Discrepancy in Tripeptide Aggregation Behavior Between GROMACS 2020.1 and 2023.3

[gdellorl]

GROMACS version:2020.1 and 2023.3

Hi,

I am experiencing a reproducibility issue between two versions of GROMACS and would appreciate your input.

We performed molecular dynamics simulations of tripeptides using GROMACS 2020.1 with the CHARMM36m force field. Over time, we observed the formation of small peptide clusters, as expected.

However, when restarting the simulation from the last frame of the trajectory and running it with GROMACS 2023.3 (using exactly the same parameter files, topology, and input settings), the clusters disaggregate very quickly.

Do you have any suggestions of what could be the cause of this issue?

Thanks in advance,

Gianluca Dell’Orletta

[hess]

There should not be any changed in the code that affect the physics. Have you tried continuing in the same way using 2020.1?

Note that MD is chaotic. You would need to run at least 3 repeats to be able to draw conclusions with some certainty.

[gdellorl]

Thank you for your answer.
No, I haven’t yet continued the simulation with GROMACS 2020.1, but I did try using version 2021.2, and the results look consistent — the clusters remain stable throughout the simulation (up to 750 ns).
On the other hand, with version 2023.3, we observe an unexpected behavior where the clusters rapidly disaggregate.

I understand the need for running replicates. However, the results so far are quite unusual — the disaggregation effect occurs within just ~100 ps, which seems physically unreasonable.

For clarity, I’m attaching the SASA profiles from the simulations run with GROMACS 2020.1 and 2023.3.

[hess]

That looks like a serious artifact. How did you run the continuation simulations? GPUs? How many MPI ranks?

[hess]

It looks like the artifacts appear very quickly. Could you check if the issue is also present with the latest version: 2025.2?