cicero with multiple samples

[Daniel Gingerich]

Hannah, 

I have 24 total samples with 26 final clusters.  Want to do cicero on each cluster separately.  As such, Before running cicero I do SVD on each cluster individually in order to capture max within-cluster variance.  My question: Should I integrate the SVD embeddings using a tool like Harmony or Seurat before constructing cicero metacells?  I think this would help eliminate bias from age, sex, pmi and other sample level artifacts.  Is the unintegrated SVD fine to use?

[hpl...@gmail.com]

Hi Daniel,

I have not tested methods of data integration/batch effect reduction with Cicero before... If you're using the embeddings just to generate a map to pass in to cicero for the make_cicero_cds step, then I see no problem with using the integrated embeddings. However if you're modifying the count matrix that's being aggregated then you'll likely run into issues due to the normalization/scaling. 

Best,

Hannah