# [AMBER] What is the best way to do triplicate runs

From: Romero, Maria via AMBER <amber.ambermd.org>
Date: Fri, 8 Sep 2023 19:32:51 +0000

Hello Amber community,

My goal is to generate triplicate aMD runs for my system, but I have been struggling to obtain results that agree with each other to a certain extent. These are the two methods that I have tried so far:
-Starting with one single pdb structure, do prep in cpptraj and minimization and then introduce random velocities in my first step of equilibration using the following script:

nstlim=250000, dt=0.001, ntx=1, irest=0, ntpr=1000, ntwr=10000, ntwx=1000,
tempi=0.5, temp0=300.0, ntt=1, tautp=2.8, ig=-1,
ntb=1, ntp=0,
ntc=2, ntf=2,
cut=10.0, iwrap=1

The results from the 3 trajectories do not agree at all. For example my ligand would randomly leave one of the active sites in one trajectory, both active sites in another and none in the third trajectory.

The second approach I tried was:

-Take 3 different starting structures from an unbiased MD simulation as starting points and using the following script for the first step of equilibration:

nstlim=250000, dt=0.001, ntx=1, irest=0, ntpr=1000, ntwr=10000, ntwx=1000,
tempi=000.0, temp0=300.0, ntt=1, tautp=2.8, ig=-1,
ntb=1, ntp=0,
ntc=2, ntf=2,
cut=10.0, iwrap=1,

Which I guess introduces velocities based on the forces, but again results are very different.

Can someone help? Should I use a different thermostat like ntt=3? My system is dimer protein solvated in water. What is the best way to introduce random velocities that would give me trajectories that are similar to each other? My big concern is reproducibiliity.