Operationalize RNA velocity section

Bioconductor / OrchestratingSingleCellAnalysis

Content for the OSCA Book.

http://bioconductor.org/books/devel/OSCA/

64 stars 37 forks source link

Operationalize RNA velocity section #29

Closed LTLA closed 3 years ago

LTLA commented 4 years ago

@csoneson to make an unspliced dataset available in scRNAseq.
other volunteers to create a dedicated R wrapper around scvelo or velocyto, or both.

csoneson commented 4 years ago

I just sent a PR for scRNAseq with spliced and unspliced counts from a part of Hermann et al (2018).

csoneson commented 4 years ago

I was also toying with an R wrapper package for scVelo based on basilisk: https://github.com/csoneson/scvelor. Still very early-stage though. Perhaps @lazappi would be interested in joining the effort?

csoneson commented 4 years ago

I was also toying with an R wrapper package for scVelo based on basilisk: https://github.com/csoneson/scvelor. Still very early-stage though. Perhaps @lazappi would be interested in joining the effort?

Well, I guess this was a bit redundant, I just noticed that you already started this with velociraptor

LTLA commented 4 years ago

Excellent. A few points:

It would be good to have a very lightweight package whose only job is to convert SCEs to/from AnnData objects. This code could then be re-used in downstream packages like velociraptor. Some care is required to ensure that our Python dependencies are kept in sync, but I'm sure it can be done, e.g., provide a function to report the AnnData versions for use by downstream packages.
Yes, please put further development effort into velociraptor. The main missing piece is a similar wrapper for the original velocyto, which is why we gave it a more general name than scvelor. Who knows, there will probably be more velocity methods in the future. Another contribution would be to expose more API points, e.g., can we plug in our own reduced dimensions or nearest neighbors? Ideally as much work would be done in R as possible, for consistency with the rest of our analysis pipelines, and we would only enter Python for the final velocity calculations.

csoneson commented 4 years ago

I think the arguments of the scVelo functions need to be exposed as well. I thought about providing a list of lists (with one element per scVelo function, being a list where any argument to the corresponding function can be set if you want a non-default value). E.g. https://github.com/csoneson/scvelor/blob/3f5102508f727e93cbc465be0928d643e60cf7af/R/scvelor.R#L64-L66

LTLA commented 4 years ago

One could indeed do that. I don't particularly care much until it gets to the velocity function, everything else we can do inside R with BiocNeighbors and BiocSingular.

lazappi commented 4 years ago

I would be interested in helping out with this if there is anything you need a hand with, it's something I have thought about but never got around to. Also happy to liaise with the scVelo/AnnData authors if needed (given I am in the same lab).

LTLA commented 4 years ago

Should you care to build an AnnData converter, I'm sure velociraptor would use it.

lazappi commented 4 years ago

I have started a conversion package here https://github.com/theislab/zellkonverter. It just has basic functions for reading .h5ad files and converting AnnData to SingleCellExperiment so far.

Comments/suggestions/contributions welcome!