Combining scRNA-seq datasets that have been processed differently

Question

[S]

Hi,

I am new to immunology and I was wondering if it was okay to combine 2 different scRNA-seq datasets. One is from the lamina propia (so EDTA depleted to remove epithelial cells), and other is CD45neg (so the epithelial layers). The sequencing, etc was done the same way, but there are ~45 LP samples, and ~20 CD45neg samples.

I have processed both the datasets separately but I wanted to combine them for cell-cell communication, since it would be interesting to see how the epithelial cells interact with the immune cells.

My questions are:

1. Would the varying number of samples be an issue?
2. Would the fact that they have been processed differently be an issue?
3. If this data were to be published, would it be okay to have all the analysis done on the individual dataset, but only the cell-cell communication done on the combined dataset?
4. And from a more technical Seurat pov, would I have to re-integrate, re-cluster the combined data? Or can I just normalise and run cell-cell communication after subsetting for condition of interest?

Would appreciate any input! Thank you.

[dthorbur]

I would suggest only using these datasets for preliminary analyses. If you are trying to compare between cell types, any results you find will be nested within the dataset effect, which you'd never be able to disentangle statistically. Realistically, there are too many factors between datasets that are likely different, including sequencing depth, platform software versions, flowcell health, library prep, etc...

The number of samples shouldn't be too much of an issue so long as they cluster, but downsampling LP to 20 would be an easy fix.

[S]

Thank you for your input! And yes I agree with you in terms of the difference in processing, though the workflow was pretty standard except for the difference in depletion (of course sequencing depth, batch effect, etc. cannot be accounted for).

We were hoping to use the results in our publication though, and if we make the caveats very clear do you think this is still something we can pursue?