Local docking simulation diverges leading to an unfolded protein; Rosetta version used?

[edikedik]

Dear authors,

After running global docking using the protocol from here and using the Rosetta public 3.14 release, the simulation completes successfully leading to eight different complex orientations with comparable scores.

Proceeding to the local docking, I'm experiencing severely abnormal behavior, with the score ramping up quickly.

Sufficiently long simulations (>5mln steps) even lead to partially unfolded chains way outside of the interface region. This problem was raised in the Rosetta Commons forum but received no answer. Eliminating potential causes I reached the same conclusion that backrub is the culprit here. (I must also note that defining atom pair constraints on the interface residues and reducing the temperatures helps to keep the structures close but still leads to clashes and diverging energy).

My wild uneducated guess would be that the MUDS scoring terms do not play nicely with the new Rosetta versions. Do you think it's plausible? Which Rosetta version was ReplicaDock2.0 developed on? I didn't find it in either of the related papers. Of course, any troubleshooting suggestions are welcome!

Cheers,
Ivan

EDIT: My initial hypothesis was wrong: MUDS is not the reason behind this as changing it to ref2015 has the same effect. The issue is thus likely with backrub when used in MC context.