I have a 10x output for a tumor sample from a mouse xenograft. To make my further analysis clear, I need to filter out mouse cells. On a cellranger github I found a similar issue. The developer advised them to map 10x read to a combined mouse-human reference that they have on the 10x website. Mouse genes in the output matrix start with "mm10..." and human genes start with "hg19...". So, the developer suggested to sum up mouse and human gene counts for each cell and to remove cells based on mouse counts to human counts ratio.
When I aligned my fastq files to the combined reference, I found out that ~86% of hg19 genome and ~9% of mouse genome mapped to the reads. I summed up mouse and human counts and got a ratio for each cell. Now I am confused to set a threshold for removing cells. There are a lot of genes that are shared between human and mouse. From that point, is that a good strategy to remove mouse cells? Which threshold should I use for mouse/human ratio to remove mouse cells wit some level of confidence?
I also thought about removing mouse cells based on unique mouse genes. If you know something about it please share your thoughts.