An obvious compromise could be a long simulation of the highest and lowest quality models. if the results of the simulations are similar it may imply the other 8 models would also be. Either that or run all 10 models for shorter times to see if they reach an earlier convergence.

It depends on what your final goal is. If you want to check which is the preferred conformation of the peptide perhaps you need a long-time simulation (ns). The starting point will be not quite relevant.

Thanks for your help. The simulation will run for 50 ns. I intend to study the impact of some point mutations on the internal interaction network of the peptide.

I recommend you the accelerated molecular dynamics (aMD) approach implemented in Amber 12 and NAMD. Also it is good idea to run at least two simulations per mutation and after the initial 50ns run you can continue the simulation to 100ns.

Good luck!

50 ns is a small amount of time to check conformational convergence of such a long peptide. Your starting point will be relevant. At least double it up, and consider using REMD or metadynamics to get better sample.

I will advise you to use peptide structure prediction software (like pepstr, pepfold) to get right model for structure. Try simulation on predicted structures.

Gajendra: He obtained homology models, I doubt using ab initio structure prediction softwares makes sense.

I advised as homology based protein modelling software mainly fails on small proteins or peptides.

This is interesting. Doesn't it just depend on the sequence similarity?

About homology modeling in case of peptide It might depend more on your template. Was it a peptide or a protein? Also are you studying the conformation of your peptide in solution or bound. As we know bound conformation of peptides can be strongly induced as seen for GLP-1 type of peptide (Nterm becoming helical upon binding).

In general I will agree with Massimo approach and use REMD is you have enough computing power. If your peptide is strongly constrained you may want to check this paper: Chem Bio Drug Des. 2010, 75 , pp 348-359 by Guo Z et al.

REMD will be a good approach to consider. You could generate an average structure of the 10 homology models. Run your simulation with the average model and analyze its convergence.

Hi,

Here is how i understand it. You have a 36 odd length amino acid sequence for which you generated a structure. First question is why are you limited to 10 models, from how i have used homology modelling you can go on to generate more, cant you? Another question, i am not really sure what you mean by metrics, but what exactly is the protocol you adapted for these model generations. At this stage 0.6 Ang backbone RMSD is not very informative, did you employ a minimization technique during model generation. The only way the highest score is reliable is if you have used some for of minimization. If you have, than you should just start with the best model in your dataset. Another question is what sort of secondary structure elements are present in the model generated? Are there any loops? Loops are tricky to handle and with these you should really not try to look at the RMSD.

Since in one of your later comments you say that you will be studying effect of mutations, its best to start with the best model generated on your dataset (if minimization was involved). However note that, if you did employ minimization and the template structure that you used was close to its minimum than it can be assumed that your predicted structure assuming a very high homology will be close to its minimum as well (all 10 models). For mutational study purposes (considering the interaciton network that you talked about earlier) it shouldn't matter if you pick up any of the 10 models. However, before the MD is actually conducted you should consider minimizing your structure. If you system is stable you will notice that all 10 models will converge to the same energy value (almost), hence as i said earlier it wont make a difference which ever model you start of with.

Hope this helps.

Best,

/A

Given the relatively low backbone RMSD of your models, I would start from the highest quality score one. However, I would expect starting configuration is not negligible and can severely affect your simulation results. This is particularly relevant if you run a relatively short MD like you are planning to do. I strongly suggest you to think about REMD. Parallel tempering should be suitable in your case, in order to sample several conformations and quantify which is the most energetically favorable.

0.6 Ångström is a very low RMSD for a 34 peptide backbone... given the low number of atoms of your system, a very long MD simulation (REMD is a good suggestion) could be performed quite easily... starting from the best conformation the system will likely evolve through conformations with an RMSD larger than 0.6Å... probably all the conformations you have got will be explored.

One question: how did you prepared your homology model? modeller?

I agree with the idea of REMD. However, the amount of computing resources and time required might be a problem. So I suggest you use Accelerated MD, a method developed in McCammon's group (Accelerated Molecular Dynamics: A Promising and Efficient Simulation Method for Biomolecules, D.Hamelberg, J.Mongan, and J.A. McCammon. J. Chem. Phys., 120:11919-11929, 2004.). This method increases sampling efficiency by modifying potential energy surface. Take a look at the paper and there are a couple softwares that have implemented this method, such as NAMD...

Are you sure that your molecule is rigid? If yes, than what are the natural environment of this protein in cell? Using long-term MD and making its environment mostly natural will help you to obtain the structure that looks like real.

I think that more or less most people here have raise several issues you should consider. First of all, what about your polypeptide. Think carefully first what exactly you know about it, is it a random sequence you have designed or a biologically relevant polypeptide. Do you have any kind of structural information about it. You mention for exactly that it is "rigid", which implies absence of conformational flexibility. Is it something you know or infer from homology models.

The second issue is what exactly you seek for with your simulations. If your polypeptide is more of a peptide and you want to examine its conformational space, you should think about the force field to be used-the new CHARMM36 would be a good choice-and the exact method of MD to be used, There are several enhanced sampling techniques, the easier and standard one is Replica Exchange. The accelerated MD is another choice but it's just one of several existed and make sure before using one that you understand how the technique works, advantages and limitations

Agree with Kostas. REMD is kind of the safe choice for conformational sampling. If you're adventurous and you like cool stuff, I love metadynamics approaches (which you can combine with REMD) but be sure to understand it well before (I'm not sure I do!)

Since there are only 36 amino acids, and you intend to sample for a long trajectory, I don't think that it is going to make a big difference if you choose any of the modelled structure. Here I am assuming that you have only primary structure of the polypeptide. No helices or beta sheets are present.