Normalization with cellranger count filtered gtf

[asilvestris84]

Dear Seurat developers, I have a question regarding the pre-processing step: I have always seen that 10x recommends using cellranger with a filtered version of the transcriptome annotation containing only the protein coding, long non-coding and antisense RNA genes; I found instead that, since the capture system of the polyadenylated fraction is not perfect, there are many other types of RNA sequenced and also at high copy number in a normal 10x single cell gene expression experiment (for example MT-RNR2). My question is what is your point of view on the fact of using a filtered annotation and if the fact of not considering the RNAs sequenced but "that should not be there" can create a bias in the normalization with Seurat for sequencing depth. Thank you!