Using a scRNA as reference to train custom CellTypist models (or load custom pre-trained CellTypist models)

[Rafael-Silva-Oliveira]

Hey! really excited about this tool and thrilled to see more progress being made on Visium HD! Currently, it seems the primary methods in enact for cell annotation rely on cell markers or pre-trained models (especially with CellTypist). I’m curious if there are any plans to use annotated scRNA datasets as references for CellTypist, enabling users to train custom models from scratch based on their own scRNA data and then apply these models to infer spatial cell types. I think the fastest way to enable this would be to at least allow users to load their own custom CellTypist models, as it currently appears to load only from pre-existing models in CellTypist (though please correct me if I've missed something on this!).

An approach similar to what's described in this article (https://www.sciencedirect.com/science/article/abs/pii/S0010482523000999), where a cell marker finder is integrated with annotated scRNA datasets, could also be helpful for using those markers to identify spatial cells

Thanks again for all the great work!

[Mena-SA-Kamel]

Hi Rafael, thank you so much for your very valuable insight. We can certainly add ability to load pre-trained CellTypist models in future releases. This sounds like a simple and reasonable ask.
Regarding the use of scMAGS, this could be useful to mention in the Readme.