I have two sequences with low sequence identity and I get an alignment
in modeller different than clustalW, which also differs from when I use
emboss matcher.
This is unsurprising, since different algorithms are likely being used
here to make these alignments.
I also have a missing segment in my template, I want the final model to
have the corresponding segments and the gaps unmodeled, same as the
template.
If you don't want part of the sequence in the model, exclude it from
your alignment. Modeller only models the sequence you give it. You may
need to add a chain break (/ character) to the alignment to prevent
Modeller from trying to connect the residues on either side of the gap
with a peptide bond.