Entering edit mode
8 weeks ago
D
•
0
Hi! After I used DoubletFinder to detect doublets in my dataset, I found out that some clusters have a really high percentage of doublets, for example, over 50%. Should I remove those clusters entirely since all the cells in those clusters might be doublets, or should I just remove the cells that are detected as doublets?
Look at nCount_RNA of cells in that cluster. Does it appear to be significantly higher across the entire cluster? Typically normalisation and scaling before clustering tries to correct differences due to depth/counts so it isn't a given that all the doublets will cluster together.
In addition, look at the markers for that cluster. Do they contain clear indications of 2 (or more) different cell types? If so, probably safe to remove.
However, on occasion, some of the doublet-finding algorithms can label legitimate transitory populations as doublets, so it's worth taking a closer look at them.