I know I am not the brightest in this topic, but I believe this question should not be that stupid.
Is there (generally speaking) a qualitative way to compare 2 workflows of scRNA seq, which only differ in their mitochondrial cutoffs?
I followed the workflow from the seurat vignettes "Guided tutorial" , "SCTransform" and "Integrate Data" and when it came to this code: pbmc <- subset(pbmc, subset = nFeature_RNA > 200 & nFeature_RNA < 2500 & percent.mt < 5)
I entered one time 30% for the mitochondrial cutoff and another time 20%.
I did this analysis in two distinct projects, but every other step is identical.
If it's possible, are there some criterias to qualitative compare the results (UMAPs, found marker genes) or does this need additional analysis steps?
It is not entirely clear to me what the goal of the comparison is. First, if you truly want to understand the impact of the mitochondrial cut-off, it would make more sense to focus on the same data set. Second, what types of metrics are you interested in? Do you want to see if the clustering is less stable? The marker genes less reliably identified? Or something else?
The goal would be, to find out if the 30% mitochondrial cutoff would be better or the 20%. And I used the same data sets for this. Sorry, I admit my sentence is misleading.
About your second point...I don't have specific metrics. I was just generally wondering, if it is comparable / if we can say "...because of this, the x% cutoff is better"?
The only thing I can say is that the annotation with the 20% cutoff was harder, bc there were a few different marker genes (compared to 30% cutoff) and these genes were not that easy to annotate to a specific cluster/cell type. But I thought, maybe there are other criterias.
If you want to decide which strategy works better, you'll have to think about how you would define "better". Cell type annotation ease could definitely be one criterion. That being said, maybe you should think about what the mitochondrial content actually represents for your data set at hand (for most data sets 30% mito content sounds pretty high; for blood cell samples I'd probably discard most cells with more than 5% mito content, but I certainly would be more lenient in, say, liver cells). It is usually helpful to check whether the mito-content might be a driver of specific clusters of cells, too. To get a better sense of the meaning of the mito content, you may want to read this recent post.
I don't know how to definde "better". That's why I am asking here. I was thrown into cold water with this project.
But thank you for your time and help. I hope my brain can do something with the information you provided.
Well, what is the goal of the project? Is it to illuminate the effect of mito content filtering?
Trying to reproduce some of the scRNAseq analysis steps from a paper in RStudio with Seurat, therby understanding what and why I am doing something and what I can deduce from the output.
I am studying biology and thought it would be a good idea to expand my horizon by choosing bioinformatics.
If you're trying to understand the details of scRNA-seq analyses, I strongly recommend the OSCA book. It doesn't use Seurat for the analyses, but it explains the "why" much better than most Seurat vignettes.
Oh, I came across a few websites, which were similary build as this, but only gave little detail/not the detail I needed. Thank you!