Hello biostars community,
I need help regarding a QC, for a 10X gene expression assay (GEX component of a multiome, nuclei), at what point are you guys filtering percent.mt ? The pbmc3k Seurat tutorial has said cells >5%, remove them, but those are pbmcs, which I understand are probably best case scenario.
I numbered the samples. I would say 7 and 8 were simply bad runs, probably lots of dying cells and the mitochondrial transcripts were adhering to the nuclei.
The others could maybe be salvaged through filtering?
Thoughts?
Violin plots:
•
written
by
Pratik
1 answer
20% is very generous for single nuc data, but an iterative filtering approach is totally fine. Exploratory analyses are useful for just these sorts of things. As a first pass, I'll set some conservative baseline thresholds (<20% mito, at least 500 reads, at least 200 genes detected, etc) and then just look at the distributions in each sample, see if any low quality populations come out in clustering, etc.
The metrics can vary significantly between cell types and tissue, particularly for whole cell kits. But a flat 10% threshold across the board for single nuc may raise eyebrows during review whether it seems reasonable or not. Some people stick to the arbitrary numbers without good reason.
You might consider using cellbender or soupX or similar tools to assess levels of ambient RNA in your empty droplets. They can help filter out some of the garbage and are easy to use. They may or may not drop off some of the mito reads in your "good" cells.
I wouldn't consider this data unusable or anything (outside of those 2 samples), but you want to be as rigorous as you can be imo.
Log in to answer this question.
Filtering for mt isn't always a good idea, see here https://pmc.ncbi.nlm.nih.gov/articles/PMC8599307/
There have been a number of studies suggesting filtering for mt might sometimes remove viable cells. What does the rest of your QC look like? If you ignore mt filtering and move on to cluster analysis, do you get a clear single cluster with high mt %