Description of feature
Hi,
I'm pretty new to the field so I apologize in advance if this doesn't make sense. When inferring paracrine signals or anything that diffuses across a neighborhood graph, I would think intuitive that spatial connectivities be weighted (assuming interaction is stronger when a receptor is closer for example). LIANA+ does this using radial kernels (Gaussian by default) and a cutoff. Historically, it seems that LIANA implemented its own version of spatial_neighbors mostly to remove the squidpy dependency but I'm left wondering how much the two functions now overlap. Could it be interesting to implement this? I guess this is related to how the graph is pruned when passing the current radius argument to any type of graph (Delaunay, KNN, radius..).
I'm doing that by hand for now, let's see how it goes. I would very much appreciate any feedback :), thx!
Description of feature
Hi,
I'm pretty new to the field so I apologize in advance if this doesn't make sense. When inferring paracrine signals or anything that diffuses across a neighborhood graph, I would think intuitive that spatial connectivities be weighted (assuming interaction is stronger when a receptor is closer for example). LIANA+ does this using radial kernels (Gaussian by default) and a cutoff. Historically, it seems that LIANA implemented its own version of
spatial_neighborsmostly to remove thesquidpydependency but I'm left wondering how much the two functions now overlap. Could it be interesting to implement this? I guess this is related to how the graph is pruned when passing the currentradiusargument to any type of graph (Delaunay, KNN, radius..).I'm doing that by hand for now, let's see how it goes. I would very much appreciate any feedback :), thx!