On Mon, Sep 25, 2017 at 5:52 PM, Andrew Schaub <aschaub.uci.edu> wrote:
>
> I understand that only
> one residue is selected each state change, so each of those residues would
> average 50,000 state changes over the microsecond simulation.
>
Actually this isn't true. Unlike GB, where protonation state changes are
very cheap (since you get to reuse the force calculation and integrate
following attempted protonation state changes), protonation state changes
in explicit solvent are much more expensive. So instead what is done is
that we try to change each residue once (in random order) in sequence.
That's why we were able to increase ntcnstph from ~10 (as was suggested in
Mongan's original paper) to 100 -- on average the number of attempted
protonation state changes for a system with 10 titratable residues is the
same. I believe you'll see this described in our paper introducing the
method.
> Thanks for the clarification on ntrelax. I reviewed the 2012 JCTC paper
> again, and noted that there was minimal difference between 100 fs, 200 fs,
> and 2 ps for the solvent relaxation time. I tried to use a larger ntrelax
> value in the hopes it might be speed up simulation speed, though this would
> have the exact opposite effect (I believe if I'm understanding this right).
>
Yes. The relaxation steps are effectively discarded MD. The more steps
you run to relax solvent, the more of your MD you simply discard (making
the simulations less efficient).
HTH,
Jason
--
Jason M. Swails
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 26 2017 - 12:30:01 PDT
