Add script to reformat NBAtlas object

[sjspielman]

If you are filing this issue based on a specific GitHub Discussion, please link to the relevant Discussion.

Part of #1151

Describe the goals of the changes to the analysis module.

The NBAtlas reference object is disseminated as a Seurat object. We will need to use this object in different formats: an SCE object for use with SingleR, and an AnnData object for use with scArches/scANVI.

Since we'll need both these formats, we might as well tackle it in a single conversion script that reads in the Seurat object and export SCE and AnnData versions to scratch. We can then use this script to convert both the full and subsetted versions of NBAtlas; the specific one we'll use will depend on some runtime investigation (issue forthcoming), but we can still convert them to be prepared.

What will your pull request contain?

Script to convert the NBAtlas object from Seurat to both SCE and AnnData and associated docs.

Will you require additional software beyond what is already in the analysis module?

Dependencies will get added to both renv and conda

Will you require different computational resources beyond what the analysis module already uses?

This can be run on a laptop, but will need quite a few GB of memory. I will indicate the usage it ends up needing in docs as part of the PR.

If known, when do you expect to file the pull request?

No response

[sjspielman]

We have received additional metadata indicating which cells are tumor:

• The full atlas has 42413 tumor cells
• The atlas subset has 17702 tumor cells

These are labeled as Neuroendocrine cells in the NBAtlas object. When exporting other versions of this data, we should update the cell type for those cells to be tumor; note that those were all determined to be tumor in the NBAtlas paper as Neuroendocrine cells with tumor calls from copyKAT.

[allyhawkins]

One thought I had is that maybe we want to retain the fact that these cells were labeled initially as neuroendocrine and change this to tumor-neuroendocrine rather than just tumor?

[sjspielman]

One thought I had is that maybe we want to retain the fact that these cells were labeled initially as neuroendocrine and change this to tumor-neuroendocrine rather than just tumor?

Yeah, this seems reasonable for retaining that original information.