Identify differentially expressed genes on the BayesSpace clusters on the pilot data

[lcolladotor]

Once #17 is done, the next step will be to identify differentially expressed genes on the BayesSpace clusters. There are several ways to do this and we currently don't have code for this at spatialDLPFC with Abby @abspangler13 though we will work on this soon.

The options are:

• scran::findMarkers() like at https://github.com/LieberInstitute/HumanPilot/blob/master/Analysis/Layer_Guesses/layer_specificity.R#L1170-L1236 or how Matt used it at https://github.com/LieberInstitute/10xPilot_snRNAseq-human. I see that there's a new function scoreMarkers() at http://bioconductor.org/books/3.14/OSCA.basic/marker-detection.html (maybe it's the same one that got renamed, dunno!).
• Using limma directly like at https://github.com/LieberInstitute/HumanPilot/blob/master/Analysis/Layer_Guesses/layer_specificity.R#L1351 and lines further below. This involves pseudo-bulking the data prior to using limma. There's different models you can use and well, that will depend on what type of marker gene you want. See our Nat Neuro 2021 paper or http://spatial.libd.org/spatialLIBD/ for more info.
• Using pseudoBulkDGE() like at http://bioconductor.org/books/3.14/OSCA.multisample/multi-sample-comparisons.html. We'll likely use this with Abby on the spatialDLPFC data.

You might want to discuss this in more detail with @stephaniehicks and @stephaniecpage to choose the right model for the type of marker you want. I mean, there's also the work we've done with Louise @lahuuki for deconvolution marker genes, but I don't think that applies to this case.

[lcolladotor]

From our conversation today with @abspangler13 and @lmweber, we think that using pseudoBulkDGE() is the way to go. We'll soon start using it with the spatialDLPFC project and see how it goes.