Necessary files to be kept for WHAM analysis?

User discussions
1

GROMACS version: 2025.3
GROMACS modification: No

Performing the umbrella sampling simulation results in obtaining huge amounts of data (tens of GB).

What are the files that should be necessarily kept to allow for WHAM analysis, and what files can be safely deleted? This is an important issue due to the limited disk space.

Please suggest whether any of the following approaches ca be considered safe:

  • prevent creating any .xtc, .edr and .log files with use of the following MDP options:
    nstenergy = 0
    nstlog = 0
    nstxout-compressed = 0
  • do not suppress creating that files but execute “gmx trjconv” just after finishing “gmx mdrun” for the currently simulated window, and select non-solvent atoms only. Then, remove the original trajectory and leave only the processed one.

If I need to re-calculate PMF (e.g. including the error analysis), do I need only the .tpr and pullf files passed to “gmx wham” using appropriate flag, or are there any other files that have to be necessarily stored?

2

gmx wham only uses the files that you provide it. Thus, if you only want to re-run WHAM analysis, the .tpr and pullx.xvg files are the only files that you need to keep.

3

Thank you for your answer.

Does this mean that I am allowed to disable writing the .xtc, .edr and .log files in each MD simulation corresponding to a specific snapshot?

The gmx mdrun command produces both pullx and pullf files automatically if the pull section is defined in the .mdp file. Theoretically, there is even a possibility to disable saving one of them by setting either pull-nstxout = 0 or pull-nstfout = 0.

Is there a recommendation which of them should be passed to gmx wham - pullx or pullf?

4

If you’re certain you don’t need any of them for later analysis or reference, sure (you’re only interested in the PMF, that is).

I’m not familiar enough with the internals of gmx wham to give a definitive answer, but if I understand correctly then yes, only one of them is needed as an input. Which one you take shouldn’t matter too much, but I would go with pullx as the positions will be otherwise just be computed from the pull force, according to the manual.