Is magicked data recommanded for cellrank pipeline?

[jiangpuxuan]

I use velocyte to get .loom from 10x results, and seurat, then transform the seurat object into adata and run scvelo + cellrank pipeline.
However, I found huge differences between the raw RNA and MAGIC_RNA results, especially in 'initial states'.

Thus, I wonder how exactly does MAGIC influence the scvelo+cellrank pipeline. Thank you for your help!