Batch effect issue in RNA velocity analysis

Question

[sarahmanderni]

Hi,

Using Seurat to perform scRNAseq analysis for multiple samples (SCT + CCA integration and downstream analysis), what would be the recommended input assay to run RNA velocity using cellrank or scvelo? I am thinking to extract RNA assay, UMAP embeddings and clusters information to provide to scvelo but the RNA assay is not batch corrected. On the other hand not sure if using "integrated" assay as input for scvelo is a reasonable choice either.

[ATpoint]

I am not aware of any method for batch-aware velocity. There are GitHub issues at the scvelo repository and then have all (by best knowledge) no satisfying answer. I would run it separately per sample and then project the latent time onto the integrated dataset, knowing that you can only meaningfully compare the latent time within sample, not across samples. I would anyway not overinterpret velocity, it is after all a quite abstract metric, at least to me, that can be used for hypothesis generation, but nothing more.