[gmx-users] gromacs+CHARMM36+GPUs induces protein-membrane system collapse?

[Ramon Guixà]

Dear all,

After several years simulating membrane proteins using GROMACS and the
CHARMM force field, I have come across this issue:
I use the CHARMM-GUI interface to prepare my membrane-protein system, that
I subsequently equilibrate using a standard protocol. However, after
several ns, namely 100 to 400 ns, systems become extremely unstable (i.e.
very large box fluctuations) until they eventually deform (i.e. membrane
abnormally enlarged, big water pores go through, and protein collapses). I
have tried several things, namely:

1 - I changed (increased/decreased) certain parameters within the mdp file
(e.g. tau-p).
2 - I started from scratch, re-built and re-equilibrated the system.
3 - I used different GROMACS versions, namely 5.4, gromacs2016 and
gromacs2018.
4 - I used different computers and different GPUs.
5 - I changed the system (other membrane proteins of the same family with
different ligands).

Sooner or later (within 100-400 ns), same thing happens over and over again
in every protein-membrane system I simulate. Intriguingly, systems never
crash (i.e. simulations keep on running despite of the observed artifacts).
The only exception out of all tests so far happens upon running the system
using CPUs only. In the former test, I observe no fluctuations so far (500
ns).
I never came across this problem before using the same set-up, could it be
linked somehow to the updated version of the CHARMM force field
(CHARMM36m)? Any thoughts?
Please find below a link to a zip file with tpr, mdp, raw (skipped)
trajectory, init/end wrapped pdb files, and boxZ plot from one test, namely
one using gromacs 2018.3+GPU:
https://drive.google.com/open?id=1s3FWbk91danJW863BN-GqsgraSIore6D

Many thanks in advance for the help!