Discordance between arrows in different velocity embeddings.

I started experimenting with scVelo and got some interesting results but also noticed some discordance between the different velocity embeddings.

Result of scv.pl.velocity_embedding:

Result of scv.pl.velocity_embedding_grid:

Result of scv.pl.velocity_embedding_stream:

Out of those 3 the grid representation seems a little off, inverting many of the thick arrows in the green and blue clusters.

Additionally, the arrows in the PAGA plot don’t seem to correspond very well either:

sc.pl.paga_compare(
  adata, basis = "umap", threshold = .15, arrowsize = 10,
  edge_width_scale = .5, transitions = "transitions_confidence",
  dashed_edges = "connectivities"
)

Did I misunderstand the output of these representations?

[VolkerBergen]

If I understand correctly, the discordance only arises when running scanpy’s preprocessing steps without running scvelo’s pp steps.

It’d be important to normalize the spliced/unspliced count data, which is not done within the scanpy pp pipeline. That would be running scv.pp.filter_and_normalize() after all your pp steps, which then only touches your spliced/unspliced counts and leaves adata.X as is when already processed.

[romanhaa]

Ok, that makes sense. I’ll close this thread then, no further questions for now 😃

Thanks again and great work! I like the dpi option for plotting. Would be useful also in the scanpy plot APIs.