Can we use resolVI and squidpy to annotate a CosMX object based on a single cell reference?

Help squidpy
1

Hi,

  1. Are there tutorials or protocols to assign cell types in CosMX loaded via Squidpy based on a single cell reference ? I think ResolVI might help, but I did not found a clear end to end tutorial

  2. When annotating CosMX cells, are there recommended cut-offs at which point cells are dropped to low quality (poor segmentation, low number of detected cells, etc) ?

Thank you,

Cristian

2

Hi, resolVI does not provide integration with scRNA-sequencing data. We hope to provide an approach for this in the near future. I filtered out below 20 true counts (true proportions X counts of that cell). Split provides a way to do this type of label transfer: GitHub - bdsc-tds/SPLIT: Profile purification of single-cell spatial transcriptomics data. Warning in my hands it doesn’t scale to datasets beyond 500k cells. ProSeg estimated counts together with Leiden clustering and LLMs will likely also do the trick (you can then train a semisupervised resolVI model afterward to clean it a bit more and get more fine-grained).