mouse+human cells in 10x scRNA-seq

Question

[Sy80]

Hi everyone,

I'm analyzing 10x scRNA-seq data generated from xenografts (mouse + human tissues). I have the following workflow to label cells as either mouse or human:

1. Align 10x scRNA-seq data to mouse+human combined genome using cellranger count.
2. Use the file generated by cellranger count (gem_classification.csv) which assigns mouse, human, or multiplet to each cell to classify each cell type.
3. Realign the fastq files to the mouse genome ( we are interested in the mouse cells) using cellranger count
4. Use the classification from Step 2 to filter out human cells in Seurat.

Would this be the right approach or is there a better alternative that you would recommend?

Thanks a lot!

[dario.garvan]

For the third step I get the error

Martian Runtime - v4.0.8
RuntimeError: pipestance 'HNP_01_P5_culture' already exists with different invocation file __HNP_01_P5_culture.mro

How should it be modified so that it doesn't baulk about the sample already being processed once?

[jared.andrews07]

Change the ID. Or output in a different folder.

Answer 1

[jared.andrews07]

This approach is fine, but is there a reason you aren't just using the output from step 2 and limiting to the mouse cells/genes within it?

[swbarnes2]

If there are stray reads in "mouse" cells which aligned to human for some reason, reliagning to mouse alone will force those reads to be assigned to a mouse gene.

[jared.andrews07]

I guess I don't have a good grasp on how much of a difference it'd make, would be interesting to know if there are specific genes where this is a particular issue.

[Sy80]

Thank you to both for your input. I also thought that alignment may improve if only a single ref genome is present, perhaps only for a small fraction of genes - though this is just an assumption. I may try both ways just to compare.

[jared.andrews07]

If you notice much of a difference, be sure to give us the update!

[Sy80]

Will do so!