One cluster in reference is completely missing in query predicted clusters - even when using a sample that was used for reference building

[florian0512]

Hey,

First of all thanks a lot for making this package and also for the great implementation and documentation!

I build a reference from SCtransformed Seurat object (buildReferenceFromSeurat()).
When I then map my query (all according to your tutorial) and finally use knnPredict.Seurat(), the query mapping is missing cluster 2 (a fairly large one).

To verify if this is coming from the data I redid the query mapping using one sample that was used for reference building - and the same problem arises with it.

I was wondering if you've experienced that/if you have an idea on how to improve mapping (maybe not starting from an SCtransformed object?).

The code I used is pretty much how it's described in your tutorial and I didn't get any other warnings than the one displayed during the tutorial as well.

Thanks a lot!

[joycekang]

Hi Florian,

Thanks for the question. Trying to see if the issue is with the Symphony mapping procedure itself or how the reference was built. Would you be able to clarify whether you're seeing this behavior with a log(CP10k+1) normalized dataset as well? We haven't tested Symphony with SCTransform as much as log-normalization. I see that cluster 19 is not being predicted as well, but perhaps that is because it is a very small reference cluster. Thanks!

[joycekang]

Closing this issue (but feel free to reopen if still not resolved!)