Re: [AMBER] Temperature-Based Replica Exchange: Number of Replicas

From: Chris Neale <>
Date: Wed, 21 Mar 2018 20:40:58 -0600

I use this one: but
the page seems to be down.

This one is simpler:

I agree with Vinicius that you’re going to need a lot of replicas, but it’s
not necessarily hopeless. You might need 150 replicas to go from 310 K to
450 K. There are groups that could handle that nowadays. And, depending on
what you are doing, you might not need to go up to 450 K.

I’ve played around with the REST/REST2 HREX approach and unfortunately it’s
a bust for folding proteins. You can either get good replica mobility
across the ladder and not good sampling enhancement at the low lambda value
(REST2) or you can get poor replica mobility across the ladder and good
sampling enhancement at low lambda (REST). I’ve looked for intermediates
but so far with no great success. I’d not waste your time on it unless you
can find a paper that at once shows good replica mobility across the entire
ladder and good sampling enhancement at low lambda (and if you do find such
a paper, I’d be keen to see it please).

The big problem is that AMBER is not going to let you do pressure coupling
with REMD. The code modification to add the pV term is actually quite
simple, but the issue is then that the domain decomposition grid does not
get re-initialized periodically and a replica starting at high temperature
and going to low temperature needs to have grids rebuilt but they are not
so the run will crash.

On Wed, Mar 21, 2018 at 5:37 PM, Cruzeiro,Vinicius Wilian D <> wrote:

> Hello Jason,
> The number of replicas required to cover a given temperature range
> increases with the total number of atoms. Therefore, for a system as large
> as yours (and mainly in explicit solvent) T-REMD becomes impractical
> because the number of replicas required for a reasonable exchange rate
> would have to be very large.
> In order to check if a given distribution of temperatures is good, you can
> run a short simulation and check your rem.log file.
> I have seen publications in which people use Hamiltonian-REMD instead of
> T-REMD for large system in order to have a good and efficient temperature
> sampling. I know they scale the Hamiltonian of each replica to a different
> effective temperature in order to mimic what happens in T-REMD and good
> exchange rates are obtained and with a reasonable number of replicas. I
> have never done calculations like this before though, therefore I don’t
> know the technical details.
> Best,
> Vinícius Wilian D Cruzeiro
> PhD Candidate
> Department of Chemistry, Physical Chemistry Division
> University of Florida, United States
> Voice: +1(352)846-1633<tel:+1(352)846-1633>
> On Mar 21, 2018, at 6:16 PM, Jason Ku Wang <<mailto:
>>> wrote:
> Hi AMBER community!
> I'm interested in running temperature-based replica exchange for a soluble
> protein of ~140,000 atoms in a periodic box with explicit solvent.
> I'm having trouble determining the number of replicas and the corresponding
> temperature distribution. Are there any resources/protocols that you have
> found helpful for determining the temperature distribution?
> Furthermore, how does one know that a given temperature distribution is a
> good one?
> Thanks in advance!
> Jason
> --
> Jason Wang
> Stanford University, Class of 2018
> ᐧ
> _______________________________________________
> AMBER mailing list
> ambermd.org_mailman_listinfo_amber&d=DwIGaQ&c=
> pZJPUDQ3SB9JplYbifm4nt2lEVG5pWx2KikqINpWlZM&r=LIQu8OlVNKmzfbMg9_5FnKrt9-
> DrdQBJXyFyocKAWXc&m=jK39qIzt9rWqlKxphJB87p4Csxzwm-FH4yZPswap5Vw&s=
> wu5Fifv3rs5gFAZnmyTm3jCqliDnkCFZlMPRrDGxQHw&e=
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Wed Mar 21 2018 - 20:00:02 PDT
Custom Search