Preprocessing enhancements

[AnnaChristina]

• [ ] add DecontX tutorial
• [ ] add more details for visualization "This is generally where I'd like to see a few comments on limitations of embeddings and visualizations. How interpretable are the distances? If you don't want to tackle this now, please open an issue on the repository and we can add this in a couple of weeks."
• [x] Separate sections into separate notebooks
• [ ] Investigate & cite more of Dmitry Kobak's papers (e.g. https://www.nature.com/articles/s41467-019-13056-x) to explain stuff like initialization for t-SNE & UMAP and equivalence. Add some more details on when to choose which or whether the choice doesn't matter at all.
• [x] Add more reasoning why scDblFinder is show cased. What is recommended? You mention that there are benchmarks but then never discuss them.
• [x] Add one UMAP that shows which cells are singlets and which ones are doublets for https://www.sc-best-practices.org/preprocessing_visualization/quality_control.html#doublet-detection and how to filter them. Remove doublet islands.
• [ ] Specify whether doublet detection should be done on ambient RNA corrected data or not
• [ ] Specify that ambient RNA correction should be done sample wise

[const-ae]

Hi Anna,

this looks like a cool effort. I just skimmed your section on "Normalization" and wanted to point to a benchmark article, I recently wrote https://www.biorxiv.org/content/10.1101/2021.06.24.449781v4. We compared many of the mentioned preprocessing methods looking at their conceptual differences and their empirical performance. I hope you will find this article useful, let me know if you have any questions.

(Sorry, if this is the wrong issue, I wasn't quite sure where the best place to post this was)

Best, Constantin

[Zethson]

Ohh great. This one is very recent :) We'll have a look.