This can be done in many programs (In Modeller, for example, you can build an alignment covering both templates). Which doesn't mean the model will be accurate, because, if there's no template available covering all domains, there's no way an homology modeling program will guess the relative orientation of each domain. You could try protein-protein docking software such as HADDOCK, but keep in mind that this level of prediction (protein-protein docking on top of single domains which are already models themselves) will hardly be very reliable. If you're collaborating with an experimental group, it would be great if you could get a low resolution structure (using SAXS, for example, which is straightforward for monodisperse soluble proteins) and then fit the sub-units inside it.

One option could be to use IMP http://salilab.org/imp/ . This tool should help you through this process, but I'm not too familiar with it.

A more step by step option would be:

a) identify the 3D domains assigned to each part of your sequence

b) find if interactions between those domains apear in the database (may be through 3DID http://3did.irbbarcelona.org/ )

c) model each one of your domains separately (my program of choice would be MODELLER http://salilab.org/modeller/ )

d) use the known interactions between domains as a guide for the relative relation between your domains (i.e. if you have modeled a domA and domB and have a PDB interaction between domA' and domB', superimpose each domain with its homolog)

e) model everything again with the new relative position between domains using as templates your previous models in their new positions; this will add the linker sequences between them.

f) finally, you might want to optimize the new model.

Of course, there is no guarantee that you will find known 3D relations for each of your domains, but it might take you a step further.

I hope it helps.

J.-

Once you have identified protein domains in your sequence, I would continue with prediction of disordered / unstructured regions, then with prediction of secondary structure elements (a consensus between several methods would be preferred) and finally with fold recognition methods for each protein domain. This would hopefully identify templates for each protein domain. To build a homology model based on these templates, you will have to align the target and template sequences and then use programs (either free, like Modeller, or copyrighted, like Accelrys's Insight) to build the structure. But a correct sequence alignment is vital for the accuracy of the model.

for having a preliminary idea/model of your complex you can use the following automated-modeling webservices:

http://zhanglab.ccmb.med.umich.edu/I-TASSER/

http://modbase.compbio.ucsf.edu/modloop/

http://swissmodel.expasy.org/

if you would like to drive your protein modeling you should use the stand-alone versions of the cites program/webservices...Remember to cite the services and the related literature in the papers you will write :) ..

For more info about protein modeling issues, you can read this review:

http://www.ncbi.nlm.nih.gov/pubmed/20433957

enjoy :)

If you need specific support, don't hesitate to contact me, also by writing a private message :)

Take a look at Rosetta CM - there is a recent Structure paper out of the Baker group (first author Song, Volume 21, Issue 10, 1735-1742, 12 September 2013) . This seems to be exactly the kind of problem that the algorithm is designed to solve. They have a server (Robetta) for academic use or you can download the full Rosetta package. I have not yet gotten the multitemplated version working locally, but I'm working on it with the help of their forums.

you could try to model each domain seperately then use a protein docking server. The better models will have close N- to C - domain-domain distance

Try the I-TASSER server and then run MDS. But as allready mentioned by others there is no guaranty for success.

NAMD tool K make a good simulation to any protein, also VMD program may help u if i understand yr question well.

There are a few good answers here already. I would recommend using MODELLER (free to academics) to create the initial model (they even have a tutorial on their site for using multiple sequences to model a single, multidomain protein). That will give you a good rough start (depending on the quality of your homology sequences). Then you can use NAMD (also free) to do extensive minimization on the MODELLER output to get a usable model.

The answers note several methods that are reasonable in themselves. The problem is that whenever you model on top of a model, i.e. you are modeling the domains and then modeling their arrangement, the accuracy of the results becomes poor to terrible in my experience without some experimental feedback. I would urge you to do small angle x-ray scattering or some experimental method to provide restraints to your possible models, e.g. see Nature. 2013 Apr 25;496(7446):477-81. doi: 10.1038/nature12070 or at least test results by mutations or x-linking.

For a similar problem I used MODELLER from the sali lab (I think www.salilab.org or so). But then of course you don't know anything about the domain orientation towards each other if it has flexible linkers. There you would need some experimental restraints to use as input in e.g. cns/haddock.

Cheers

Janosch

for this case, in my opinion you should use an interactive method not automated ones. You use the latter in cases where you have exactly one experimentally solved protein and another sequence that you want to model.

An excellent package to combine different templates interactively is SYBYL-x.

You might want to take a look at the paper

R.M. Bhaskara, A. Padhi & N. Srinivasan (2013) Accurate prediction of interfacial residues in two-domain proteins using evolutionary information: Implications for 3-D modelling. Proteins (in press).

http://onlinelibrary.wiley.com/doi/10.1002/prot.24486/abstract

and try extending the ideas in this paper to model the multi-domain protein structure.

Most of the proteins have multiple domains. Can you elaborate how many amino acids are there in each of you 7 domains? 450 aa is not a very big protein. Which domain in it you find the template? You may try using PSI-BLAST to check each domain sequence similarity etc. to identify a closely related templates.

This modeling of protein with multiple templates can be best done using Prime module from Schrodinger 2012 molecular modeling software. In that, once the query is loaded and run for blast, the multiple templates showing similarity to the query sequence covering the maximum length of query are selected and aligned.

Once aligned, the template sequences should be selected totally the template will be considered for modeling. Once the desirable templates and query selected, the modeling should be proceeded using energy based.

Once the model is built, it should be evaluated using Ramachandran plot and ProSA analysis.

I would use Rosetta Docking. Rosetta will be significantly better than other docking approaches if there is any backbone changes upon joining the individual domains into a complex. I might use other approaches to narrow down the likely binding sites on the protein first if those are unknown. Rosetta should be able to help you confirm your answer after the fact by predicting disruptive point mutations you could test.

The first step is to identify the domains. This can either be done manually if you are familiar with the protein [1] or automatically using a tool like DOMAINATION protein [2] domain identification or ThreaDom[3]. The next step is to model the individual domains minus the connecting loops. The  I-TASSER server [4] is probably the best software for this based on multiple runs of the CASP blind modelling competition. Once the structure of the individual domains has been determined, they can be used as templates for the final full sequence. [5] Note that predicting the orientation of the two domains relative to each other is a difficult problem. Experimental constraints, for example the overall shape from small angle x-ray scattering SAXS-assisted protein fold recognition [6] or from NMR, can greatly assist in finding the solution.

http://www.sdsc.edu/pb/edu/pharm201/15/c20.pdf

Thank you very much for the in-detailed answer of the question.