Analyzing metadata varying across cells of a donor

[pakiessling]

Hi, thank you for the nice tool.

I was wondering how I could apply this tool to my research.

I have scRNA samples of different donors. For every donor cells from different locations in the tissue were included.

I would like to detect patterns in gene expression between the different regions.

When I include metadata indicating the position of the cell in the tissue (something like "border zone", "remote zone", "ischemic zone") I get the warning message:

 "You may have included metadata that varies across cells within each donor/sample. 
              We recommend only including metadata that varies across donors/samples."

Can I use scITD for my purpose and would I need to structure my data in a different way for that?

Thank you for your help

[evanbiederstedt]

@pakiessling

Thanks for using scITD

Could you provide code above and some basic reproducible example so we could follow along? That would be helpful

Best, Evan

[j-mitchel]

The reason we recommend not including variables that change across cells of the same donor is because a common downstream analysis is to test for associations between donor metadata variables and each donor's factor score. If you don't intend to evaluate those associations, you can just ignore the warning.

If you're interested in identifying coordinated changes among cells in different regions, I would suggest appending the region name to the cell type name (e.g., endothelial_border, endothelial_remote, etc.), essentially considering cells from a different region to be different cell types. I suppose another option is to consider a donor + region combination to be treated as separate donors, and then you could test for factor association with region as I believe you intend. Let me know if this helps or if I misunderstood what you are trying to test!

[pakiessling]

@j-mitchel Thank you, this is exactly what I wanted to know. Is there a recommendation on the minimum number of cells my cell types should contain?

[j-mitchel]

There's no surefire threshold, but we generally recommend at least 50 cells per cell type per donor (based on simulations)