[gmx-users] Hamiltonian Replica Exchange

[Szilárd Páll]

On Fri, Jun 27, 2014 at 11:36 AM, Thomas Evangelidis <tevang3 at gmail.com> wrote:
> Thank you all for your comment!
>
> The helix dimer is not very stable and I know that from extensive
> accelerated MD simulations I've done with AMBER using various boost values.
> Now I want to use a different force field available only in GROMACS, and
> though partial solute tempering HREX as the most suitable enhanced sampling
> technique considering the available computer resource I have.
>
> I know about the PLUMED implementation of partial tempering scheme. My
> question was if I can do the same with GROMACS because the speed with
> plumed is significantly impaired compared to serial execution of GROMACS:
>
> 3.10 ns/day / Replica with GROMACS 4.6.5 double precision + PLUMED 2.0.2 on
> a single node
> 5 ns/day with GROMACS 4.6.5 double precision on a single node (serial
> execution, only one replica)

Note that unfortunately your slowest replica will determine the
overall speed of the run and if you have significant performance
differences between replicas, you may end up wasting (5-3.1) ns/day.

Also note that if you were comparing an Nx(1 thread/replica) run's
performance with a single threaded run, the comparison is *not* valid!
Most modern processors use frequency boosting and if you only use one
(or a few) core(s) you'll get much higher clock frequency compared to
what you get when all cores are used! Try to run vanilla GROMACS 4.6.5
with -multi too and you'll probably not get 5 ns/day anymore.

Could you upload some log files to e.g. pastebin and share them here.
Based on those we could tell where the overhead is - although you may
want to verify what's the slowest replica's performance in a separate
run because I think the multi-run's overhead can't be separated from
the PLUMED overhead without code change.

Cheers,
--
Szilárd

> I don't know if it is the replica exchange scheme that affects the speed so
> much, or the "communication" with PLUMED.
>
> On another note, I would like to know you opinion about a couple of ideas I
> have:
>
> 1) In order to reduce the replicas needed to achieve 30-50% exchange rate
> for a given lambda range, I though to selectively heat only the main chain
> and CB carbons of the two disordered ends.  Although I don't see anything
> wrong with this (except that the system may explode in low lambda
> replicas), I would be interested to know the opinion of more experienced
> people about possible caveats of the above mentioned rationale.
>
> 2) Do you reckon partial solute tempering HREX is suitable to monitor
> protein-ligand interactions? In my understanding, ligands will tend to stay
> more in touch with "cold" residues of the protein than with "hot" and hence
> the results may be misleading. Do you think I should also heat up the
> ligands to speed up their diffusion?
>
> thanks,
> Thomas
>
>
>
> --
>
> ======================================================================
>
> Thomas Evangelidis
>
> PhD student
> University of Athens
> Faculty of Pharmacy
> Department of Pharmaceutical Chemistry
> Panepistimioupoli-Zografou
> 157 71 Athens
> GREECE
>
> email: tevang at pharm.uoa.gr
>
>           tevang3 at gmail.com
>
>
> website: https://sites.google.com/site/thomasevangelidishomepage/
> --
> Gromacs Users mailing list
>
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-request at gromacs.org.