From Seurat to AnnData and scVelo

[Jeff87075]

Hi, apologies for such a basic question but I'm pretty confused about the input files required for running the scvelo analysis workflow. I am familiar with Seurat but am new to python and the ultimate objective is to run the RNA velocity analysis but with the UMAP embedding and clustering I've processed in the Seurat object.

I've been trying to go through the tutorial and know that both the adata and ldata should be loom files but I'm uncertain as to what exactly is the difference between the two and how/where they can be generated. Any guidance would be appreciated, thanks!

[WeilerP]

@Jeff87075, sorry, not sure what you mean by ldata. Could you elaborate?

I'll try to briefly outline how to proceed on going from a Seurat object to AnnData and using scVelo:

Conversion from Seurat to AnnData

When you have your Seurat object data you can, for example, convert it to AnnData using anndata2ri. See, for example, here how you can use it in a Jupyter notebook. This will keep the UMAP embedding (should be added as adata.obsm['X_umap']) and clusters (stored as column in adata.obs).

Alternatively, if you want to do everything in R, you can use Seurat and SeuratDisk:

library(Seurat)
library(SeuratDisk)

data <- ReadRDS(data.rds)
SaveH5Seurat(data, filename = "data_from_seurat.h5Seurat")
Convert('data_from_seurat.h5Seurat', dest='h5ad')

Disclaimer: I previously ran into problems using the latter approach when the neighbor graph shape did not match the count matrix (e.g. more nodes than observations because they were subsetted after graph construction).

Loading loom files and aligning data

import scanpy as sc

adata_from_seurat = sc.read()
adata = sc.read_loom(LOOM_FILE.loom)

# Remove observations/features not present in data from Seurat
adata = adata[
    adata_from_seurat.obs_names.intersection(adata.obs_names),
    adata_from_seurat.var_names.intersection(adata.var_names)
]

# Add Seurat UMAP embedding and clusters
adata.obs['seurat_clusters'] = adata_from_seurat.obs['seurat_clusters']
adata.obsm['X_umap'] = adata_from_seurat.obsm['X_umap']

[Jeff87075]

Thanks for the clear explanations. @VolkerBergen I actually have the ldata files now so I just have to figure out how to obtain the adata loom file and I think I have got it figured out using the as.loom() function. But I was wondering then what's the difference/relationship between scvelo and velocyto? Or in other words is the output files from velocyto an input file for scvelo and in the future if I do not have the files from velocyto to begin with, do I just skip the scv.utils.merge(adata,ldata) step and follow the workflow as stated by @WeilerP ?

Also one more thing I'd like to clarify @WeilerP for this step:

adata_from_seurat = sc.read()
adata = sc.read_loom(LOOM_FILE.loom)

If I'm understanding it correctly, the adata_from_seurat is the AnnData file converted from Seurat but then where does the loom file for adata come from? Many thanks.

[WeilerP]

Yes, adata_from_seurat is the AnnData converted from Seurat. The loom file comes from running the velocyto pipeline velocyto run .... This also answers your other question: The output of the velocyto pipeline, i.e. the loom file, is used as input for scvelo. It provides the unspliced, spliced count matrices used to calculate RNA velocity.

The second part of the workflow I outlined is almost the same as the other one. The difference being that you align the AnnData objects yourself and manually pick slots and columns to add.

[Jeff87075]

I see, sorry just to make it very clear, so can I take it as the output loom file (with the unspliced/spliced count matrix) from velocyto pipeline is a must/prerequisite for running scvelo? Also is it the adata_from_seurat can also be a loom file instead of the AnnData H5AD file type? I'm asking this because the conversion to AnnData using the SeuratDisk functions seems to be giving some errors. Thanks a lot.

[WeilerP]

Yes, unspliced and spliced count matrices are needed for running scvelo. They are the basis of RNA velocity in general!

And yes, adata_from_seurat can be read as a loom file as well. In general, scanpy supports reading a vast number of file types into an AnnData object. See here for the official documentation.

[WeilerP]

@Jeff87075, did this answer your question? If so, would you please mark the discussion as answered?