There have been a few methods proposed for integration (or batch correction) of scRNA-seq datasets, such as Seurat CCA, MNN Correct, Scanorama, and Harmony. The concern is generally about the maximum number of cells that they handle, but I haven't seen any discussion about the minimum number of cells. I am confident they can all handle 10k cells reasonably well and will fail with 10 cells, but where do you draw the line? Is there a method that works best for small datasets?
For example, with plate-based platforms like Fluidigm, many experiments only have 96 cells and potentially much less after quality filtering. How can those be used?