Trajectory analysis of a lineage across different tissues


I was introduced to the scvi-tools a couple of weeks ago when I performed some RNA-protein analysis with TOTALVI and I came back for more :slight_smile: My next question now is to construct a trajectory analysis of cell type x in different tissues. In other words I have two scRNA-seq datasets, one from tissue A and one from tissue B. The cell type x migrates from tissue A to tissue B during inflammation. What I want to do is order the cell type states in tissue A and tissue B without overcorrecting and from a literature search, it seems that SCVI has the highest potential to do so. I just wanted to get some feedback from the developers on this question. My understanding from reading the Xu et al. (2921) Mol Sys Biol, and particularly Fig. 3, is that SCVI would quite accurately overlay very similar states, but keep separated more distant cell states.


I take it that tissue A and tissue B were collected separately (i.e., no sample multiplexing like hashtags)? In this case I would run scvi.model.SCVI two ways: first, by registering the tissue as the batch_key and second without doing this. Our benchmarks suggest it won’t overmix when you do batch correction, but first you should establish to what extent these batch/tissue effects exist.