I am exploring the method for neighbour graphing of multiplex imaging data. Normally I would run scanpy.external.pp.harmony_integrate to combine a large number of samples then sc.pp.neighbours(use_rep=“X_pca_harmony”) for the neighbour graph . Does phenograph need to be directed to the modified “x_pca_harmony” to correct for batch effects first?

It looks like it uses whatever is in adata.obsm["X_pca"] and there’s no way to set something manually so you’d need to move things around in adata.

Alternatively you can use leiden clustering sc.pp.neighbors and then

We had some feedback from a developer that neighbour graphing by sc.pp.neighbours for multiplex imaging data wasn’t a great fit and that phenograph gave better results. Are there any scanpy authors here who might be able to comment on the sc.pp.neighbours algorithm compared to phenograph?

You can always just use the actual phenograph package.