Inference of the Receptors for unknown Ligands

[dslv3y]

Dear NicheNet developers and maintainers,

Thank you for developing the framework! I seek some advice and would greatly appreciate any help. My data is a perturbation snRNA experiment with some uncharacterized ligands (one per experiment with 4 biological replicates, plus control). My main goal is to retrieve the activated receptors. I wonder what would be the best practices to use NicheNet for such analysis?

My idea is to just run sender-agnostic analysis with all possible ligands as a "set of potential ligands" and then retrieve top-ranked receptors. Is this a good approach? Maybe I somehow could use an artificial LR network (one-to-one connections receptor-"pseudoligand") for the prior "pseudoligand"-target network construction?

Another thing is that I would like to add more membrane proteins as receptors, for which there are no known ligands, but there are signaling pathways. So I wonder if I could introduce artificial weights for such "receptors".

P.S. There is a RIDDEN tool, which is applying the receptors-targets matrix to the data, but it's a bit different approach, I would also like to use NicheNet.

Thanks a lot in advance!

Best,
Dmitry

[Eisuan]

Dear Dmitry,
Thank you for your interest in NicheNet and apologies for the delay in our reply.

My main goal is to retrieve the activated receptors. I wonder what the best practices would be to use NicheNet for such analysis?

This application could be an exploratory analysis that might help with further experimental validation using methods suited to assess the involvement of the prioritized receptors. However, this use case is stretching beyond the experimental scenarios used to develop and validate NicheNet, as the whole approach focuses on ligand-target gene associations.

You could prioritize ligands using a sender-agnostic approach, then consider the related receptors for further validation.

Another thing is that I would like to add more membrane proteins as receptors, for which there are no known ligands, but there are signaling pathways. So I wonder if I could introduce artificial weights for such "receptors".

You could definitely add some resources based on your needs. However, you are describing a scenario in which you do not have a characterized gene expression footprint for those Ligand-receptor pairs, which is the type of information that seems to be the most informative in the context of our framework. Thus, I am not sure that re-leveraging the current network with this type of information would lead to a useful model.

Bests,
Daniele