Provide predicted aligned error (PAE) for locally folded proteins

[jolakopec]

Would it be possible to output PAE for locally folded proteins?
I know, that it is possible to get PAE from the ATLAS, but how can I get pae for proteins folded on my computer?

[naailkhan28]

You can get the PAE using this code snippet (adapted from the ESMFold ColabFold notebook):

with torch.no_grad():
        output = model.infer(sequence)

#Retrieve PAE values from the model outputs
pae = (output["aligned_confidence_probs"][0].cpu().numpy() * np.arange(64)).mean(-1) * 31
mask = output["atom37_atom_exists"][0,:,1] == 1
mask = mask.cpu()
pae = pae[mask,:][:,mask]
np.savetxt("filename.txt", pae, "%.3f")