[AMBER] Finding a good REMD temperature ladder

From: Eugene Yedvabny <eyedvabny.berkeley.edu>
Date: Wed, 9 Jul 2014 11:01:39 -0700

Hello Amber community,

I am stuck on the age-old problem of REMD simulations- picking a good
temperature ladder for my protein. Historically I've used this REMD
calculator <http://folding.bmc.uu.se/remd/>, which Jason has recommended
several times, and didn't really second-guess it too much. But with the
addition of remlog analysis functionality to cpptraj I've wanted to verify
the 'appropriateness' of the ladder for a new protein I am simulating
(short 43-residue disordered mutant of amyloid beta).

Here I've linked <https://www.dropbox.com/s/bj0vxpl5xoaxw0g/remlogs.tar>
the results of a 2 ns REMD simulation with 0.5 ps exchange rate across 64
replicas from 287K to 450K. These files were generated with cpptraj 14
remlog command for crdidx and repidx exchanges across time. The temperature
ladder for run 1 was taken from the aforementioned REMD calculator and
slightly modified for run 2 to address dips in acceptance rate.

As you may see from the attached graphs, there's a weird break in exchanges
around the 55th replica. Replicas below are exchanging (albeit without
round-trips within the 2 ns simulation) and replicas above are exchanging,
but for some reason there's almost no exchange between the two groups. The
behavior suggest there's little overlap between the 55th and 56th replica's
energy distributions, but I can't rationalize why that would be the case
for that particular temperature pair.

Has anyone seen this kind of behavior in REMD simulations using a ladder
from the online calculator? Any suggestions on how to increase the mixing
rate and break that barrier would be greatly appreciated! If you also have
links to other suggested REMD temperate calculators I'd be happy to give
them a try.

Thank you,
Eugene Yedvabny

AMBER mailing list

Received on Wed Jul 09 2014 - 11:30:02 PDT
Custom Search