I'm trying to model a certain membrane protein. There are two template
structures with decent sequence similariry (~ 40 %) and not-so-great
resolution (~ 3 angstrom).
When building a model against either template nothing in particular
sticks out in the logfile, but when building a model model against both
templates combined there are lots of distance violations (CA-CA and N-O)
in the log, however this model gives the lowest DOPE score.
This sounds reasonable. When you have two templates, you're going to
have lots of bimodal restraints. When one template says a given distance
is 10A and the other says 20A, Modeller doesn't restrain that distance
to 15A, but to either of those two minima. So whichever one it picks the
other will be violated.
DOPE scores the model on it being "native protein like" without taking
the templates into account, so those violations aren't going to bother it.