Closed dbdimitrov closed 1 year ago
OK. I ran two different techniques to filter doublets in R and both results in different number of doublets detected than scrublet. We could perhaps just leave the doublet detection code as a best practice and just use total_counts to keep consistency between R & Python.
Also, doublet detection takes ages in R (even when parallelized)
We currently filter cells by total counts (very minor since the data is already largely preprocessed), but if we want to stick to best practices best to change it to doublet removal by sample :)
This is also related to the Seurat -> SCE issue because doublet removal in Seurat is not great, while in SCE it's largely comparable to scanpy.