Sorry still new to this. I have ~100 samples with 7k cells at least from the human brain (diseased and not-diseased). So far I have been using the Cell Ranger filtered_bc_matrices (for my single nuclear RNA-seq data) and getting decent umaps (albeit with some noise). I have not used any ambient rna based correction tool. I thought that 10X filtered matrices where good enough, but I read that for single nucleus RNAseq that may not be the case. Any input and suggestion is appreciated. Any tools that you all personally suggest or are filtered_bc matrices from Cell Ranger good enough?
Thanks
•
written
by
biotrekker •
SoupX, but with the raw, not the filtered folders. The point is to use the empty droplets (removed in filtered, that is what filtered means), to estimate the "soup" aka the ambient contamination, and then subtract that from the filtered cells. You can do that, but I would not say it's mandatory unless you see clear evidence, like expression of conflicting genes in celltypes where these are not supposed to be present. This could be an indication of ambient contamination. But actually, if things look good so far, better figure out the biology of the samples than caring about ambient RNA.
ATpoint Does order of operations matter? I'm doing raw and filtered bc matrix from cell ranger > soupX (autoEst, no QC) > QC > doublet detection. Even though this is snRNAseq data there is considerable MT contamination in some samples.
Is what I'm doing fine?
Is their anyway to run soupX in python, i've been trying with rpy2 and anndata2ri
I think you shouldnt do QC before doublet detection. This is what they recommend in the scDblFinder vignette.
Log in to answer this question.
