Masked target structure support

[biochorl]

Thanks for this protein design tool.
I was wondering if there is the possibility to set a config option to mask the structure
of the target or the input PDB needs to be fixed (soft/hard?) as start condition for the generation/evaluation

[nvdreidenbach]

The target is a hard requirement. Are there certain binder design applications where you would not want to condition on the target?

[biochorl]

I intended how to deal with the situations when the target is a part of a PDB structure, but you Need to target a flexible region not solved structurally and with an inaccurate predicted conformation, or when is completely disordered, like a Linear peptide

[biochorl]

I can add after some tests that also side-chain conformation is fixed for the conditional binding design, which could cause problems in case you do not work with targets having the holo conformation to target. What do you think? Would it be possible to add some soft conformational variation on the target by operating on the partial latents every tot steps?

[nvdreidenbach]

Yes the current model works by taking in a frozen all atom target. We are thinking about relaxing this constraint in the future.

In the meantime we had success designing sugar binders by changing the ligand conformation used by the model during each step in the search process. If you have multiple conformations you could try similar strategies but this is not implemented currently in the code. As for partial targets if the residue of interest is at the binding interface there likely is not a quick fix as the model was not trained with chain breaks at the interface.