snRNA seq analysis - huge computational demand due to large number of entries

Question

[applepie]

Hello Everyone:

I have been conducting single nuclear RNA sequencing (snRNA-seq) analysis of data downloaded from gene expression omnibus (GEO). (The accession number of the dataset is GSE200366, which is a mouse dataset)

I have used the Seurat Packages to load one of the feature-barcode matrix. However, when I tried to view the metadata, it turned out that there were 6,794,880 entries (as shown in the attached figure).

Regarding this, I have two questions:

1. Would it be too computationally demanding to analyse the the snRNA-seq with such enormous number of entries?

2. How can I smoothen and quicken the process of downstream analysis even in the presence of such large number of entries ?

Thanks!

[GenoMax]

Please follow a standard workflow such as (or Seurat): https://bioconductor.org/books/release/OSCA/ These account for this sort of thing.

Answer 1

[predeus]

You've added "raw" matrix, which has lots of barcodes that are pretty much empty droplets. What you need is a filtered output - using an algorithm like EmptyDrops, you can identify the barcodes that most probably contain cells (in your case, nuclei), and only load that part. Normally, Cell Ranger or other tools would make a filtered matrix for you. It seems like the authors has only shared the raw matrices, however. You can install R package named DropletUtils and run emptyDrops function on your raw matrix.

Hope this helps.

[applepie]

Ok! I really appreciate your help.

One more question: Can emptyDrops be directly applied to seurat object?

[predeus]

I don't remember - I think you might have to convert it to SingleCellExperiment

Or, alternatively, you can read the Cell Ranger output you're looking at as SCE, run emptyDrops on it, and then save the filtered matrix.