Production runs

Overview

Teaching: 10 min
Exercises: 0 min

Questions

• How long should I run my production runs?

Objectives

• Performed production runs using pmemd.

Production runs

Finally, we are moving forward to the production runs, which are usually performed in an HPC environment. Therefore we won’t use these workstations to run the simulations, but we will go over the files needed to run these simulations and will use a simulation we’ve run for you.

The mdin of the production runs is very similar to the equil_classical.mdin but no restraints are applied this time. We will perform 5 runs of 10ns.

Production runs
&cntrl
  imin= 0,                         ! Run molecular dynamics.
  nstlim=5000000,                  ! Number of MD-steps to be performed.
  dt=0.002,                        ! Time step (ps)
  irest=1,                         ! Restart the simulation and read coordinates and velocities from the restart file provided in -c
  ntx=5,                           ! Initial file contains coordinates and velocities.
  ntpr=1000, ntwx=1000, ntwr=1000, ! Output options
  cut=8.0,                         ! non-bond cut off
  temp0=298,                       ! Temperature
  ntt=3, gamma_ln=3.0,             ! Temperature scaling using Langevin dynamics with the collision frequency in gamma_ln (ps−1)
  ntb=2,                           ! Periodic conditiond at constant pressure
  ntc=2, ntf=2,                    ! Constrain lengths of bonds having hydrogen atoms (SHAKE)
  ntp=1, taup=2.0,                 ! Pressure scaling
  iwrap=1, ioutfm=1,               ! Output trajectory options
/

We execute the simulations using pmemd.MPI which uses similar options as sander. These commands may change slightly depending on the system you use to run the simulations.

mpirun -np 48 $AMBERHOME/bin/pmemd.MPI -O -i mdin -p system.parm7 -c system.md0.rst7 -x system.md1.nc -r system.md1.rst7 -e system.md1.ene -o system.md1.out -ref system.md0.rst7 -inf system.md1.info

The resulting files are the same than before, sander produces several output files:

• system.md1.out: Log file with thermodynamic data of the system stored during the run.
• system.md1.rst7: coordinates and velocities to restart the simulation.
• system.md1.nc: trajectory in netcdf format of the simulation.
• system.md1.info: Information on the performance of the simulation.

Usual points to take into account when running MD simulations

Replicates

As in any experiment, a number of replicates should be performed (The minimum amount of replicates is 3). You should equilibrate each replica separately and initiate it with a different seed number.

ig=-1,                    ! seed for the pseudo-random number generator will be based 

This line in the sander_heat.in file sets a random seed number to ensure the different assignement of velocities after the minimisation step.

Length of the simulation

How long should a simulation be depends on the process you want to study. It is worth to note that so far the high end of the length of the typical MD simulations is around the low µs scale. However, biological processes happen on a wider range of time scales beyond the timescale available to MD simulation.

NVT or NPT ensembles

It is OK to use any of them except when your particular system requires a determined NPT ensembles such as membrane protein simulations.

Key Points

• Production runs are usually run in an HPC or local cluster environments.

• The length and number of replicates of your runs depends deeply on the hypotheses you want to prove.