Merge vs Integration for Cell Typing

[Rayan21100]

Hi everyone,

I have 4 Spatial transcriptomic dataset from 4 different patients. I'm interested in doing the cell typing for all the patients, because I'm interested in rare subsets, I wanted to merged/integrated them to have more cells and be sure to detected them.
My plan was to:

• SCTransform
• RunPCA, FindNeighbors, FindCluster and a DE analysis to separate immune from stroma cells for each patients separately
• Create a new list of objects containing only the immune cells of each patients
  (1)

Then integrating this list of object with:

• SelectIntegrationFeatures
• PrepSCTIntegration
• FindIntegrationAnchors
• IntegrateData

Then on the integrated object:

• SCTransform
• RunPCA, FindNeighbors, FindCluster and a DE analysis again on the integrated assay of the integrated object

Afterwards, I will split the object as my integration is only to do cell typing, I'm not interested in patient differences. I just want to do the cell typing across all the cells I have and boost my statistical power by having as many cell I I can

I have several questions about this:

• What do you think of this workflow ?
• Is there some advantage to use Merge rather than Integration ?
• Should I normalize again (SCTransform) my list of seurat objects containing only immune cells ? At (1)

Thanks in advance !