I would like to refine my models of a homopentamer with loop refinement.
In the past (Mod8) I would use loopmodel, but noticed that there is an
"improved" loop modeling routine dope_loopmodel() which uses (among
other things) the DOPE score in the refinement process. In various parts
of the manual it mentions that the DOPE score was developed for
single-chain models and shouldn't be used as an evaluative tool for
multichain models. Does this also meas that the dope_loopmodel() routine
shouldn't be used for loop refinement of multichain models? What's the
downside of using it for multi-chain models?
As with any statistical model, you should be careful about using it for
systems not covered by the original training set. DOPE actually appears
to work rather well for scoring multiple-chain models, protein-protein
interfaces, etc. but I'm not aware of any comprehensive published
benchmark for such systems, hence the warning in the manual. And yes, as
a method which uses DOPE, dope_loopmodel would be subject to the same
warning.
On the other hand, as far as I recall the 2000 Prot Sci loop modeling
study (loopmodel) also did not include any multi-chain models in the
derivation of its statistical potential...
If so, this would be a shame since there are also other enhancements
(Leonard_Jones potentials, etc.) unrelated to DOPE that would be nice to
be able to use.
Regular loopmodel also uses a Lennard Jones potential. But it is rather
straightforward to customize the loop modeling method yourself by
subclassing loopmodel if you want to play with different scoring terms,
etc. This is exactly what dope_loopmodel does - see the
modlib/modeller/automodel/dope_loopmodel.py file.