Graph iLISI gives a super low score for well-integrated data

[dariotommasini]

Hi scIB team,

Thank you for a great tool!

I had a question on ilisi_graph. I ran as such:

scib.me.ilisi_graph(sm.samap.adata, batch_key="batch", type_="knn")

I get a low number, indicative of poor batch mixing. Here is my kNN graph:

This graph is sorted by type, then by batch. This should be a good integration e.g. high cLISI and high iLISI. Can you think of a reason why this kNN graph would receive a low iLISI? It does indeed receive a high cLISI. Note that I used an integration method (SAMap) that deliberately lowers within-batch connectivities. However, even considering that, the neighborhood of each cell seems to include multiple species, but I get a iLISI score of 0.027. Thank you for the help!