Choosing a cut-off for mitochondrial gene % depends on things like the organism and type of tissue. This paper recommends a general cut-off of 5% for mouse and 10% for human tissues, but I am curious to see what people consider a good cut-off range specifically for brain single-cell (or single-nucleus) RNAseq data for mouse and human samples.
Plot a violin of the metric, look at it, then decide. Keep in mind what filtering is for, so to remove outliers from the rest of cells avoiding bias. Meaning, if cells are similar and there are no outliers, maybe no filtering at all will do just as fine. If possible, do it per celltype, since different celltypes can have different characteristics. There is no universal cutoff. Use the best ML tool which is the human eye/brain, look at the data.
Also note single nuclei sequencing generally has much more restrictive mitochondrial % cutoffs (like 1-2%), as there should really be limited mitochondrial reads in properly prepped/isolated nuclei. There is not a threshold that will work for both technologies for the same cell.
Otherwise, in agreement with ATpoint , just look at the data and see what it shows. I've generally found relaxed, very conservative thresholds (<20% mito reads, 200 genes detected, at least 1000 reads) combined with outlier analyses for each celltype or cluster sufficient.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6