I have two all atom molecule dynamics systems of a protein. One is a wildtype and the other is a mutant. They differ by amino acid sequences. I am running my simulations in Gromacs. Each system has ran for 150 ns, showing certain system and physiological properties converging over the final 50 ns. Prior to this, each system will have gone through a series of short simulation steps where I relax position restraints on the protein and where I switch from the Berendsen pressure coupling to the Parrinello(sp?!) pressure coupling.
I am not entirely happy with sampling structural analysis from just one trajectory per system. My options could be:
1) Run the system for a lot longer and take results over a longer period of time, or,
2) Run triplicates per system; three for wildtype and three for mutant.
Taking three per system would involve three completely different sets of velocities, thus requiring the whole equilibrium procedure of relaxing through two different pressure coupling, relaxing position restraints on the protein etc... That's a lot of work all over again.
I was wondering whether it would be appropriate to take the final velocities from the system equilibrium step (just before I ran the 150 ns simulation, so I am talking about the final velocities from 1 ns where I have removing my position restraints from the protein backbone) and performing a 150 ns simulation, but the first 1 ns is a simulated annealing step of raising the temperature and then dropping it back down again.
My question: having used a set of preserved trajectories, by raising the temperature and then dropping back down, have I now accessed a new set of velocities and thus a different region or direction into conformational space, or will I simply shadow the earlier trajectory just a little sooner (having raised the temperature)?
I doubt a brief temperature spike will yield anything productive. Replicate simulations are the standard approach and are trivially easy to set up, just change the seed and re-use all existing inputs.
The best approach would be real replica exchange (temperature or Hamiltonian, though with the former you may be prone to unfolding, which would be unproductive), or some form of locally enhanced sampling (in AMBER, not GROMACS AFAIK).
I doubt a brief temperature spike will yield anything productive. Replicate simulations are the standard approach and are trivially easy to set up, just change the seed and re-use all existing inputs.
The best approach would be real replica exchange (temperature or Hamiltonian, though with the former you may be prone to unfolding, which would be unproductive), or some form of locally enhanced sampling (in AMBER, not GROMACS AFAIK).
Always something new to try! Thanks for confirming my suspicions regarding the temperature spike.
It's never very appropriate to heating up a protein system in MD. Normally you want to find the thermal equilibrium in NVT or NPT at T = 300 or 310 K. Your protein will unfold with increasing temperatures.
- Badges
- Science topic
- Similar topics
- Chemistry
- Theoretical Chemistry
- Computational Chemistry
- Molecular Dynamics