Best workflow for evaluating proteins designed with NCAAs.

[tsamdin]

I'm currently using Rosetta to redesign a protein of interest that includes 7 non-canonical amino acids. The output structures I'm getting from Rosetta seem promising with regards to their scoring and by visual inspection, but I would like to further refine these models to the best sequence before going to the bench.

From the literature, it seems like the next step is to use RosettaFold All-Atom to evaluate how likely my given sequence is going to fold into the structure I get from Rosetta. Alphafold and collabfold don't appear to have readily accessible pathways for incorporating NCAA residues in the input sequence (please correct me if I'm wrong).

However, I'm unsure if I should be using RosettaFold All-Atom for this task, the publication and the github page say that work on NCAA residues is ongoing and should be possible, but I haven't seen any more recent publications demonstrating the ability (again please correct me if I'm wrong).

Instead of using RosettaFold All-Atom, would it be appropriate to run a simple protein in a box simulation using GROMMACS, to evaluate the thermodynamic stability and changes in CA RMSD to make decisions about a particular model/sequence?