Prioritization based only on ligand activity (ignoring ligand/receptor expression) in MultiNicheNet

[donghyeon321]

Hi,
I would like to ask whether it is possible to modify the prioritization strategy in MultiNicheNet to rely exclusively on ligand activity (NicheNet-based target prediction), while completely ignoring ligand and receptor expression–based criteria.

My specific goals are:

1. I want to prioritize ligand–receptor interactions only based on their downstream target activity, not on:
   • differential expression (DE) of ligand/receptor
2. The reason is that in my dataset:
   • I am specifically interested in cases where ligand/receptor expression does not change, but target activity differs (i.e., hidden regulatory layer beyond expression)
   • There is also a large batch effect, so I prefer using MultiNicheNet instead of standard NicheNet
3. I am aware of the scenario = "lower_DE" option, but:
   • this only reduces the weight of DE criteria
   • I would like to completely remove all expression-based criteria, not just down-weight them

Specific questions

• Is there currently a way to:
  • disable ligand/receptor expression, DE, and fraction-based criteria entirely, and
  • compute prioritization scores using only ligand activity?
• If not, would modifying the internal prioritization step (e.g., setting weights of all non-activity criteria to zero) be a valid approach?
  • Are there recommended functions or code sections where this can be adjusted?
• From a methodological perspective, would the authors consider this type of prioritization valid or meaningful within the MultiNicheNet framework?

Thank you for your time and for developing this valuable tool.

[rseurinc]

Hi donghyeon3,

Thank you for your question. This certainly can be meaningful, especially if you think that there is an issue with the capture efficiency of your ligand in your data. Note that the ligand activity is still based on DE, in the sense that you calculate the DE between conditions in your receiver cell. If you have batch effects in your data, MultiNicheNet allows you to correct for them, provided that your experimental design enables this. So the DE could more accurately reflect the actual DE between conditions than in the regular NicheNet.

It is easy to retrieve a ranking solely based on the ligand activity. If you calculate the prioritization table using generate_prioritization_tables, the resulting table will contain a score for each individual criterion and for your purpose you can use the column max_scaled_activity to do the ranking. Note that you will get the same scores within a contrast for all ligand_sender_receptor_receiver combinations that have the same ligand and receiver cell, as the ligand activity just reflects what upstream ligand can best explain the DE in the receiver cell.

You could also add a new scenario with setting all other weights to 0, but this would require adapting the code.

Hope this helps,

Ruth

[donghyeon321]

Thank you for replying,
I have several questions about interpreting NicheNet results.

Setup

• Receiver: cell type B / Senders: all other cell types
• Ligand of interest: expressed in cell type A, receptor expressed in cell type B

Results

• Overall ligand ranking: ~60th, focused ranking: ~30th
• AUPR_corrected: relatively low
• But 9 target genes linked to this ligand are all significant DEGs (8 down, 1 up)

---

Q0. Is batch-corrected DEG input acceptable for NicheNet?

I used MAST with batch as a covariate to generate DEGs, then used these as input for NicheNet. Is this a generally accepted approach?

---

**Q1. Low AUPR but strong target gene evidence — still meaningful?

The ligand ranks low (~60th overall), but all 9 linked target genes are significant DEGs. There is also reason to believe this ligand has protein-level functional differences between conditions (e.g., isoform switching affecting functional domains) beyond what gene expression captures.

However, I'm also aware that these target gene changes could be coincidental, or driven by other higher-ranked ligands rather than this specific one. NicheNet doesn't distinguish between these possibilities.

Can this interaction still be interpreted as biologically relevant despite the low AUPR/ranking, especially when additional functional evidence exists? And is there a way to assess whether the target gene changes are more likely attributable to this ligand specifically, rather than being a byproduct of other signals?

---

Q2. How to connect ligand activity to the direction of target gene changes?

NicheNet's AUPR doesn't provide directional information. My 9 target genes show a consistent directional pattern (8 down, 1 up).

I want to argue that this directionality is consistent with the proposed change in ligand activity between conditions — not just that these genes happen to be DEGs.

What analysis strategies would allow me to validate this directionality? For example, in silico perturbation (CellOracle, GEARS) to predict whether KO of this ligand pushes target genes in the expected direction, or checking known downstream pathways. Is there a recommended approach?

---

Q3. Receptor expression vs ligand activity pointing to different receiver cell types

• Receptor expression: low in cell type B, high in cell type C
• Ligand ranking in each cell type: higher in cell type B, lower in cell type C
• All three cell types are physically adjacent

I want to focus on the A → B interaction. Is it valid to prioritize ligand activity over receptor expression level when deciding which receiver cell type is more biologically relevant?

---

Q4. What level of causal claim is justified?

Given all of the above (low AUPR, strong target gene evidence, directional consistency, possible protein-level differences), is it appropriate to claim this ligand–receptor pair causally drives the target gene changes? Or should it be limited to "plausible upstream candidate consistent with observed changes"?

What would be the minimum additional validation needed to strengthen a causal claim?

Any thoughts or pointers would be really appreciated, thanks!