Hi, the problem was solved by deleting "inifile = 'mgl.pdb'" in my
script. mgl.pdb is my template pdb file, without any further change. If
you still think this is not normal, I'll send the input files towrrow
when I get to lab.
If you specify an "inifile" it needs to be a set of coordinates of your
model (usually from a previous optimization), not a template. If you
leave out "inifile", then Modeller will automatically build an initial
model (with a .ini file extension) based on your template(s) and the
alignment. This is usually what you want, although sometimes the .ini
file has such a bad structure (e.g. knots in the sequence or other bad
clashes) that the optimization is unable to recover, and you may be able
to manually tweak the initial model to correct this.
Unfortunately, Modeller does very little checking on a user-provided
initial model, and expects it to match the .ini file (i.e. same residue
and atom order, just with different coordinates). So you can get weird
errors if this is not the case.