SingleCellExperiment/Seurat should contain dgCMatrix instead of dgRMatrix

[pakiessling]

Similiar to the discussion in theislab/zellkonverter#34

A lot of tools like https://github.com/immunogenomics/presto/ will try to call functions from the Matrix package that error when encountering a dgRMatrix.

[Artur-man]

Looking at my code, I hardly use dgRMatrix, most of my converted matrices are in dgCMatrix

[lazappi]

Can you share an example of what your use case is? I agree that dgRMatrix should mostly be avoided but there is a difference here in that AnnData is transposed (cells x genes) compared to SingleCellExperiment or Seurat (genes x cells) so which format is more "natural" is different.

[pakiessling]

I use anndataR to integrate R methods in my mostly python based workflow.

I will routinely read data from on-disk .h5ads into Seurat / SingleCellExperiment to then apply tools e.g. presto

These tools (Is suspect all tools calling common {Matrix} functions) will then crash with an error message about the matrix having the wrong format.

I would expect a Seurat / SingleCellExperiment object generated by anndataR to be out-of-the-box compatible with functions taking Seurat / SingleCellExperiment objects as input

[lazappi]

Can you please give a code example? If you are using {anndataR} functions to convert to Seurat/SingleCellExperiment then I agree that should return a dgCMatrix.

[pakiessling]

Yes here is what happened:

library(anndataR)
library(presto)

sce = read_h5ad("adata.h5ad",  to="SingleCellExperiment)
wilcoxauc(sce, 'cell_type')
> Error message about incorrect matrix format.

I think that this should "just work".
Less sure about how to handle things when staying inside an anndata, but isnt the main draw of using anndataR to be able to use R packages on anndatas? Won't a lot of them always fail for dgRMatrix input?

[lazappi]

Thanks for the extra detail. I will update the issue title to make it a bit clearer. @rcannood @LouiseDck this is pretty important for anyone who wants to read a .h5ad to a SingleCellExperiment/Seurat. As part of the conversion when matrices are transposed they also need to be converted from dgRMatrix to dgCMatrix.

[LouiseDck]

I was following the discussion here, and read the issue on {zellkonverter}. I am convinced that the conversion should probably produce dgCMatrices.

I am not sure whether they should be dgCMatrices in the anndata representation --> if you want to interact with the matrices in the anndata object, you might want to run more specific methods where a row major format makes more sense? On the other hand, if you want to run these specific methods, you might be more aware of the differences and be able to easily convert it by yourself?

Do you have any thoughts @rcannood ?