I have run a Seurat pipeline to integrate 4 different scRNA datasets, and have been able to successfully run FindAllMarkers() for my clusters. In total, 12 clusters were found, and we are fairly confident about the identity of 10 of them. However, for 2 of the clusters, only a handful of genes are found, and don't really have good separation between groups:
DIFF | GENE
0.196 FADS2
0.173 PTPRF
0.17 TMA7
0.167 MT-ND2
0.164 MT-ND1
0.157 ERGIC1
0.156 SCD
0.153 SEC61A1
0.15 ASPH
In cases like this, where this aren't good markers for a group, what is the typical procedure? Lower the resolution of the cluster identifier and run again?
•
written
by
Dirk
2 answers
Every experiment is different, but you usually want to try multiple resolutions. For example, some smaller populations will not separate with lower resolutions, but larger populations will be split into multiple clusters with higher resolutions.
You can also change the thresholds for FindAllMarkers() if you just want more markers.
Clustering is always going to do its best to identify different population, whether they're actually biologically different or not. So it's also perfectly valid to say those two clusters are likely the same population/cell type if you don't feel the identified markers are robust/believable. When manually annotating, you're always going to have to deal with edge cases like this as they come.
Clustering will never perfectly identify all subpopulations, though as @igor mentioned, running with multiple resolutions should be a standard part of your workflow. Those 5 genes could represent a small subpopulation in one of the clusters that just isn't getting pulled out on its own, in which case higher resolution may help reveal it (and allow you to subsequently find more/better markers for them). Maybe.
Sure, if you feel that it's not worth separating. How granular people try to get with their annotations/subpopulations varies wildly, so you should rely on your expert knowledge of the biology/experiment to determine if that separation is valuable or meaningful.
Log in to answer this question.