after af2_initial_guess/predict.py, the distance between binder and target becomes much longer

[rocke2020]

Hi @nrbennet

thanks very much for your shared codes and paper!!
I use RF diffusion to generate, ProteinMPNN-FastRelax, and finally af2_initial_guess, all the default configs.
I test short peptide length 6-10 amino acids, and longer length 26-30 amino acids for the 2 different target proteins.

After run 1000 designs and all the same phenomena:
In the new pdb which is generate from af2 prediction, the distance between binder and target becomes much longer.
In the pictures blew, the green color is the pdb after ProteinMPNN-FastRelax; the cyan color is the pdb after af2_initial_guess prediction.

Questions:

1. is this noraml or designed on purpose?
   in my case, all the pae_interaction are between 22-28, too larger, that's low quality. When the designed structured from ProteinMPNN-FastRelax is low quality, the distance between binder and target becomes much longer?
2. Maybe some target protein structures are very hard to design good bindiers.
   could you supply a target pdb and binder design configs to let me have a pae_interaction < 10 for 1000 designs?
   thanks very much in advace!!
   thanks for your innovation on af2_initial_guess, a really good idea!!
   pdbs.zip

[Dan-Burns]

@rocke2020, did you find an answer to this?