Actually I used compare.py as a means to narrow down the list with the 8
potential templates, not to generate an alignment for comparative
modelling.
That's what I figured - compare.py won't give you a useful modeling
alignment. But there is no requirement to compare your templates with
compare.py either.
Certainly my option 3 (letting Modeller align the sequences) is not a
good idea as T-COFFEE's method is more sophisticated and incorporates
structural information. So I'm between 1. and 2.
Right - align() is the simplest alignment method, so will not be
competitive with anything other than simple sequence-sequence alignment
methods. So feel free to use T-COFFEE or one of Modeller's more powerful
alignment methods, such as SALIGN.
1. edit my T-COFFEE-generated consensus alignment by leaving only the 4
sequences I need, and create an alignment for Modeller
2. generate a new structure-based alignment with T-COFFE with only the
selected 4 sequences and feed it to Modeller
Why not build both and see? (1) sounds better to me, but it's impossible
to be sure.