Error: In function [: "j" must not have duplicated values

[alasaa]

When I use 'DoubletFinder' with seurat5 + BPCells, I got the Error:

sweep.res.list_obj <- paramSweep(obj, PCs = 1:40, sct = FALSE)
[1] "Creating artificial doublets for pN = 5%"
Error: In function [: "j" must not have duplicated values
▆

1. └─DoubletFinder::paramSweep(obj, PCs = 1:40, sct = FALSE)
2. └─base::lapply(...)

3. └─DoubletFinder (local) FUN(X[[i]], ...)
   

4.   ├─data[, real.cells1]
   

5.   └─data[, real.cells1]
   

6.     └─BPCells:::selection_index(j, ncol(x), colnames(x))\
   

packageVersion('DoubletFinder')
[1] ‘2.0.4’
packageVersion('Seurat')
[1] ‘5.0.1’
packageVersion('BPCells')
[1] ‘0.1.0’

Previous posts have raised similar questions, but ours seems to be different?
#161

Although the authors said "DoubletFinder should work with Seurat V5 now", I still cannot use 'DoubletFinder' with seurat5 + BPCells.
Is it my problem? Could someone help me, please?

[veeramv]

Re-install Double finder its updated >2.0.4 it works well now :)
remotes::install_github("chris-mcginnis-ucsf/DoubletFinder")

good luck :)

[ggruenhagen3]

I'm facing the same issue. I'm using DoubletFinder v2.0.4, BPCells v0.1.0, Seurat v5.0.0, and R 4.3.1. From the error it seems that the issue is that DoubletFinder is trying to subset by a list of cells that contains a single cell multiple times, but BPCells does not allow indexing that way. Seems like this issue might require attention from the DoubletFinder and/or BPCells team.

As a temporary solution, I found that converting your assays to the old versions allows DoubletFinder to work.

obj[["RNA"]] = as(obj[["RNA"]], "Assay")
obj[["SCT"]] = as(obj[["SCT"]], "Assay")
sweep.list = paramSweep(obj, PCs = 1:30, sct = TRUE)

[nitinmahajan20]

This worked for me as well, but I had to use v3 paramSweep_v3

obj[["RNA"]] = as(obj[["RNA"]], "Assay")
obj[["SCT"]] = as(obj[["SCT"]], "Assay")
sweep.list = paramSweep_v3(obj, PCs = 1:30, sct = TRUE)

[Pentayouth]

New doubletfinder only works for those Seurat V5 objects without BPCell preprocess

[simang5c]

Hi,

I've encountered the same issue you're facing. The problem arises due to the type of data you're working with. Specifically:

1. In paramsweep.R, line 18, you are trying to load the counts matrix using the following line:

data <- seu@assays$RNA$counts[, real.cells]

However, the object seu@assays$RNA$counts is stored as a BPCells data type. This is an optimized data structure used to handle large single-cell RNA-seq datasets efficiently. Because of this, the matrix is not fully loaded into memory when you access it like this. Instead, it's handled in chunks or "iteratively," meaning you're not dealing with a typical matrix that you can manipulate directly.

2. To correctly load the matrix and perform operations on it, you need to convert it into a format that can be handled in memory, such as a sparse matrix. You can do this using the following command:

library(Matrix)
data <- as(seu[["RNA"]]$counts[, real.cells], "dgCMatrix")

This will convert the BPCells matrix into a sparse matrix (dgCMatrix), which is memory-efficient and allows you to perform standard matrix operations.

I hope it helps.

Cheers :)

[chris-mcginnis-ucsf]

@haibol2016 thoughts?

[haibol2016]

@alasaa, @chris-mcginnis-ucsf , @simang5c

I think @simang5c's solution works. I personally haven't try BPcells yet, which seems very helpful for scRNA-seq data with millions of cells. I will try later. After my update, the package is tested and I am sure it works with Seurat v3 and v5 smoothly.

Haibo

[ggruenhagen3]

I think as long as a cell isn't attempted to be retrieved multiple times, BPCells could still work, which would be ideal to get its speed-ups.

[yxsee]

Just to chime in, this error impacts both paramSweep and doubletFinder. I found that data <- as(counts[, real.cells], "dgCMatrix") only works if real.cells contains all cells. Otherwise, the entire matrix needs to be loaded into memory by converting to dgCMatrix format before subsetting: data <- as(counts, "dgCMatrix")[, real.cells] This will work if the matrix is not too big, but may incur high memory usage for large datasets.