I'm using Modeller 9v2. My protein is a homodimer, and my template is
also a homodimer. There is one molecule of AMPCPP and 2 metals bound to
each chain of the template. I adjusted the PDB file (2BW7) so that the
metals and nucleotides followed directly after the appropriate chain
TER's. I should add that the modeling runs fine in the absence of
nucleotide.
You could try using the BLK residue type (.) instead of GTP (y) in your
model sequence - it is entirely possible that the CHARMM forcefield
parameters for GTP are not doing a great job. As for your metal, your
model contains MG but your template contains CA, so since you have no
equivalent template atom, the MG gets no homology-derived restraints on
it and it can wander (Modeller is not smart enough to recognize
automatically that one metal could be considered equivalent to another
here.) To fix that, you could either create manual distance restraints
on your metal ion to hold it in place, or edit your template PDB and
change the metal ion there to MG.