Is it reasonable to keep only protein-coding genes in highly variable features in scRNA-seq data?

Question

[carolofharvest]

Hello,

In my human cell line data I have a lot of psedo-genes and Seurat::FindVariableFeature() function also includes them into highly variable features. I noticed some feature selection tools removes mitocondrial, ribosomal and pseudogenes from variable features so they cannot affect clustering.

What would be the best course of action?

Answer 1

[ATpoint]

Why would you exclude them? It's genuine biological signal. Just because we/you don't know what these genes are doing doesn't mean that they are not valuable to drive separation between cells. An argument to remove them though would be that they have no annotated function, thus difficult to explain, but on the other hand, many protein-coding genes have no bona fide function as well. Just leave them.