This would be a tall order. Like Felix said, you're best hope would be to find a model of a homologous multi-protein complex that you could use as a template for building a predicted model. If such a model does not exist, you could NOT perform de novo predictions of how a multi-protein complex assembles, even in the case where a reasonable model for the structure of each subunit exists.

A fruitful approach for modeling multi-protein complexes that cannot be studied directly by X-ray crystallography has been to perform docking studies in cryo-EM images of multiprotein complexes. The minimum criteria you need are high-resolution cyro-EM, and structural models for each subunit.

I think no. Tertiary structure does not depend only on the amino acid sequence but also on the solvent, pH, ionic strength,...As a first approach, probably you can try to simulate your specific proteins with a computer code as GROMACS speciyfing all these conditions

I think the curical question is, if there are already some homologue proteins with already known 3D structure available. In the ideal case for your whole complex. In the not so ideal case only for single domains. In the latter case, you would have to exactly know how the complex is assembled and reproduce this. Did you already do a quick search in the PDB? If available, homology modeling (maybe also threading) could give you good results.

Overall, this sounds like a very huge system and a very complex task to do.

This would be a tall order. Like Felix said, you're best hope would be to find a model of a homologous multi-protein complex that you could use as a template for building a predicted model. If such a model does not exist, you could NOT perform de novo predictions of how a multi-protein complex assembles, even in the case where a reasonable model for the structure of each subunit exists.

A fruitful approach for modeling multi-protein complexes that cannot be studied directly by X-ray crystallography has been to perform docking studies in cryo-EM images of multiprotein complexes. The minimum criteria you need are high-resolution cyro-EM, and structural models for each subunit.

I would mostly agree with Daniel and Felix - what you are asking will be very challenging, and possibly impossible.

I would worry first about whether or not you can model your individual subunits. Making a good model is a skill in its own right, and you will probably need to find someone who is quite good at it to help you (especially to determine how reliable your model is).

Only then, if your individual models are OK, can you attempt to build a ternary complex. Definitely, this will be much easier if there is a homologous complex available. Although (here I will disagree with Daniel), it is possible to make a de novo prediction of how two proteins will interact (this is protein-protein docking, and is assessed at the CASP competition). However, this is undoubtedly expert territory, and you would need to collaborate with a specialist group if you are to have any chance of getting a good result. You would also need to have some pretty good experimental evidence to back up your model if you are going to convince anyone that it is accurate.

I think that your best bet would be to find a potential collaborator locally, and let them in on what your proteins are, so that they can assist you in determining how doable this is.

@Nicholas- haven't been following CASP so closely in the past few years. Has anyone succeeded in docking protein-protein interactions for membrane proteins? If so, then I retract my statement on the feasibility of such an approach, even if the state-of-the-art is limited to binary pairing of proteins. It is easier to envision that these docking program can succeed with small, cytosolic proteins, but I didn't think the field could yet tackle membrane proteins.

Either way, I fully agree that this effort would require extensive collaboration with experts in the field of homology modeling and protein structure determination.

@Daniel - I think that Uttam is only (!) looking to get a complex of his three cytosolic proteins (I might have misinterpreted). I agree that membrane proteins are likely to be at least an order of magnitude harder to dock, as they are with everything else.

A ternary complex of two proteins is likely to be challenging enough to produce a convincing model of, I agree!

Yes it is possible! Reliable depends on the sequence homology of course with proteins of known structure. You can use SwissMOdel, and also CS-ROSETTA (for this you may need access to some special databse - NMR related). For the complex is more complicated, but you can also try: CHEMERA (I am a user, and close to the lab where it was/is developed). You fsrt have to have the models of the proteins, and in case you have some evidence/experimental data where the proteins will interact is easier to filter the solutions. For a ternary complex, you first need to have the docking of the two proteins and then dock the third. It is possible, but of course the analysis and filtering of the solution is more difficult.

Try also to see in the Capri webpage.

@Nicholas- yep, missed that point. Cutting out the membrane receptor from the picture, I agree that docking could work in theory.

@Nicholas and Sofia- can you point to some references showing successful modeling of protein complex in an analogous situation (starting only with primary sequence and structures of homologous proteins)? This might help Uttam assess the feasibility of these modeling approaches.

Can look at mine. And look up for the references for the program.

Rossetta could be the best choice at the moment: http://robetta.bakerlab.org/

Thank you Nicholas, Daniel, Sofia, Felix and Lago. I realize that the modelling would be complicated and the result might not be promising at all. There could be a domain swapping between two proteins and that will make the whole thing a lot more complicated.

Structure of some of the domains of the homologs are available and homology/ab initio modelling of the individual proteins could be possible. But modelling followed by successive docking would definitely decrease the statistical significance of the resultant model in each step. I agree it would have been better if the structural information of the homologous complex were available. Cryo-EM is a good option but it will be a whole project on itself.

@Sofia – I used SwissModel before. I'll try Rosetta. And did you mean UCSF-Chimera?

@Daniel - I don't know any of the top of my head. I was involved in successfully docking a model to a crystal structure back in 2003, and I know that the methods have improved since then.

I think that, as ever, the question will depend on how good an alignment can be made to extant structures.

Dear Uttam,

It was really informative to read this entire discussion, especially changing positions of some of the people with learning new details. As with every science problem it turns out that with some modifications something is doable after you hack at it, and hack at it.

My only 2 cents here is that over the years people developed methods for the exact task as you are facing that were not mentioned here.

1) There are several databases on the web that describe the protein-protein interactions on many different levels including details (Biogrid, String, Mips, Pips etc.)

2) There are sophisticated protein-protein docking methods, some of them quite successful (Rossetta, Zhang)

3) There methods that allow you to map the most likely interacting surfaces by looking at remote homologies and co-evolution (i.e. Lichtarge, from Baylor)

4) There are methods for using any experimental constraints and building the complex (Sali, San Francisco, Gerstein, Yale)

5) You can use some of the newest ontology derived methods of association that use the web info normally not integrated by other sources.

But, I generally agree this is a very difficult problem. Modeling even a single component would be a challenge. My experience with membrane or peripheral membrane proteins is that they undergo significant conformational changes upon binding, so you would be stepping into a substantialy new territory by predicting one of the active conformations of these components, selecting it and using it productively. This would be like executing the decision tree before you know how it looks like. How would you decide which of the models (if you get several models for contributing factors) to use in quaternary assembly? Proteins are not stiff, they are mobile and malleable. Good Luck.

Hi Uttam,

Hi Uttam,

Unfortunately, this is hardly possible due to fundamental limitations. Even in 'simple' chemistry, a primary structure does not say you much about the realistic structure in the *specific environment*. Hydrogen bonding, atomic tunneling, rotational barriers, and many many other things make this impossible. Detailed quantum calculations, computationally possible for small proteins, are prohibitive in terms of zillions of 3D configurations to be sampled to minimize energy. Also, for highly flexible molecules like proteins, the energy calculations themselves do not make much sense due to the degeneracy of states of protein molecule when a multitude of states have the same (or very close) energy and form "bands" (like electronic bands in metals). Protein dynamics is a new challenging world... Simple estimates are possible, of course, but they are hardly very helpful.

Unfortunately, i cannot suggest something constructive and ready to use. I think the coming developments in protein dynamics will deal with the collective modes, quantum effects, in particular, for atoms H in the numerous hydrogen bonds, degeneration phenomenon for protein conformations, atomic tunneling under various barriers, etc. I think such an in-depth approach will be able to explain the nature of denaturation and similar collective phenomena.

Due to many conjugated bonds in aromatic amino acids a native protein may hold an extra energy and be in a coherent/collective state via pi-electrons. Having an extra energy, a protein is a reactive particle, and so a catalyst, an enzyme - in biological terms. Also, a native/living coherent protein is a unique optically transparent non-dispersive structure. Denaturation is a result of destruction of highly coherent/collective state of a protein, when a protein loses its extra energy with its conversion into a ‘regular’ quasi-crystalline structure incapable to accumulate and reuse energy.

I think the protein dynamics should probably be approached in the context of *expression* processes as well. In reality, proteins are not isolated (macro)molecules; they are the result of expression dynamics controlled by genome, and so they get developed and 'immersed' into the 3D 'structured' matrix (in cytoplasm, nucleus, etc), their real environment. The proteins catalytic properties (in enzymes) is also an indication to involvement of electronic states controlling reactivity as well as involvement of metal ions (Cr, Fe, Cu, Zn) and of sulfur S, phosphorus P with reactive d-electrons.

The proteins are a different beast from another perspective as well: a protein likely a most ‘suitable’ organic molecule to keep/reuse energy as its flexible 3D structure has lots of soft degrees of freedom. Folding, sheets, globules, etc configurations may be the trial ones (in selection dynamics) to check what configuration has the best energy reuse capability in the specific environment. Hydrogen bonding is likely a key to the proteins properties and dynamics.

So, the "context" for discussion and modeling of protein dynamics should probably be wider. It will take time...

Wish you the best,

Val

Dear Uttam

Modelling such a complex will be difficult and not very reliable. We did work on a similar problem a while back where we had to model 2 different proteins forming a complex and linked to each other by a short linker. Our models were able to explain the experimental behavior - i.e. the binding of the ligand to the complex. You can find some details of the methods here:

Influence of positioning of carbohydrate binding module on the activity of

endoglucanase CelA of Clostridium thermocellum.

Best wishes and good luck