My REMD it is giving me 0.0 of probability

User discussions
1

GROMACS version:2023
GROMACS modification: No
Hello everyone,

I’m running a replica exchange simulation without PLUMED and could use some guidance. Here’s the command I’m using:

srun -n SLURM_NTASKS gmx_mpi mdrun -deffnm protein_1 -s protein_1.tpr -multidir {1..240..15} -replex 1000 -gpu_id {gpu_ids} -nb gpu -pme gpu

After 50 ns of simulation, I’m seeing the following probabilities for replica exchanges:

Repl ex 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Repl pr .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00

My system comprises about 80,000 atoms in the protein and 79,000 in the solvent. I initially aimed for 275 replicas, using temperature values from Virtual Chemistry with a target exchange probability of 0.2. However, due to resource constraints, I’m now using 16 temperatures ranging from 310 K to 450 K in 8 K increments.

Since all the probabilities are zero, I’m looking for advice on how to proceed. Should I adjust the temperature range or step size, or would integrating PLUMED be a better option?

I’d greatly appreciate any suggestions or insights you can provide!

Best regards,
Jorge

2

Try with 0.5 K increments first (or even lower), just to see that you get replica exchanges (310 K to 317.5 K). Try simulating for 10 ns. If you still don’t get exchanges we’d need to see more of your parameters.

If Virtual Chemistry (which I haven’t used myself) suggested 275 replicas (approximately ~0.5 K per increment if the temperature range was the same) for an exchange probability of 0.2 I think you will need to reconsider your temperature range or use more than 16 replicas.

3

My rough estimation (with Temperature generator for REMD-simulations) is that 200 replicas might give ~0.2 probability and 110 might give an 0.01 probability. So, 16 replicas is optimistic.

4

Thank you very much for your reply #MagnusL, then there is no option to reduce the number of replicas, right?

5

With such a large system, REMD requires very many replicas (grows with the square root of the number of degrees of freedom in the system). There are other REMD-like techniques (e.g. local REMD and partial REMD) where you only affect a subset of the system. I don’t think (but don’t know for sure) that they can be used natively in GROMACS. I would think that Colvars (available as a plugin in GROMACS 2024) or Plumed (still only available by patching GROMACS) could help you.