how to inform modeller to reproduce this part, of just the whole
protein which as little deviation as possible (ie. to closely
resemble the template) ?
If you want it to look *exactly* like the template, you can simply
unselect this part of the protein:
(This will only work, of course, if you have 100% sequence identity in
this region, or can supply a good starting model.)
Otherwise, you can ask Modeller to include more restraints, which will
more tightly constrain the target backbone to resemble the template.
This was discussed recently:
Of course, this will only work if you have reasonable sequence identity
between target and template. If the identity is very low, then you will
probably have a hard job building a good structure using comparative
modeling in any case.