Fix fastMNN corrected feature output

[danielStrobl]

A fastMNN API change causes anndata2ri to break for newer batchelor versions as LowRankMatrix can't be converted to python. Converting to dgCMatrix fixes that.

Not sure if we want to lock to an older batchelor version or do this fix here.

[mumichae]

Thanks for spotting that! I was wondering why fastMNN suddenly stopped working but didn't know that it was related to a newer version. Will test if the pipeline runs through and get back to you on this

[LuckyMD]

I think future proofing is better. Thanks for fix @danielStrobl! @mumichae you have power over merging, so if you give it a thumbs up, please merge :).

[mumichae]

I'm still trying to run things, and as it stands, the code doesn't seem tested yet (I got a syntax error). I still need a bit of time though, as I'm having issues with my R environment atm.

[danielStrobl]

this was missing a bracket, was just quickly transferring my fix without testing from openproblems: danielStrobl/SingleCellOpenProblems@2092d72

[mumichae]

I just realised that we are most likely looking at different problems here. In Openproblems we are using R 4 and Seuratv4, whereas the pipeline uses R 3.6 and Seuratv3. For some reason, I am not getting any errors with fastMNN anymore after recreating the R environment.

Sooner or later, we should move to R 4 and only keep the 3.6 and respective dependencies for reproducibility reasons. Once that is underway, this fix will be useful, but it will break pipelines running with older versions.

[LuckyMD]

Yes, ideally R 4 as soon as possible.

[lazappi]

Just FYI R 4.1 is the current version and R 4.2 comes out in a month. I would probably just skip ahead to whatever the current version is when you get around to this.