I am trying to read trajectories to calculate some properties from a .gro file generated after "trjconv" on a .xtc file in GROMACS. This will be required in a FORTRAN90 code. But the .gro file trajectories are consuming a few Terabytes. Is there any way to read the trajectories directly from .xtc / .trr files within a FORTRAN code?
Dear Mondal
Could you be more specific? If you would like to analyze gro file you can generare a gro file using trjconv.
I suppose you want to bypass trr file creation and directly analyse the trajectory. However, that would require a lot of tweaking with the source code to record the output in fortra format. I can suggest try increasing the output control options for coordinates and velocities in the mdp file. You would obtain a coarser trajectory file consuming much less memory and that can be proceesed using fortran.
@Jamoliddin Razzokov - Actually I want to bypass using "trjconv", because it is generating the .gro files which is consuming few terabytes of space. I was looking for a way to directly read from the compressed trajectory (.xtc/.trr) directly in a F90 code.
thanks
@Chirodeep Bakli - Thanks for your answer. But increasing the number in the output control (nstxout, nstxtcout etc.) will not serve the purpose. I need to dump the trajectories every ~4-5 fs. Can you suggest any other options ?
Thanks Stephane, Actually my calculations need the trajectory of solvent molecules (to investigate dielectric relaxation). My simulations are 5ns long with ~10,000 waters and I want to write the trajectory in every 5fs. If I increase the time step of saving data I am going to lose some interesting information regarding H-bond formation or breaking.
So yes, a few TB of data is getting stored when I run trjconv on a .xtc file.
Sayantan: I wonder what "interesting information" one can obtain by observing hydrogen bond formation and cleavage in MD simulations at 5 fs timescale?
@MARTIN : should I increase the resolution? i.e. less that 5 fs? this is not the time step of the MD integrator though. I am trying to save the trajectory in every 5 fs. Can you suggest how can one look into the H-bond formation and cleavage using MD simulation?
Yes, I have a protein in the system. I have already written a code which calculates the properties and time correlation functions (located at a given distance of few angstroms). But that codes reads from the .gro files. So I wanted to know if there is a way to use .xtc or .trr directly.
To get a good average for time correlations, I think I need longer runs. Otherwise won't there be unnecessary fluctuation in the way of relaxation?
Thanks
Sayantan: If you are interested in hydrogen-bonding formation and cleavage at femtosecond timescale then you are OK. But is that timescale relevant from the perspective of the studied system? What if a hydrogen bond (somewhat arbitrarily defined by distance and angle criteria I assume) "breaks" for 5 fs? Will such an event affect the (function of the) system or is it just a noise that can/should be ignored? If the system is very complex (e.g. a protein in an explicit solvent), the energetic information is far superior over the geometrical information. Therefore, I usually store energies every 10 femtoseconds, but structures only every 1 picosecond. I view MD simulation as an ensemble generator rather than an event simulator, even when I model enzymatic reactions. I use the structures to rationalize the energies (and I often fail) rather than the other way around. If I cannot reproduce the experiments quantitatively using MD simulation (I always fail), then the MD trajectory is just a movie composed of a meaningless sequence of frames. But even such nonsense can sometimes lead to good ideas for additional calculations or experiments.
Yes, you are somewhat right but it is also important to look into these tiny changes say, in a particular region. Dynamics is indeed important (at least for a physical chemist) but I think we shouldn't overlook the structural aspects as well. Maybe to see what I want, I need to save every 1 fs. Working on it.
By saving structures at every 1 ps you won't be able to calculate correlation functions very well. The initial decay will not be accurate due to poor averaging. Say for water the dipole auto-correlation decays within ~8 ps. So you only have 8 data points in the graph before 8 ps. That's why longer runs and low timestep data writing is required in my case.
Do you calculate correlations with those 1 ps resolution trajectories ?
I have coupled my fortran code with molfile plugin of VMD. It takes care of all standard type of binary format. Now whatever file can VMD read my code can read too.
Let me know if I could be of any help
- Badges
- Science topic