Best way to subset when using scvelo on Seurat embeddings

[Leena-Abdullah]

Hello,

I am trying to use scvelo with my seurat object and so I project the velocity grid onto my umap embeddings when using scvelo. My umap is divided into two partitions and I wanted to apply velocity and velocity_pseudotime on the two partitions separately. I was wondering what would be the best way to subset my data and then calculate velocity/Pseudotime?

##subsetting to just look at clusters from one partition followed by preprocessing and velocity calculation in scvelo:
clusters = ['0', '1', '2', '3', '5', '6']
adata_subset = adata[adata.obs['clusters'].isin(clusters)]

2. ##use the grouping option in the velocity and velocity_pseudotime:
   scv.tl.velocity(adata, mode='stochastic', group=['0', '1', '2', '3', '5', '6'])

3. subset cells from clusters = ['0', '1', '2', '3', '5', '6'] in seurat and recluster them and then use scvelo.

Is there any difference between option 1 and 2? I personally don't think option 3 is the right one as it might result in some cells getting reshuffled between clusters. Am I right to think this?

Thank you for your help!