All the element in OT matrix are the same

[DruncBread]

Hi,

I encountered a problem when running PASTE:
pi_temp = pst.pairwise_align(adata_list[i+1], adata_list[i], backend = ot.backend.TorchBackend(), use_gpu = True,numItermax=3000)

All the element in the OT matrix pi_temp are the same, which means the mass was devided equally.
And the alignment result was not good.

But when I set the param alpha to 0, the problem solved but the spatial information was not used.

I noticed that even I set alpha to a small value like 0.01, the problem above still occured.

It would be great if you could help me solve this problem! Thank you!

Best regards.

[RonZeira]

It can happen when the optimization gets stuck in a local minima given by the initial equal mass distribution. One workaround is to initialize the alignment with an alignment which is not equally distributed. For example, solving the problem using either just the expression or just the spatial coordinates to get an initial alignment and then using this alignment as a starting solution to the combined optimization.