Is it possible to construct a prior model from mouse data sources? about nichenetr HOT 2 CLOSED

Hi @mariusmessemaker

You raise a valid issue here. First, know that we also included some mouse data sources into the general model (at least for the one-to-one orthologs), but these were strongly in the minority. Because this model is indeed tailored more for human, you can expect that it would perform a bit better for human than mouse, but I don't think the difference is very large. In our paper, we evaluated the ligand-target predictions based on >100 ligand treatment datasets of both human and mouse origin, and there is no indication on these datasets that human would work better than mouse.

Of course, you can also explore the possibility of constructing a mouse-specific model with only mouse data sources. This is some work, but is not so difficult. In the vignette https://github.com/saeyslab/nichenetr/blob/master/vignettes/model_construction.md, we show how to build a model yourself. The main thing to consider though, is that you should have enough primary data sources of mouse to do this.

So, and this is valid for applying NicheNet to all other species other than human, you should make the tradeoff between the homology between species, and the number and quality of data sources that are available to build a new species-specific model. For example, for mouse, you could use the human model, but for Drosophila making a new model from own data sources would be more appropriate.

In conclusion: it is possible to build a prior model only from mouse data sources, but I doubt it would work better than using the human model.

from nichenetr.

Thank you for your extensive answer @browaeysrobin!

from nichenetr.