My other question is: what if the profile score for each of the residues
get lower (after MD simulation) but the global score actually increases?
How could we compare the two models then? Would you say one is better
than the other because of (global DOPE or profile)?
You could perhaps argue that the structure of that part of the protein
with a lowered DOPE score has improved, at the expense of some other
part of the protein. But DOPE was designed and benchmarked primarily to
score proteins, not individual residues, so I'd say the more reliable
indicator is the global DOPE score.