Is there any logic behind deciding the time and time steps to run an MD simulation? Is this related to the size of the molecule?
hey, thnx for replying and sorry for not writing the full form...its Molecular dynamics
You have to consider your computational resources, the refinement of your structure, the complexity of the structure, the number of naturally disordered regions (since they are more mobile than well-defined structures like a-h or b-s). In all cases the higher the better, why don't you try a decoy run, using just a small number of MCMC steps (or FTs) and see how your indices vary between each step, and how many hours (days) does your cluster take for such run, then you can extrapolate to a higher number of steps expecting that at certain point you would have many conformations covered and no more steps are required.
The total time you are running your MD simulation depends on the time frame of the phenomenon you are studying. If you are looking at an unfolding process you should know how long the process takes. That is, is it a 4 ns process or 50 ns process.
For the time steps; if you are running a long simulation, many ns, then your time steps will be longer so that the your mdcrd file is small enough (< 1G) to handle.
Many thanx to Victor and Gloria, I'll surely try as you've suggested.
the timestep depends on your force field. Basically if it is too long the atoms move too much between two integrations and the simulation is therefore unstable. If you don't know more use 1fs.
PS: running MD seems easy but the results can be meaningless if not well implemented. Be careful...
As a general principle, the simulation length should be as close to the biological event you are trying to model. But as a rule of thumb, longer simulations (in the microsecond regime, for protein folding for instance) are much better than shorter MD runs. Even then you may still need a way of confirming that your system has converged to experimentally relevant ensemble. You might also want to consider advanced sampling methods such as replica exchange MD (REMD), and coarse grained forcefields which will allow you to sample the phase space better.
With respect to time-step, most MD set-up will employ between 1 and 2fs. If you are using a coarse grained potential or virtual sites then you can use longer time-step without your system becoming fatally unstable. I have been able to use 5fs and 7fs with virtual sites.
If you work in materials or composite you can also use the peridynamic theory
your time step will be dictated by how long the shortest event in your simulation takes to occur e.g. in protein structure MD simulations, this will probably be the C-H bond vibrations (though the SHAKE algorithm allows you to use a longer timestep than usual here)
The duration of your simulation is determined by how long it takes in nature for the phenomenon you wish to observe to occur - although as mentioned above, generally the longer the simulation, the better...
Time selection and time step selection depends upon what structural interactions we want to probe e.g. H-bonding, confrontational change, binding interaction etc. each of which its own time scale of existence. It is not much depending on the system size (box size) but depends on size of the molecule definitely if going to investigation conformation changes of large proteins or other biomacromolecules (time scale for such process may vary from less than a nanosecond to few seconds)
For the timestep: take the largest frequency among all vibrations in your system and divide the corresponding 'period' ('inverse of the frequency') by 10. In this way you will have your fastest motion sampled at least ten times during each vibration.
For the simulation time: calculate the mean square displacement of some of your simulations and take the square root. If the corresponding distance is a suficiently long distance in terms of travelled distance by the molecule in question then it is ok. If not increase the simulation time accordingly. This is valid for molecules diffusing in some media. For conformational analysis you may better consider somehow if a significant fraction of all the conformations are being sampled: quite difficult perhaps, but I can not give you a more accurate answer if you do not describe your problem in more detail.
The answer could be found in "Determination of Proper Time Step for Molecular Dynamics Simulation" by Jong-In Choe and Byungchul Kim - Bull. Korean Chem. Soc.
2000 , Vol. 21, No. 4
Thank you Nutan for asking the question, thank you Hisham for the paper and all who gave answers. I have learnt useful things here..
- Badges
- Science topic
- Similar topics
- Biological Science
- Biochemistry
- Enzymology
- Enzymes