fix per sample custom config

[alexvpickering]

Linked Issue

Also fixed in data-ingest: https://github.com/biomage-ltd/data-ingest/pull/34

Background

The function add_custom_config_per_sample wasn't subsetting the dataset. This created a bug where the entire object was being used to calculate default cellSizeDistribution, numGenesVsNumUmis and doubletScore thresholds. For example all the doubletScore thresholds end up the same:

This is the same value as if I calculate the threshold with the full seurat object:

Whereas e.g. for an actual sample subset we get something different for each sample:

[JuanluOnieva-biomage]

It's weird because so far it has been working well. Moreover, if it gives wrong parameter for numGeneVsnumUMIs it should also fail for the cellSizeDistribution. What could happen is that since the numGeneVsnumUMIs' config parameter depends on the number of cells (p.level = 1/numCells), in the case that of larger experiments with similar size across samples we will have similar value.

Nevertheless, I like the refactoring. I think it is better to use the subset function (which belongs to the SeuratObject package), however filtering by columns seems to give the same result so I leave it to your choice and I approve the PR.

[alexvpickering]

It seems to have been a bug introduced by the fact that we have both sample and samples as meta.data columns:

# samples was unique samples
samples <- unique(scseq$samples)
samples
# [1] "T0_1"   "T0_2"   "Ctrl_1" "Ctrl_2" "Gent_1" "Gent_2"

dim(scseq)
# [1]  18689 267867

dim(subset(scseq, samples %in% "T0_1"))
# [1] 18689  2103

sample <-  "T0_1"
dim(subset(scseq, samples %in% sample))
# [1]  18689 267867

I'm a fan of the slice notation because it's universal for R objects and clear when your subsetting columns/rows. Also not going to have this bug lol :)