Re: [AMBER] How many complex trajectories for mmpbsa computations ?

From: Liao <>
Date: Sun, 10 Oct 2021 18:39:36 -0500

Let me quickly add-another way to interpret the reviewer’s comment; I see that you mentioned providing the average coordinates picture of the highest affinity compound. That would be the ideal pose found. If running the simulation multiple times, would you still get this same pose? If not, then this may not be the ideal pose-I think that’s what they’re worried about. There needs to be some extra corroboration that those 24 ns are actually representative enough.
Presumably you started with a docking pose, and it’s trustable, then I don’t any point in running long simulations. Unless you’re looking for protein changes, etc. If you want to confirm or improve that docking pose, then the long simulations could be the point.

> On Oct 10, 2021, at 6:08 PM, Liao <> wrote:
> Please allow me to also comment and discuss here - any feedback would be greatly appreciated.
> If talking about the protein-ligand binding energy calculation based on MMPBSA, I believe from my experience the foremost important issue is that the ligand is in the correct binding pose within the pocket. The reason why multiple short simulations give better correlation to experiment than a single long simulation, could be because in the long simulation the ligand moves out of the ideal pose (if we start from a known crystal structure). Please see this paper ( table 3:
> They actually drew to a conclusion at the end of this section 3 that "Therefore, long MD simulations do not necessarily lead to accurate binding free energy calculations when the single-trajectory protocol is used." (if not using seperate trajectories for complex, ligand and protein).
> This is totally consistent with the explanation that if one already found the correct binding pose, then if the ligand leaves the pocket or correct pose in a longer simulation the energy will be off. While it probably shouldn't leave so soon in reality, but it could happen in simulation. So as long as one has found a good binding pose, the binding energies are better off based on multiple short trajectories (or even single point after good minimization? I'm yet to experiment more with that), given all trajectories are sampling that same pose. A very long simulation merely for the point of obtaining binding energy, AFTER one already has the desired binding pose, could be counter-effective. What are the experiences of other folks here, on this specific topic?
> Thanks very much!
>> On Oct 10, 2021, at 3:20 PM, Cenk Andac <> wrote:
>> Dear Amber community,
>> We have submitted a research manuscript to a scientific journal. The
>> manuscript is about MD simulations and MMPBSA binding energy
>> vomputations that involve extensive use of amber suite of programs.
>> We have 11 compounds ( 4 reference compounds + 7 compounds synthesized in
>> our laboratory) targeted to a protein. We have implemented ~370 ns MD
>> computations for each ligand+receptor system. Then we computed MM-PBSA
>> binding energies for each complex systems using 100 snapshots extracted
>> from the the last 24 ns of trajectory.
>> We then rank-ordered the binding energies.
>> In the manuscript we discussed binding energies for the reference and
>> synthesized compounds. We also put in the manuscript a nice picture of
>> averaged coordinates of the highest-affinity ligand in complex with its
>> receptor
>> One of the reviewers responded to us with the following request :
>> *Reviewers request:* A single MD simulation is not sufficient. For each
>> compound, three replicas of 200+ ns MD simulations should be performed.:
>> *Our respond: *Thank you for the invaluable comment.
>> running several MD simulations for a complex system will not affect MM-PBSA
>> computations much provided that the MD simulations equilibrated well. The
>> idea with MM-PBSA method is that multiple snapshots (best is more than 50!)
>> are collected over a certain time interval from equilibrated trajectory.
>> Energy computations are implemented for each snapshot and finally averaged
>> out to give rise to an average binding affinity value over dynamics motions
>> of the complex system at equilibrium in solution. Of course, different MD
>> simulations of the same complex system will yield different snapshots but
>> should give rise to very similar averaged MM-PBSA energies at equilibrium.
>> *The reviewer then responded* : Three replicas of 200+ MD simulations are
>> required for each compound to make sure that the simulation runs are
>> consistent, or to determine a way of using certain average. This was a
>> trend before and is a convention now in MD simulations.
>> My question here is that is it really necessary to run 3 replicas of MD
>> simulations to average out ligand+receptor interactions ?
>> Thanks in advance.
>> Jenk Andac.
>> _______________________________________________
>> AMBER mailing list
> _______________________________________________
> AMBER mailing list

AMBER mailing list
Received on Sun Oct 10 2021 - 17:00:02 PDT
Custom Search