Question about running pathways activity inference

[MigleMi]

Thank you for this cool tool!

I have a question regarding running pathways activity inference for scRNAseq data. I was wondering if it is only appropriate to run mlm method or also ok to run decoupleR::decouple. I tried both and seem to get a bit different results and even more what I would expect with using decoupleR::decouple with default parameters. Is that only ok to do for TF analysis or should I change something for pathway activity specifically?

Many thanks,
Migle

[PauBadiaM]

Hi @MigleMi,

Personally I would use always ULM for its simplicity and reported predictions scores in our benchmarks (see decoupleR's manuscript). MLM is nice because it can model interactions between pathways but this is a problem when they are too overlap, which produces unexpected scores. decouple is not a method by itself, it just runs multiple methods sequentially.