WNN neighborhood graph construction failing

Hi everyone,

Currently facing an issue where I am trying to run WNN in muon to integrate RNA and ATAC. The problem is that the neighborhood graph calculation is failing because there are cells in the ATAC graph that appear to have no neighbors. After looking at the source code for the WNN implementation in Muon, it looks like the algorithm expects every cell to have a neighbor. This makes sense considering how the neighborhood graph is calculated, so I can’t figure out why some cells wouldn’t have any neighbors. This problem does not appear to occur with RNA data. Appreciate any insight, thanks!

Hey @abbass2,

I don’t know if it’s still relevant but we’ll be happy to debug it if we could have a sample of the data to reproduce that. Also, was the neighbourhood graph for the ATAC modality re-generated after cell filtering?