search

carmonalab / GeneNMF

Methods to discover gene programs on single-cell data
85 stars 3 forks source link

Mitochondiral genes removal? #16

Closed nandobonf closed 2 weeks ago

nandobonf commented 3 weeks ago

Dear authors, thanks for developing this very useful package. I am currently analyzing a set of 33 samples QCed same way with GeneNMF default parameters but i get many meta-programs related to mitochondrial genes. Would it make sense to you to filter out these genes before hand from the seurat object? Or for example, do you suggest using another feature selection strategy? Thanks in advance for your reply!

mass-a commented 2 weeks ago

Hey Ferdinando, yes I think it's reasonable to filter out genes from specific compartments (e.g. mitochondrial or ribosomal), if there are reasons to believe they may mask more interesting biological signal. In fact GeneNMF has a hvg.blocklist parameter to exclude specific gene sets from the feature selection. You can also have a look at SignatuR for some precompiled blocklists. Cheers -m

nandobonf commented 2 weeks ago

Thanks for the swift reply! Now, I'm in the phase of filtering to get high-confidence MPs. In the original paper, MPs are filtered based on genes (>5), silhouette and sample coverage. Once identified these high-confidence MPs, Is there a way to easily subset the output of getMetaPrograms()? It would be very useful for the heatmap plotting without hard coding the similarity matrix etc. Thanks again!

mass-a commented 2 weeks ago

Hi again, I agree it would be good to have a way to easily subset the output of getMetaPrograms(), others have also requested this feature. We will work to implement a function to drop MPs from the object, hopefully soon. I'll keep you posted.

mass-a commented 2 weeks ago

FYI there is now a dropMetaPrograms() function in GeneNMF v0.6.2 (on GitHub, soon to CRAN) that allows removing MPs from the GeneNMF results. The subset object is also understood by the heatmap plotting function. Cheers -m

nandobonf commented 1 week ago

Awesome! Thanks for the enhancement!