[Feature Request] registration_wrapper(): create and check validity of eventual pseudobulk factor names before starting

[berniejmulvey]

I recently ran registration_wrapper() on a very high-dimension dataset (~80 or ~200 clusters across 47 samples, depending on the granularity) and twice had runs go >8 hours before seeing warnings about invalid factors in the limma steps. For example, the first time around, the cluster names were character strings but with leading digits (e.g., "038_Nonneuron_OPC"). There was a single cluster of cells that had been annotated with a "/" in the name, which I didn't even consider the factor invalidity of.

Since the duplicateCorrelation step for datasets of extra high complexity can take several hours, it would be more resource efficient to have a pre-run check that stops the wrapper from running if invalid factor names are going to be generated down the line, rather than wait for limma to encounter them.

[lcolladotor]

Can you provide more details about the error you got and maybe make a small reproducible example with a fake dataset?

See https://youtu.be/8bBo3B7N8YQ and https://reprex.tidyverse.org/ for more details. Though the basic idea is to write code that you can then run with reprex::reprex() that shows the problem you encountered. Just make a tiny example with random data but with the right values in the colData() that led to the error you are describing.

Thanks!

[berniejmulvey]

Ah, so more specifically, it's whitespace characters that are not picked up by the initial check run by SpatialLIBD (second spatial_registration call in the attached code). Here's an HTML output from reprex. spatiallibd-cell-type-label-error-reprex.html.zip

[lcolladotor]

I'm delegating this one to @lahuuki now that she's back analyzing spatial data