Filter out mouse reads from single cell RNA-seq of PDX tumor

Question

[fifty_fifty]

My goal is to get an expression matrix with gene expression values for each cell without any mouse cells in there. I need to analyze single-cell data from a PDX tumor.

I have read papers about the tools for deconvolution of mouse and human genes from PDX tumors. However, I am not sure if they can be applied to single-cell data.

I have fastq files with sc RNA-seq, I aligned them separately to mouse and human references using STAR. I used the resulting . bam files as an input in XenofilteR to filter out mouse reads. Then I tried to quantify the resulting filtered .bam file using featureCounts from this tutorial. But the output expression matrix did not have cell columns.

I will appreciate if you share any concerns or advice.

Answer 1

[swbarnes2]

I aligned them separately to mouse and human references using STAR.

Why didn't you align to a combined reference?

I would think you should be able to spot the cells with mouse reads, so it's probably easier to filter at that step.

Then I tried to quantify the resulting filtered .bam file using featureCounts from this tutorial. But the output expression matrix did not have cell columns.

I'm pretty sure FeatureCounts by itself won't handle UMIs or cell barcodes You have to use something else as well to get the counts split by cell.

[fifty_fifty]

I would think you should be able to spot the cells with mouse reads, so it's probably easier to filter at that step.

Well, I thought about doing that but I do not know how exactly to remove mouse cells then.

[swbarnes2]

Well, what have you tried?

[fifty_fifty]

I was planning to concatenate mouse and human fasta and gtf files and map my fastq files to those using STAR. But then I saw STAR Log.final.out file and did not see if it distinguishes the cells with mouse reads.

I would think you should be able to spot the cells with mouse reads, so it's probably easier to filter at that step.

This is my bottleneck right now. Could you please briefly describe how to filter out cells with mouse reads at that step?

Answer 2

[TC_Chang]

Have you ever tried to map your fastq files to the combined human and mouse reference genome? The matrix files that CellRanger generates will show mouse features. Maybe, you filter out the mouse feature in this way?