I would advise a different route to predicting binding partners of disordered proteins. Molecular modelling is almost certain not to give you useful data-mining for what your protein's intrinsically disordered/unstructured regions will bind to. Your best bet is to combine homology and co-expression analyses, also look at PPI data. Databases like STRING and D2P2 might be useful:

http://d2p2.pro/

http://string-db.org/

You could use Discovery Studio or Insight Ii but are not free.

Modeller is free though.

There is also Raptor but don't know if it's free or not.

Anyhow, this will take you some time and any of these can't give you an automatic 3D model.

I have a time but not tools. I had modeller on my classes but I will have to focus on it. I need to confirm my experiment and propose a model of interaction between two proteins. I will try with this programs, thank you.

You can use freely downloadable softwares i-TASSER, or modeller, and some automated tools like 3dzigsaw, and swiss model, once the modeling of your proteins is completed, check their binding using Vakserlab (online tool for protein protein docking) freely available and very easy to use...

The problem with the packages like I-Tasser (also Phyre2 can do something similar) is that the confidence, that the model is really good, is low. In order to make a really good model, to have knowledge about some hot spots in the structure, you need to use packages as Discovery Studio or Modeller.

I must say, using I-Tasser for instance is an easy way and could use it further in perfecting the model. It is interesting though what validation score will get compared with other softwares.

It is not trivial task. If I understood correctly you should perform the following steps:

1) Homology modeling to create the structure of the first protein. (Modeller is ok, but Schrödinger is perfect). If you don't have a good template.. this is not good..

2) You need do to know the binding epitopes of both partners - perfectly by some biochemical analysis. If you don't have such information then you have to use some good prediction tools for that such as ODA Molsoft (it a free web server) or again Schrödinger package. Otherwise you are going to the black box. It is a good idea also to perform some electrostatic calculations by APBS (free).

3) Perform protein-protein docking by some good tools as such for example HADDOCK (free), ICM, Schrödinger or some of the mentioned above.

4) Run MD study because personally I can't trust only on the docking score, especially in this case (amber 12 GPU version is a perfect choice or NAMD). After obtaining this "final structure" you can/have to run MM-PBSA calculations to obtain some free energy of binding confirming your docking results, indeed withhold the entropy contribution...

I hope that after few days can give you an example paper, but first have to resubmit the corrected version to the journal... :(

Thank you very much for an information about different packages and programs. Unfortunately my protein has only partial structure solved (DNA binding site) and second one has only pdb model of 1 domain. Protein consist mainly of unstructured regions so I will have to predict them. I will try with homology modelling but now I'm just preparing the experiment and gathering information about possible paths and approaches. When I will start the experiment I will have much more questions. Thanks again.

If you don't have good templates available for homology modeling, the current state of the art for de novo structure prediction is the Rosetta program from David Baker's lab.

I agree, it will be difficult to model an unstructured protein because there is no templade to use. Moreover, these proteins can adopt a "structuerd" conformation in the presence of the "protein partner". Actually I agree with David, you could try the Rosetta server ...if you are lucky... :-)

I would advise a different route to predicting binding partners of disordered proteins. Molecular modelling is almost certain not to give you useful data-mining for what your protein's intrinsically disordered/unstructured regions will bind to. Your best bet is to combine homology and co-expression analyses, also look at PPI data. Databases like STRING and D2P2 might be useful:

http://d2p2.pro/

http://string-db.org/

first ask yourself the following questions:

- Do I need all the sequence of the protein to model?

by this you can either select all the sequence or just a restrained domain which you want to study

- is there any homologous structure found in the PDB database (you may answer this question by using Blast or FASTA)?

use Blast or FASTA servers to search over all the PDB in order to find an homologous for your protein (or your domain)

- can I get a highly quality alignment with the selected homologous structure?

if you can do this, than you can get a high quality model of your protein, if not try to use THREADER and find if there is any match with any of the known folding.

I recommend using EXPRESSO to align the target with your template.

Try to see use different alternatives (different templates, different alignment software...etc). load tour model in your preferred viewer, if you get a satisfying structure its OK, if not,

well unless you have accessibility to a great computational power to try an ab initio prediction using Rosetta, you can't really do mach thing.

I hope this will help you

Based on your information I believe that Julian is right.

I don't think that using any "magic" software will solve your problem. Moreover, your main aim is "to check if my protein could interact with other proteins." thus I can't imagine how you will achieve that starting with 2 structures with large RMSD difference to the native ones (I don't know how large are yours). Next, as it was mentioned above " Moreover, these proteins can adopt a "structuerd" conformation in the presence of the "protein partner". That's why I suggested some MD method to be applied for a refinement, but received a negative vote -probably it was stupid suggestion :). Also what about the identification of the binding epitopes? Ops..I see, Rossetta can predict this directly from docking score:) Is the Rossetta is really so good?

I thought that there is already good tool for problem like mine. The other approach is modification of amino acids and express different mutants of proteins. But honestly speaking it is too much work (because I am already working on a parts of proteins) and the interaction seems be very specific. I hoped that some basic MD could just point the region which I should focus on. I suppose I will have to use different tools than MD because one of the proteins doesn't have any model either homolog model.

I think, unless you will use robust and complicated simulation including molecular modeling, docking and thermodynamic simulations, you are only going to get a dirty solution to your question. I wouldn't recommend using MD, because this kind of simulation is computationally demanding and require long trajectories simulations after which you will not be sure whether you reach the naive state or not.

Try at least to obtain a model for your protein partners as much accurate as you can, preferentially with comparative modelling. than try to use protein-protein docking (Cluspro server is good in doing that http://cluspro.bu.edu/login.php ).

Then by looking to the docking prediction, you can target only few residues to mutate rather than generating a lots of mutants. (Don't forget to check the structure integrity either by CD or NMR).

Crude modelling approach without basic informations such as SANS data, HD(X)MS-data, biochemical data of an unknown protein, it wouuld be difficult to desgin the 3d structure of protein through I-tasser or modweb , rosetta.

David,

I hesitated a bit to try to address your question. I hesitated because there is a certain level of contradiction in your stated goal. The protein cannot be ordered and disordered at the same time. So your question is addressing conditionally folded or unfolded proteins. We do not fully understand the protein folding problem in general and even more so conditional folding. So, the general answer must follow Julian's advice. I personally do not believe that Rossetta or any other program would even remotely address your problem. Using energy minimization principles to derive protein structures is ill conceived and applicable only to the "most stable structures" as David Baker asserted himself in one of his commentaries to Science. Heuristic elements included in Rossetta and Itasser are not helpful in this case because of lack of examples (and this is certainly true for unstable sequences).

However, I decided to comment because it may help you a bit. A modern view of proteins clearly indicates that they share characteristics of a solid and liquid phases of matter. So in the protein structure there are solid-like and liquid-like elements with variable contributions (Proc. Natl. Acad. Sci. USA 106(2009)10505). If you can, by any of the computational tools, get a description of some of them that would help you to define your protein, and at least to know which parts of your protein are not predictable in terms of final structure. So maybe you will be able to identify hinges versus completely disordered elements. (I wrote a short commentary on the subject: FEBS Lett 586(2012)1675)

The second aspect is that the experimental databases are much better source of protein protein interactions than crude modeling tools. So focus on better defining your problem versus searching for modeling tools that seems like a shortcut but in reality it is a misleading path.

Dear Boguslaw,

thank you very much for your reply. I suppose that this unstructured protein has to have a very specific structure in protein-protein interaction, because I observe a huge wavelength shift during fluorescence experiments. I have to perform few more experiments to confirm that. My idea of modelling an unstructured protein was to produce as many random models as I can and then try to dock random models with second protein which structure I know. The criteria would be an environment of tryptophan residues (more hydrophobic). I would not like to determine the exact structure of the protein but just predict how could look like protein-protein interaction which explains phenomenon I observe.

Do you know if your protein is helical when it is bound? If so, then if you model several random structures with the helix in the right location, it may be possible for you to try to dock these structures. A program such as TraDES could allow you to do this. However, in my experience trying to come up with a physically relevant structure in by sampling random structures in the case of a protein that is not helical is not likely to be fruitful.

Dear Dawid,

I decided to write you again because I have some updates and still believe that your idea is good and you can achieve valuable results by combination of several computational approaches. I don't believe that this possible by a single method or software.

As I wrote you above I used combination of approaches to solve such not trivial problem. In my case I didn't have any information where is the binding epitope of my second protein (P2) and even whether the binding epitope of the first one (P1) is properly folded at my conditions and is able to interacts with P2. However using the above step by step scheme I was able at least to identify the main structural elements involved in the protein-protein interactions and suggest a binding mode despite the limited experimental information. Surprisingly several weeks later my results were confirmed by detailed experimental study published in Cell and the same structural elements were identified. Sounds like a good story but makes me to believe that this is not impossible mission.

All the best,

Let me just clarify why many people (with whom I agree) discourage you to use any kind of Molecular Dynamics (MD). What I say here is obvious to them, but maybe not to everybody.

Although in principle one could use MD, the time-scale of the events you want to simulate is far too large. So the problem is not (mainly) in the accuracy of the method but in the sampling it would require. Sampling hundreds of nanosecond for a large structure is already challenging, but if you start just from the sequence, in that time-scale nothing interesting or useful happens.

Hi Ramon,

I thing that phrase "to use any kind" is a little bit too strong. I agree that the sampling point is critical here, but it has been shown that methods such as accelerated MD can give access to millisecond time scale in a direct comparison to conventional MD. The aMD is a great for sampling. Moreover if he focus only on the binding site and harmonically restrict remaining parts he can use a lot of other MD approaches to achieve a good sampling.

Indeed the above is valid if he has at least some reasonable template for initial homology modeling step. I don't say that he can obtain the folded/native state starting from totally unfolded one by MD:)

You can try modeller.