GROMACS version: 2021.3
GROMACS modification: No
Dear users,
I have read some papers performing MD simulations in “replica”.
Looking for the term in the forum, many questions are about the GROMACS replica-exchange function. Asking for some authors, told me they performed individual simulations.
What is the best way to perform the MD simulation in replica?
*Or do I understand the replica exchange in the wrong way?
Thank you for your help
A “replicate” is simply another simulation of the same system but from a different starting condition (e.g. different initial velocities).
A “replica” in replica-exchange is a simulation performed in the context of others with which it may exchange based on energy.
Thank you for your answer @jalemkul,
The difference in the initial velocity is the most common way to produce the replicates? I didn’t read in those papers the source of initial difference.
Ok
Most commonly applied, yes. This is because it is very easy to simply generate further tpr files containing different velocities using all the same inputs for grompp.
However, there was a paper a few years ago (which I cannot seem to find from a quick search) that IIRC suggested this wasn’t the best approach in terms of improving sampling given typical simulation timescales. If I find the paper I shall include a link to the work in this thread.
Depending upon the system you are studying, other ways to generate replicates could include things like using a different force field or using a different starting structure (e.g. from different experimental sources or generated models). I tend to use combinations of such strategies to perform multiple repeats.
All that said, you need to always keep in mind what you are trying to observe in your simulation, how much computing power you have, and how you might validate your results. This should guide you in how you go about your simulation strategy (e.g. many short simulations vs fewer longer simulations vs applying enhanced sampling methods, etc.)