I think what you propose is reasonable, but you will also want the same overall structure. That will really depend on the kind of TM protein you are looking at. If it is a [?]GPCR[?] for instance you will expect not 1 but 7 transmembrane domains. You can still start with a homology search but you will also want to determine what kind of domains are present in the protein, for instance using [?]Interpro[?].

If you have a 3D structure of the TM model you want to compare with, you will also want to study how that changes when you change the sequence in more depth. You can for instance do that with [?]WhatIf[?].

But maybe what you really are after is 3D structure comparison, for which you could use [?]3D-BLAST[?], or dedicated programs like [?]Matras[?].

Update: This paper "[?]GOSSIP: a method for fast and accurate global alignment of protein structures[?]" was published recently and could be useful as well.

You may find this paper interesting: PRALINE™: a strategy for improved multiple alignment of transmembrane proteins

I do not know ifyou are familiar with Discovery studio. If yes, then there is separate technique to model transmembrane proteins like GPCRs. Refer Webinar series by Accelrys. Here they have explained the steps for modeling a GPCR protein. In addition, I would also suggest a review Critical assesment on homology modeling of membrane proteins (if you need a copy of it email me kipokh@utu.fi)

MODELLER from the Sali Lab is very popular for both soluble and membrane proteins: http://salilab.org/modeller/modeller.html

The recently released and creatively named MEDELLER is an online tool and might be an easy alternative: Paper: http://bioinformatics.oxfordjournals.org/content/early/2010/10/04/bioinformatics.btq554.abstract.html

Online server: http://portal.stats.ox.ac.uk/userdata/proteins/medeller/gui/home.pl?app=MEDELLER

Try multiple template......... May be you can have a look at I-Tasser serever if you are not so found of Modeller software (command line) and then evaluate the model through SAVS server.....

This is the simplest two step procedure for automated homology modelling ;).

Protein modelling depends upon the quality of target-template alignment result.

Wish u luck