Re: [AMBER] Reweighting aMD conformations

From: Matthew Guberman-Pfeffer <>
Date: Wed, 14 Jul 2021 09:42:51 -0400

Dear Charo,

Thank you for the quick and helpful reply. I re-noticed in the amd output that the unboosted energy is printed, which caused me to wonder “why is re-weighting even needed then?”

Some specific follow-ups to your comments:
> So, first of all, my suggestion is to use the smallest possible boost that allows you to access the transition you are interested in studying.
Would that require prior knowledge or systematic testing with different boosts to find out how high the barrier(s) is/are?

> However, even doing so, it is highly likely that you will
> have some weights that are quite high, and higher than the others. This is practically certain to happen if you have a large protein complex, as you describe.

Yes, the protein complex is a dimer, each part of which has 407 residues. From the output, I see that the boost potential on the total energy ranges from 0.1 to 28.5 kcal/mol, and the dihedral boost ranges from 79.9 to 203 kcal/mol. The corresponding averages are 8.9 and 135 kcal/mol. My ethreshp and ethresdd values for the simulations were -543391 and 12149 kcal/mol. These thresholds were computed based on the suggestions in the Amber tutorial.

> In turn, this means that the algorithm
> you use for the reweighting has to be carefully written, to avoid various numerical instabilities that can occur. Currently, I am working exactly on such a simulation, and I wrote a code that uses
> MPFR (on top of GMP), so I can choose the precision I want in the calculation. The averaging itself is done by using logarithmic sums and a few other such tricks.
If you would be willing to discuss how to write my own code or to share your's off list, I’d be very grateful for the help — or given the context “boost” over the conceptual and technical barriers involved.

> --
> Dr. Charo I. del Genio
> Senior Lecturer in Statistical Physics
> Applied Mathematics Research Centre (AMRC)
> Design Hub
> Coventry University Technology Park
> Coventry CV1 5FB
> UK
> _______________________________________________
> AMBER mailing list

AMBER mailing list
Received on Wed Jul 14 2021 - 07:00:02 PDT
Custom Search