Error: Cannot add new cells with [[<-

[bbimber]

Hello,

We've been running scGate on diverse 10x data for a long time with no real issues, but one dataset is posing a problem. I'm trying to run a basic command like this:

seuratObj2 <- suppressWarnings(scGate::scGate(
    data = seuratObj,
    model = model,
    pos.thr = 0.13,
    neg.thr = 0.13
))

and it gives the error:

Error: BiocParallel errors
  1 remote errors, element index: 1
  0 unevaluated and other errors
  first remote error:
Error: Cannot add new cells with [[<-

If I just make a marginal change to pos.thr, then it will pass:

seuratObj2 <- suppressWarnings(scGate::scGate(
    data = seuratObj,
    model = model,
    pos.thr = 0.14,
    neg.thr = 0.13
))

Is it possible something about the order in which pos.thr is applied could result in some chunk of code mangling cell barcodes to something like this?

I'm happy to share the seurat object and model if it is helpful.

The model is:

levels   use_as           name                                                                                               signature
1 level1 positive          NK.RM KLRD1;NKG7;GNLY;KLRB1;CD3D-;CD3E-;CD3G-;CD8B-;LOC710951-;LOC114677140-;LOC711031-;LOC720538-;LOC705095-
2 level2 negative  Complement.RM                                                                                          C1QA;C1QB;C1QC
3 level3 negative    Platelet.RM                                                                                          PPBP;LOC703451
4 level3 negative Erythrocyte.RM                                                                                            HBA1;HBB;HBA

Also here:
https://github.com/BimberLab/RIRA/blob/master/inst/gates/master_table.tsv
https://github.com/BimberLab/RIRA/blob/master/inst/gates/NK.RM.tsv

[mass-a]

Hi,
thanks for the detailed issue. My hunch is that something breaks if, with some parameter combinations, only one cell passes the filters. If you could share the problematic dataset (also by email) and the SessionInfo(), that would be helpful for us to debug.
Thanks!
-m

[bbimber]

@mass-a: sure, thanks for the offer. You can download the file here (it's about 40mb); https://prime-seq.ohsu.edu/_webdav/Labs/Bimber/Collaborations/Public/%40files/scGate/seuratObjError.rds

Can you please let me know when you get it so that I can take down that link? Thanks.

[mass-a]

@bbimber I got the object, thanks. I'll look into this soon.

[mass-a]

Hi again,
I was able to reproduce the error with your data and code.
It's a bit mysterious why this happens only with this particular parameter configuration, but the culprit appears to be one of the dimensionality reductions pre-stored in your object (adt). If I set it to NULL, scGate runs fine:

seuratObj <- readRDS("./data/seuratObjError.rds")
seuratObj@reductions$pca.adt <- NULL
seuratObj@reductions$adt.tsne.pca <- NULL

model <- scGate::load_scGate_model("data/NK.RM.tsv", master.table = "data/master_table.tsv")
seuratObj2 <- suppressWarnings(scGate::scGate(
    data = seuratObj,
    model = model,
    verbose=T,
    pos.thr = 0.13,
    neg.thr = 0.13,
))

Computing UCell scores for all signatures using RNA assay...

Running scGate on level 1...
scGate: Detected 1181 non-pure cells at level 1
Running scGate on level 2...
scGate: Detected 2 non-pure cells at level 2
Running scGate on level 3...
scGate: Detected 1 non-pure cells at level 3

### Detected a total of 14 pure 'Target' cells (1.17% of total)

Can you check it also works for you?

Best
-massimo

[bbimber]

@mass-a I will give that a try. Can you write more about what code within UCell is involved and why those parameters would matter?

This is a wild guess, but does UCell internally divide the seurat object or assay into subunits, which are split? I ask this b/c if you try to subset a seurat 5 assay object (but not seurat 4 assays) down to 1 cell, seurat throws errors.

[mass-a]

Hi,
UCell scores are calculated only once on the complete object, so this cannot be the source of the error.

I debugged individual commands within scGate for your example, and the error occurs when calling subset on a Seurat object composed of 15 cells. The same subsetting command works when I drop the dimensionality reductions, so I suspect something is off with the dimreds. The odd thing is that it happens only with this particular subset of cells (even by subsetting a different selection of cells from the 15 cells); I don't know why.

To avoid further problems, I made a quick patch to scGate to drop unused dimreds in the internal functions; note that this only applied to internal objects, so the returned object after scGate contains all the original dimreds. This is now in the master branch in GitHub.

[bbimber]

Hello,

I was able to narrow this down further and show that if I just set:

seuratObj@reductions$pca.adt <- NULL

then scGate will pass. I cannot see anything wrong with the reduction object. The cells are identical between the reduction and the main object.

thanks for looking into this.

-Ben