Hi,

I will be less enthusiastic than my colleagues... If you plan to do drug-design, as it is mentioned in your post, you certainly want to perform molecular docking calculations. This will require a very good knowledge of the side-chains position of your targeted protein that defined the binding pocket. I hope that you have an idea of the binding pocket on the protein because the searching of drugable cavities is a complex and challenging process (maybe an art).

MD can help you greatly, but you still need a good starting point since it helps for minor loops rearrangements. Keep also in mind that sometimes long MD simulations (from 50ns to 1µs) is required to explore some loops reorganizations.

My advice is: try but stay in touch with experimental people who can provide you some crucial informations (biochemistry, inhibition...).

Regards

I dont think there is any problem if you prepare the model using human sequences. It all depends how much similarity you get with the human sequences, if you get significant hits then you can wisely use it for modelling and then future docking. Performing MD will defnitely improve the quality of the structure.

yes true Dipon Das......

if the sequence identity is very low is more difficult to improve the model using the MD simulation. If the similarity is under the 30% the model isn't a useful tool for your purpose; but if you want you can try..........there aren't problem if in order to model your protein the software (or web-server) use the human sequence, the important is the family of proteins (they should have the same function) and the high degree of sequence similarity. All the rest is nothing!!!!!!!! greetingsssssssss

Thank you all for your valuable suggestions and quick reply.

Hi,

I will be less enthusiastic than my colleagues... If you plan to do drug-design, as it is mentioned in your post, you certainly want to perform molecular docking calculations. This will require a very good knowledge of the side-chains position of your targeted protein that defined the binding pocket. I hope that you have an idea of the binding pocket on the protein because the searching of drugable cavities is a complex and challenging process (maybe an art).

MD can help you greatly, but you still need a good starting point since it helps for minor loops rearrangements. Keep also in mind that sometimes long MD simulations (from 50ns to 1µs) is required to explore some loops reorganizations.

My advice is: try but stay in touch with experimental people who can provide you some crucial informations (biochemistry, inhibition...).

Regards

@Florent Barbault thnx a lot for ur advice. i'll keep it in mind.

Hi,

if you have already some active compounds, I would do a benchmark and look whether the known actives score well compared to some random decoys. If yes, then I would give it a try and make a virtual screen. If they do not score well, you can switch to ligand-based screening and go for similar ligands first. I agree with Florent, MD simulations will not necessarily end up with an improved structure, especially when your starting structure is to far away from the target structure. If you have a homology model, you may use WHATIF for a check of consistency.

good luck

Sören

One more question also comes in mind many times and i cnt find the answer yet...regarding the identification of active site/ binding sites.....

I kno many servers/softwares to detect the same but i dont have good idea which pocket should be chosen among the many as all servers/softwares suggest atleast 3-10 or more binding pockets in any protein. What should be the criteria to select the best pocket..additionally, the main problem is..if i dnt have any literature proof/ experimental findings?????

HI MR.Nutan Chauhan ,

MODELING any new structure its depend upon the similarity score of ur template. In complicated case just try to use more tool for template search, then there will be chance for u to get some idea. research is full of creativity only. If you doing modeling for particular drug discovery then you should be very careful. computational biology way of modeling protein is never be an that much accuracy, " better we can say as may be or may not be "

once you finalized your structure, read some article about ur protein or closely related protein family about the function of every residues such pi, hydrophobic, hydrophilic region & folding nature of those residues & for what type of drug u plan to design, its mode of action, whether it will directly bind with the residues or it will bind with other receptor region to force the activity indirectly. these things will be very important in he case of docking study for drug discovery. after that just try to find the pocket region or active site region of ur protein & need to understand the principles and intermolecular function with the close residues present in ur protein active sites.

if you try to model with very low score template then your structure will not be perfect you can notice some times one end of the structure will be in long tail like structure without make any loop with other residues. these are common error while using poor template. pyres is the very good online server for template search. if you not build good structure then you cant able to done docking using some impact software's. because most docking software's will be very sensitives in the case receptor protein preparation step. but in the case of Discovery studio these steps are very user friendly. anyway i am not that much genius in computational biology. better you try to read some review papers about problem in modeling protein structure.

all the best for your research work Mr.Nutan Chauhan .

1. As the source of my protein is a protozoa and my main goal is to find better lead compounds, can I use a human protein sequence as template to model my protein (i-tasser exclusively used human sequences)? Will there be any problem if I use this modeled protein (generated from human protein structure) for further docking purposes? If yes, why?

Answer: Not sure. I would need to know some physical variables (what condition it's in; is it easily dispensible, etc.). So I'm presuming you're using docking to predict the best model to represent the protein structure, right? Have you also considered homology modeling?

2. Can the quality of this model be improved by MD simulation? To what extent?

Answer:

The advantage of MD simulation is that you'll have conformational sampling. From a stat-mech perspective, you'll have an ensemble of structures at your disposal. You'll also be able to understanding residue-specific binding attractions/repulsions. I myself found benefits to combining docking (protein-protein) with molecular dynamics simulations to get a better overall understanding of protein-protein interactions.javascript:void(0);

However, don't be careless when you're setting up the simulation. Lots of people in the field of MD tend to do sketchy things like deleting residues to speed up simulation without a rigorous justification, or run it at a certain pH/temperature without justifying it. (Ask yourself if you want an NPT, NVT-ensemble and why based on the system). If you're going this route, it's best to have an experimental biologist with a solid foundation in physical chemistry as your wing-man.

Hope this helps!

If you dont know the active site, you can try to dock the known high affinity ligands to grid generated from all protein and see, where are poses with best score.

@Sakthi Vel, @Jonathan Reyles thnx for ur very valuable suggestions. I will study more about MD simulation and the parameters.

@Fredj BEN BDIRA I have tried Robetta server too...it gave error in predicting the structure of my protein.

@Vladimir Garaj, thnx for ur advice.

I kno I can do blind docking but I wonder how much good are results from blind docking... If I do blind docking by creating grid over whole protein then I know the s/w will try to dock the ligand all over the protein and will suggest me the best binding pose with less energy. But if I try this procedure in many softwares then definitely the results will vary and will give diffeent binding poses as the search algorithm are different in different s/w. then i this case what result should i consider the best?

And if I want to validate my docking results experimentally then there is less chances that the ligand will bind to the suggested binding site residues (suggested by software through docking).

so lastly, Please suggest me how reliable blind docking is?

I use GLIDE software for all the docking experiments.

Can anyone suggest me how reliable Arguslab software is in context to docking? I tried this s/w and found very much variations in docking results with the change of the size of grid.....

Hi

It seems that your model is not very accurate. Did you tried with Modeller program. I think you can imporve the model using errat plot technique. Finally you have to minimize the model also before doing docking. best of luck

@Gopi Mohan, dear thnx for ur reply and suggestion...but i cnt use Modeller as i have no good quality template for homology modeling of my protein.

Nutan Chauhan,

Docking protocol will be wary from each software's depend upon the preparation of the molecule & method of force field applied. There is no single docking algorithm or scoring function that can correctly predict the binding affinities of ligand in molecular interaction. For this reason it should me necessary to make conclusion of ur drug-receptor interaction with two-three docking program. The simple way of selecting the best interaction by binding score & energy. In this case some program will predict good binding pose result value in Negative score & some program predict in positive score.

After finishing docking & if you notice that ur drug molecule make some hydrogen bond interaction with the receptor protein then you move on to pharmacophore identification for your result. pharmacophore step will provide the clear conclusion for your different docking result. but the main thing is you should make clear idea about in which atom of ur drug and amino acid residues shown Hbond interaction. Example H33 (Lignad atom) form hydrogen bond His23 (protein) then go to literature survey about the His23 residues function & mode of action with the closely related protein family. after that you can get clear conclusion to choose best docking pose for your result.

It is very important to notice the inter-molecular interaction of bond formation ur docked residues with other closely related residues in the active site region. For example, in ur protein active site region have 30 residues. after docking u notice that H33 (lignad atom ) form Hbond interaction with His in at 33. then you need to identified how other closely related residues make inter-molecular interaction. these steps is necessary for make a proper conclusion.

My suggestion you can use Glide & Discovery studio. these two docking programme are gud for docking study then Arguslab software. you can also try with Autodock ( freely available & effective) but Autodock docking prediction will be same as Glide which you used.

Anyway i am not that much genius in this field, so if anyone find mistake in my suggestion please inform me. I am very interest to learn from you peoples.

Hello Nutan,

After going through ur question regarding to....

Protein Structure Modeling -- as u have less relevant template or u r going with human protein having similarity with ur target one.

Use of Molecular dynamics in your condition and finally molecular docking.....???/

first Thanx to all dear freinds to give such a valuable suggestion to Nutan....

1) Nutan if template structure is not upto mark then u can go with human protein that u have stated.

2) either as if you are not getting appropriate template structure you can go for fragment based modeling also if you have time u can try in once.

3) where as MD will obviously help you to understand the molecular behavior of ur protein.....

4) where as u can go for certain analysis such as Protscale aon Expasy website - in tool section, where u can check the buried residues in ur protein and even the hydrophobic regions in an graphical display that will give u a predicted results...

5) Regarding to active site u can use online server such as Qsite finder where u can get volume based active site ranging from 1-10 with specific atom name and number to a specific amino acid residues of ur protein.

6) as u have certain ligands u can go for virtual screening --- use igem dock here u can specifically interact ur active site with respect to ur ligands with less computation time and resource .........

if u have any further doubt feel free to ask.

Nutan, Arguslab is not very reliable. If you have Schrodinger Suit, you can use Sitemap to identify binding site (I have not used it yet). If you purchased only Glide, do the docking with whole protein and choose binding site with most poses. As Sakthi wrote, Autodock is also good for docking and Autodock Vina is even better due to my experience.

You can perform threading ..then use Schrodinger Glide suit for docking.

My suggestion is that you cannot rely on one software (i-tassar) for such poor model. if your model is not correct then whatever protocol you use for docking etc will not finally work out in search for some lead compounds. So I suggest you to try with different protein modelling software as well to build different homology models and check which model is most accurate. As a bioinformatician you should try also PSI-BLAST to improve your template selection.

I was going to recommend Rosetta from David Baker's lab, but I see Fredj BEN BDIRA beat me to it. :-)

I have not found yes a truly reliable tool to predict the structure of a protein lacking good templates. But there are a few tools that you should try.

1. Have a look at this paper (open access):

http://www.cicy.mx/Sitios/Journal/pdf/2011/October/HighDefinition/IBCJ_2011_1_1_26S.pdf

which I published in a not widespread journal (one that just started) to support the journal itself, but I feel is a valuable paper.

2. If you look at the paper you will find that a good source for the prediction of proteins with no known solved homologue is a somehow old server, SAM-T08. Unfortunately, SAM-T08 assembles your model from small pieces and combines a lot of information leading to locally defective models. I.e. the models may have unusually long bonds or unacceptable atom clashes. But those defects can be corrected with MM using flexible software. In my experience Hyperchem 7.5 is one the most reliable for the job, but that is commercial. Modeller can also be used, because it not only models your protein, but it can minimize an external PDB (look at details in the paper mentioned above).

3. An alternative is the use of Quark, a completely ab initio prediction guided by distant templates.

4. Now the most essential part is to analyze the quality and appropriateness of your model. To score the quality you may use a number of servers. Finally, to score the appropriateness you will need Rosetta-design and HMMer, please look at:

http://dx.doi.org/10.1371/journal.pone.0012483

This last piece of information is unique, because it tells you if the 3D coordinates of your model's backbone are an acceptable host for the aminoacid sequence you intend to model.

Good luck...

Try using HHPred , it might give you good template.

You also need to check how much sequence identity(similarity) is there between both template and target sequences. Also very important to know which are all the active site residues in target for a drug like molecule to bind. Also whether all these active site residues are conserved in both template and target as well. Side chains of these active site residues will contirbute immensely while docking. Since there is no optimal algorithm yest discovered involving dynamic docking between target and ligand. For that MD will be immense help and try to do procheck analysis after MD. Also another tip is that if you have disulphide bridges in your protein then check all those residues are conserved in both template and target.

Regarding the quality of the model, MD could assist with various issues that might be present in a model. With experimental structures this reduces such issues as steric clashes, optimises side-chain orientations, relaxes over-restrained models etc. As Florent mentions, loop reorientation can take some time depending on the size of the loop. However, if you have a poor model you may find that the structure becomes worse, rather than better, so be prepared for that. You mention the % of residues in allowed and disallowed regions of the ramachandran plot; is this the only way in which you are assessing the quality of the structure?

I'll be honest, I have never used i-tasser, so I can't comment on the structure it has given you. However, one thing I was wondering was if it had given you a structure with a well populated fold?

Answer to the first question is you have to see that what region of this proteins sequence is conserved that will give you some idea which are functionally important amino acid but designing a drug based on human sequence would not be advisable.

Extended MD simulation can improve your structure to a greater extent. How big is your protein depending you should run around 100ns which would definately do some favour to your structure allowing its side chain to move freely. Force field are not that consistent and MD simulation are random ensembles which is not easily reproducible so its difficult to comment to what extent. Good ness of the structure also depends on your initial coordinates.

Hope this helps

Theres a competition for protein folding called CASP you can google that if you are not aware. They say which are the best servers you can model your structures with some of them and compare the result and do the MD simulation on one or two of them depending on the availability of the computing facility. You can also try Bhageerath_H for modelling of your protein you can find that on www.scfbio-iitd.res.in

I would suggest to do the BLASTP searches with different "Expect" values, i.e. other than the default Expect value, which is 10. You can run BLAST with Expect value 50 and analyse the result.

no restriction on species in protein structure prediction, you can use any structure for the modeling template. In the case of low homology sequence the most important is the structure based alignment. I tasser is among the best web based structure prediction. or you can try threading method.

I am not recommend for MD unless only for side chain minimization.

Thanks alot to all for their valuable reply.

@Hasni Arsad I was also thinking the same regarding species. But I did get to know that one of the researcher has got his degree canceled just because he used a Human protein as template for structure prediction of his protein of interest (bacterial protein). He was also working in drug designing. That's why I wanted to know the reason for this fact.

If any suggestion you have please do share. Will be great help.

@Florent Barbault and Ashutosh Shandilya: I appreciate your answers regarding MD. But I have a doubt regarding the selection of the time steps/time for running an MD. Could you please suggest me some logic on what basis the time should be chosen. I know it all depends upon the size of a protein. But I don't have idea about what should be the criteria in choosing it. Is there any sort of range for this?

Hi Nutan,

The idea is the same secondary structure will give the same tertiary structure. meaning that, whatever species you use as template the structure will be the same. But you need to consider the location of expression, protein group and the protein complex.

- First of all, check the literature. Has somebody done the same as you, on some other system? How?

- Even if there's no protein with similar sequence, is your protein part of a family with known conserved structures (like, e.g. GPCRs)? Then you can try homology modeling. Otherwise I wouldn't rely much on the results of ab initio models, even if they're getting better and better by the minute.

- If you have confidence you have structurally similar proteins, use HHPred to get better structural alignments on the potential templates. Try to contact some expert on alignment and homology modeling to know how to do that (I am not one)

- Once you have several good alignment for several templates, you can use MODELLER and do homology modeling. Do *a lot* of models and check what are the best ones, using several score functions (normalized DOPE for one) and some model quality online tools like PROCHECK or VADAR. Look the score *profile* to see if parts of the molecule you're interested are modeled correctly. Try using multitemplate modeling.

- MD will help refining/relaxing your model, but only to some extent -it's not a panacea and it could even make things worse in some cases. And it takes time. Use a good forcefield (e.g. AMBER) and do a *long* run. Try also using REMD.

For docking, I would advise you to get all the experimental information you can, and then try HADDOCK to get information-driven docking.