thank you for your suggestions! I have some follow-up questions though.
1. Did you mean some simplified replica exchange with MODELLER or a real replica-exchange simulation? Because simple temperature replica exchange is most probably not feasible for a system of this size - number of replicas necessary to achieve the same exchange rates grows as ~ sqrt (N) and will be too high in this case.
2. As for increased temperature MD for loops - you imply that I keep the structure restrained and only let loops relax? I was under the impression MODELLER's refinement does something of this sort in CG model, which should be a lot faster than AA MD anyway.
3. About the normalized DOPE - again, I was under the impression that it only matters when you are comparing different proteins. My task is only to compare different models of the same protein.