CATGAGACACGGGA" "CGTAGCCTGTATGC" "ACTCGCACGAAAGT" "CTAGGTGATGGTTG" "TTACGTACGTTCAG"
[1] "CATGGCCTGTGCAT" "TTACGTACGTTCAG" "ACAGGTACTGGTGT" "AATGTTGACAGTCA" "GATAGAGAAGGGTG"
[1] "CATTACACCAACTG" "GGCATATGCTTATC" "ACAGGTACTGGTGT" "CATCAGGATGCACA" "ATGCCAGAACGACT"
[1] "GAGTTGTGGTAGCT" "GGCATATGGGGAGT" "AGAGATGATCTCGC" "GAACCTGATGAACC" "GATATAACACGCAT"
[1] "CATGAGACACGGGA" "GGGTAACTCTAGTG" "TTTAGCTGTACTCT" "TACATCACGCTAAC" "CTAAACCTGTGCAT"

Proposed solution

Not sure if removing the default seed with break other things, but this should probably be better documented as calling WhichCells overwrites the user seed, which is definitely unwanted behaviour.

[nicolaromano]

Another example of this behaviour

library(Seurat)

set.seed(12345)
print(rnorm(1)) # 0.5855288

set.seed(12345)
print(rnorm(1)) # 0.5855288

set.seed(12345)
id2 <- WhichCells(pbmc_small, idents = 2)
print(rnorm(1)) # -0.6264538

set.seed(12345)
pbmc1 <- subset(pbmc_small, idents = 1)
print(rnorm(1)) # -0.6264538

set.seed(1)
print(rnorm(1)) # -0.6264538

# Other Seurat functions behave as expected
set.seed(12345)
DimPlot(pbmc_small)
print(rnorm(1)) # 0.5855288