my approach here is to do a clustering analysis to find groups of similar frames, and then report the frame nearest to the centre of the largest group, cpptraj is a utility that will do clustering for you

Apparently, for it's a MD simulation, the 'final' structure is just one of the possible structures but not the 'best' one. 

The potential energy is helpful to find the best configuration but MD cannot ensure that you get the most stable geometry.  However, at least it is better than the final structure.

I have no idea about the system you are research, so I do not know if the simulation is long enough.  But basically, you should wait until the structure wandering around a position, and take the equilibrium structures.  After that, an averaging should be performed over them.  But perhaps you do not need such accurate result.

I do not know which package you are using.  Maybe the ideal answer has been calculated by it.

The entire point of making a MD simulation is to show the dynamic properties of the system, so it's difficult to choose one representative structure out of it. I'd rather draw a number of important features of the system in question on a plot (end-end distances, ramachandran plot and so on) and not rely on a single frame.

But if you really need it, I would do as Yu suggested: make sure that the simualtion really converged and there are little structural/conformation changes. Then take either an average of selected, last frames or take the frame with the lowest energy (you can also minimize it) or take the frame with the lowest RMSD to all other frames.

If you're looking for a good protcol,

Book: Homology Modeling, Volume 857 of the series Methods in Molecular Biology

pp 137-173Date: 11 January 2012

A Practical Introduction to Molecular Dynamics Simulations: Applications to Homology Modeling

Alessandra Nurisso, Antoine Daina, Ross C. Walker

http://link.springer.com/protocol/10.1007/978-1-61779-588-6_6

The article by Nurisso et al. details a protocol in AMBER for what you are trying to do.

Check your RMSD per frame to initial frame, and if it is relatively stable and shows convergence then take the final 20% (200 frames) and do an average structure of that. Once you obtain an average structure you will need to minimize the structure to remove any hot spots (of which there will probably be many). Hot spots (steric clashes) will be bond lengths which are too short or too long, or un-realistic angles, dihedrals, etc.

my approach here is to do a clustering analysis to find groups of similar frames, and then report the frame nearest to the centre of the largest group, cpptraj is a utility that will do clustering for you

I agree with the last comment. Clustering of all conformations excluding the beginning of the MD trajectory and then analysis of the most representative clusters representing at least of 50-60 % of the full conformational ensemble. Make sure that these clusters are visited several times during the simulations for convergence issue.

Oups. I did not see you have 1 ns MD trajectory. What is the amino acid length of your peptide and the goal of your study?

I agree with the last comments. Clustering analysis should be very useful for your aim. In case of disordered protein be careful on how you define the "distance" between the structures. In several cases RMDS is Ok, however if the structure is disordered other measure such as GDT-TS could be more appropriate.

Oups indeed Philippe is right, microseconds would be more normal.  Try using CUDA or similar.

Clustering and rmsd/energy fluctuation analysis of selected structures during MD.

Thanks to all of you for the suggestions. It helped me to understand different approaches for MD analysis as well as selecting representative structure from MD analysis.

@Andrew-> I was more interested in ab initio folding of peptides and therefore can't take RMSD from initial frame as reference.

@Philippe-> I am working on different peptides and their upper limit of length is 50 amino acids

@Joshua -> Should I go for microsecond MD because my system consists of peptides and not proteins and can I perform a short MD in nanoseconds?

If your goal is to determine the most probable structure of your peptide, use PEP-FOLD2 server, except if it is more than 50 amino acids. In that case, use QUARKS or I-tasser

Use following script if you are working with Desmond.

$SCHRODINGER/run trajectory_average_structure.py  -b  desmond_md_job-out.cms "desmond_md_job_trj[400:600]"

Where 400 and 600 are frames corresponding to 8ns and 10 ns respectively (you would like to change this as per your system)

Mohd Athar Is it the script use for determining the frames to analyse?

Philippe Derreumaux Do you mean the clustering is done starting from the frame that reach equilibrium till the end of simulation?

Thanks in advanced.

Muhd Hanis Md Idris , this script is for computing the average structure out of the given frames in desmond.

Moreover, average structure may not always represents real structure, therefore I will recommend clustering based on rmsd.

thanks

the most representation structure (in 1 ns), well that means, you are looking for the structure that has the minimum free energy. For sure there are some server which compute the free energy energy and contributions for a given snapshot (or frame) such as foldX (http://foldx.embl.de/). You can try. The optimal way will be to project the Free Energy of your trajectory for a given set of Collective Variables. and Then from the profile identify the frame which has the min F.