If you are doing ordinary molecular dynamics then after equilibration, all properties should become apart from fluctuations, time independent. That is the equilibrium state.

I took a protein of resolution 2 angstrom but when I performed heating it shows RMSD fluctuation of around 3 -4 angstrom from native. I am not sure this is right or not. Basically I wish to run MD on modeled protein to make it go towards native using amber force field again it shows rmsd of 5-6 angstrom from original structure. I am not sure whether it is going in the right direction. while heating I kept the system restrained.

I started heating of protein from 10-300K (which is body temperature) slowly in step by step manner. Protein denaturation should not take place as the temperature of the surrounding is around 300K.

@AshutoshShandilya

It is normal for RMSD to increase on heating. If your starting model structure is different from the crystal structure by 7A RMSD and you want to do simulated annealing to see if your model reaches closer to the crystal structure, it may require very long simulations! However, you can use constraints corresponding to the crystal structure in your simulation and you may achieve the desired result. Staphane Abel's comment alludes to crystallographic packing forces that constrain the structure.

But, why do you want to do this?

For crystal structure preparation, crystallographers need to predefine certain parameters like bond lenghts and angles. Usually these parameters are not the same like the ones in force fields used. Therefore, already due to this point, your protein will develop away from the crystal structure. Plus the point that the packing inside a crystal is very dense, so in a "native" model surrounding, a protein will start to "relax". However, a RMSD of 7 Å sounds very high to me - what did you calculate them from? usually, one picks C(alpha) atoms of domains with a clear folding.

@Gopalkrishnan

I did run a simulation of around 250ns for 170 aa protein in explicit solvent (TIP4PEW water box). If RMSD increase is normal then how we can go towards native or better structure. For example I have to design a molecule for a particular modeled target I wish my structure should be energetically stable and RMSD should not exceed 3-4 angstrom. I hope I am making myself clear with the question

I will just add the details minimization was done with and without restraint, heating with small steps of 10K in 10ps till 300K and equilibrating as well. I wish to get a better structure which is close to native from a modeled structure (which may have RMSD of 7-10 angstrom from the native). So is there any way for bad structured protein to go towards the native structure using MD simulation. Or what ever result I am getting is fine.

look at the RMSF of the chain, sometimes the increase of RMSD is attributed to the N or C terminals which are highly mobile. The core of the protein should generally have a lower RMSD. another thing is the crystal strucutre temperature when it was analysed, its usually room temperature that is slightly lower than body temperature so fluctuations is inevitable.

@shahir shamsir

RMSF of C and N terminal are undoubtedly high. Research groups have folded small proteins using MD simulation meaning when I do the simulation it should go towards the native. The very reason I heated till 300k is because room temperature is around 300K. RMSF of other atoms barring some are reasonable. Basically the problem is with the topological conformation of protein which is taken during sampling of structures from the trajectories.

@Ashutosh

You should consider Stephane's and Shahir's suggestions. If you identify the regions of the structure that are more different, you can cajole them to emulate the native str.

As mentioned earlier, simulated annealing with restraints (corresponding to X'al str), followed by MD without restraints should take you closer to the native structure.

Not sure the kind of proteins you are simulating but some native structure are intrinsically disordered especially if there are no secondary structures within the core.

First, you have not mentioned about sequence identity between template protein and modeled protein. So, stability of modeled structure will depend on the similarity between sequence.

If MD starting structure is already deviated by 7 A from template structure then one can not expect that during simulation, RMSD will decrease. But, you can check the stability of the structure during the simulation. You can do PCA to check the convergence of simulation. If PCA will shows the convergence, then it means trajectory reached at the equilibrium and you can use it for further analysis. If starting structure will be incorrect, you will not be able to see convergence in 250 ns.

Increase in RMSD from crystal structure is normal, as system relax during the equilibration period. Also, crystal structure is static without any Kinetic Energy, but during MD, every atom have kinetic energy that causes fluctuations and rise in RMSD. But, If it is increasing like 7 A, then either protein is highly flexible or something is going wrong during simulations. So, you can check for both possibility.

You did not provide details about how large is the increase you observe.

Soon or later, the RMSD should reach an asymptotic value. However, be aware that the meaning of what you observe may depend on the details of your MD. In particular, are you sure that during the MD evolution there is no translation of the center of mass of your protein ? And what about possible global rotation of the protein ? Both effects would result in a spurious (trivial) increase of the RMSD with time.

I'm agree with Giorgio, because some program not corrected the mass center when you calculate RMSD. Furthermore, first you should see if the program remains the global translation of the protein. This is considering that MD was carried out well in terms of the topology and parameters.

Hi,

I am not sure if someone else has already answered this. But if i took this problem correctly, you are trying to go from a bad structure to a native one. Assuming your model is absolutely good (which ofcourse isnt possible) what you are attempting is something similar to anneling. What you need to concider are alot of things here. One of the top of my head is that if this so called bad conformation is in a local minima, it will never achieve the conformation you are looking for. 7 A of difference is huge. To prevent local minima preventing you to get to the right structure, increase the temperature far above the 300k one and do multiple equilibration runs to see if you are getting stuck in the same state. Once you do this you can be absolutely sure you are on the right path. Then, i will suggest something which i havent tried myself but its worth a shot. The velocities are assigned from a distribution during equilibration runs. You may want to control the assignment of velocities through a regression mode analysis which would be biased towards your reference. This ofcourse is theoretical but doable none this less.

On an off note, how diverged is your reference. Can the target even get there in the first place?

A bit more information would help. However, anneal your system to ensure you are not stuck in any minimas. If you are, well no matter what parameters you set, you wont get out of there without giving the system enough energy to jump the barrier.

Hope this helps.

Best,

/A

Replica exchange was mentioned earlier - good idea. In general, brute force simulations in which you take a starting structure and simply hit "go" can take a very long time to give dubious information. It sounds as though you are using RMSD (relative to some crystal structure) as a measure of the protein's folded structure - ok, good. My suggestion is to use metadynamics. Metadynamics is a method for discovering underlying free energy surfaces as a function of some order parameter(s) of interest. See Barducci et al, Computational Molecular Science, 1, 826-842 (2011) for more info.. Or look it up on Wikipedia and check out the references there. There is a plugin called PLUMED that can be easily installed with AMBER, LAMMPS, GROMACS and a number of other common MD engines that implements metadyamics.

So let's say you want to use RMSD as a measure of the protein structure - you can use metadynamics to find a free energy as a function of the RMSD (relative to your crystal structure) and see if there is indeed a free energy minimum in the region of low RMSD. My recommendation would be to perform metadynamics for two order parameters (simultaneously): the RMSD and the potential energy of the system. PLUMED makes this computation very straightforward to set up.. If you are indeed stuck in a local minima, metadynamics will ensure you escape those minima and explore the global configuration space (I don't want to describe the method here -others have done that very well in the literature so go check it out!). Who knows? Maybe the protein you are modeling differs sufficiently from the crystal structure such that this 7A RMSD region is the global minimum. Or maybe not. Brute force will never tell you reliably for a complex system like a protein. Until you do a real free energy calculation, it will be difficult for you to make any claim either way. Metadynamics, my friend. It might be just what you need. Check it out.

I am agree with Dr. Kumar. First, you have to check the reliability of your model, which is highly related to the sequence identity between the template protein and your target protein. A good guideline is the figure 5.6.13 in Dr. Sali's article published in Current Protocols in Bioinformatics. (http://www.currentprotocols.com/WileyCDA/CPUnit/refId-bi0506.html)

Second, you can evaluate your model by using the web service "Protein Structure & Model Assessment Tools" in SWISS-MODEL (http://swissmodel.expasy.org/workspace/index.php?func=tools_structureassessment1&userid=USERID&token=TOKEN). If the statistics looks good, then you can try MD to improve the structure.

A bad modelling result would be trapped in local minima and probably can not be resolved by a regular MD procedure.

The answer is obvious, thermal agitation exist, including in crystals. This is also reflected in the ‘B’ factor (that, by the way, includes de thermal factor and the slight differences in individual protein structure along the studied crystal). A crystal is not a rigid structure. If the RMSD crease continuously something is wrong in the simulation.

J. Raul Grigera