Movement is a tricky question. Wide-scale sampling is very tough in this context, because local-only movement of the receptor side of the interface will break the chain and/or cause lever arm effects unfolding the opposite side of the protein. Loop modeling the interface won’t cause that problem, but of course requires that the interface be made of loops.
Minimization of the interface is more achievable. This application will do peptide + minimizable backbone docking: http://www.rosettacommons.org/manuals/archive/rosetta3.4_user_guide/d9/deb/doug_dock_design_min_mod2_cal_cal.html I think this one may have hard-coded assumptions about the system we’d need to get around.
You can also try doing FlexPepDock against a relax-generated ensemble of your receptor, or running FlexPepDock results through relax mode.
.
