Cell type annotation exploratory notebook

[sjspielman]

Closes #170

This is an extremely draft PR that begins exploring the use of SingleR to annotate the Gawad data. As of the time of filing, this is done with 2 references (where 1 reference is almost certainly way better suited to the data than the other) on 1 SCE object, and exploration is minimal. The notebook is also just sitting in the root level for now, final location TBD. I just wanted to file something before the weekend in case folks are curious.

[sjspielman]

Here's the current version of the rendered HTML: cell-annotation.html.zip

[allyhawkins]

This generally looks pretty good as a start! My sense right now is that we want enough confidence in our cell type information to validate integration. So a few just overall comments, some of which might be for a separate PR/ discussion:

• I think it might be helpful to take the top cell types similar to what you did in the intro notebook for training, but also what we've been doing throughout the notebooks for UMAP plots. There's a lot of cell types here and on the UMAPs I can only really distinguish 5-6 discrete colors. So it's hard to tell which dots are really which cell types. Could you start with the top 5-6 cell types and then if that doesn't cover enough cells maybe try more?
• I think we need to add in the CITE-seq here, but maybe this is for a secondary PR. I liked what you did in the intro training notebook where you plot the max ADT for each cell. I think this dataset gets a little more difficult because there are a lot more cell surface markers, but that might be worth starting. Or like what was suggested earlier, splitting the markers by lineage and then going from there.
• Also we should be careful here because we are now working with tumor samples... so some of these cells are normal and some are tumor which will make the lines between cell types potentially less clear.

[sjspielman]

I've updated this notebook substantially largely with feedback from our meeting on Tuesday 11/15, and removed the draft label. This should now be ready for a formal look: cell-annotation.nb.html.zip

Edit - should say still working on violin plots but this is a hefty start!

[sjspielman]

@allyhawkins I've re-factored the code based on your comments, and I've added some violin plots!

One question I have is whether we want to split this notebook up? I'm specifically thinking if we want to place "comparing references" into its own notebook.

[sjspielman]

Argh, realized I pushed a knitted version while turning off the library-specific UMAPs and heatmaps (viz=FALSE for run_SingleR function). I'm re-running it now and the next commit will have those plots!

[sjspielman]

See notebook with all viz here: cell-annotation.nb.html.zip

[sjspielman]

Officially 5 celltypes! cell-annotation.nb.html.zip