Patch-seq mapping tutorial clarification

[xpliu16]

• Mapping Tasic data back to Tasic taxonomy is pretty confusing for the example. For one thing, it is unclear which things should come from the query and which from the anno. For instance, annotations$cluster and annotations$class are read in from the annotation prior to mapping, whereas I think in a more typical flow they would be read in from the metadata and mapping result? And it's only being done to subsample the data to produce a query?

• Instead, it may be more helpful to show the preprocessing from Allen data files, including the proper log transformation of the cpm.

• In the complete_patchseq_analysis vignette, it says to read_h5ad(file.path(refFolder,"reference.h5ad")), but tasic_2016 has only AI_taxonomy.h5ad - as pseudocode this is understandable, but not fully clear. Also, in NHP_BG_AIT_114, I initially use complete_AIT_114.h5ad instead of AIT_114_taxonomy.h5ad, which was also confusing. Would help to clarify alternative use of read_h5ad() as opposed to load_taxonomy().

• Suggest discussing importance of filtering ROI and QC before trying to make sense of the results and suggestion for kinds of things to QC on and starting values. As I was trying to sanity check the results to see if I had run the mapping correctly, they made very little sense before doing that filtering. Of course, it is also possible that particular QC values are too aggressive and that's something the researchers should adjust on a case-by-case basis.

• Requirements and flow for running patchseq QC aren't super clear - I'm just starting this though, and will report back more details if applicable

[jeremymiller]

This is super helpful. I'll keep this in mind when I redo the tutorial in the next couple of weeks. Keep the suggestions coming!

[UCDNJJ]

Thanks Ping, I agree its confusing to use the same data for query and reference. Would you or someone on the PatchSeq team be willing to create a data R package with some of the published cortical patchseq samples from mouse?

This would be really helpful in the long run for scrattch.mapping documentation and examples.

We are using Tasic2016 because its very convenient to library(tasic2016data) and have some RNA-seq data :).