Feature request: defining ground truths

[semenko]

Thanks so much for this incredible work!

I'd love to see (if you're adding features), options to define ground ground truths based on known marker genes, outside just file_known_markers — or perhaps an option to derive file_known_markers from celldex?

I'm currently using Seurat v4 Azimuth (which unfortunately only has a few good reference databases available) and/or singleR, which has a nice workbook.

[ktrns]

Dear @semenko,

Thanks so much for your positive feedback, and your suggestion. Are you actively using our workflow?

There are a few features that we would like to add in the future, and mapping cells to a reference is a great point. You can always join our efforts of course!

Best wishes
Katrin

[kosankem]

Similar to the issue #46, I have put together a small R markdown script which is using the output of the scrnaseq script and performs cluster annotation with reference datasets using SingleR and Clustifyr subsequently. I can also share that with you if it is of any help.

[semenko]

Thanks! I've started using the workflow -- it's great!

I plan to expand use of it in January with a much larger sample set. My end game involves using EcoTyper to define cellular communities -- which requires ground truth annotations of key cell populations (e.g. "B.cells", "Epithelial.cells", etc., details here).

I'm also trying to apply a recent ML model Pollock to label things more cleanly. If things work out nicely I might make a PR.

@kosankem That script would definitely be helpful, if you're able to share.

[kosankem]

@semenko, I apologize for the late reply. You can find a updated version of the annotation_with_reference script under https://github.com/MHH-RCUG/scrnaseq_modular_extensions. I would be glad to hear your feedback.