Re: [AMBER] Jarzynski SMD

From: Jason Swails <>
Date: Fri, 17 Oct 2014 17:38:47 -0400

On Fri, Oct 17, 2014 at 3:39 PM, George Tzotzos <> wrote:

> I’m trying to explore possible pathways for the exit of a ligand from the
> binding pocket of its receptor.
> I’d appreciate any advice on the procedure that I’m following, which is as
> follows:
> 1. plan to repeat each SMD simulation 3-4 times from different starting
> coordinates taken from a 20ns MD trajectory. Is the number of repeats
> adequate?

​There is a tradeoff with Jarzynski averaging of multiple SMD simulations.
It takes a small number of work profiles that are very close to the PMF in
order to closely approximate the "correct" free energy profile. The slower
you go, the more like a reversible process the SMD simulation becomes --
don't forget that a reversible process is the limit of such a
transformation happening infinitely slowly.

The faster you pull your reaction coordinate, the less likely you will get
a "low" work profile (which means it will take far longer to converge your

So you have a tradeoff -- the faster you pull the RC, the sooner your
calculations will be done. But you will have to do more of them to get a
converged result. If you have a lot of processors available, then you
might as well pull faster and run more simulations concurrently -- that
will get you your answer the fastest. If you are getting a converged PMF
with only 3-4 SMD simulations, you are probably pulling too slowly.​ Or
you may not have nearly enough simulations to get a truly converged PMF.

> 2. I’m changing the distance of the centre of mass of my ligand to that of
> 10-14 residues in the proximity of the original position of the ligand in 1
> Angstrom increments. Is incremental change of distance between the centers
> of mass of 1 Agstrom OK?

​This sounds more like umbrella sampling to me. There are 2 approaches to
SMD that I'm familiar with -- constant velocity (i.e., the center of the
"pulling" potential moves with a constant velocity along the reaction
coordinate) and constant force (i.e., there is a constant force driving the
transformation along the RC). In neither case do you "incrementally
change" the location of the potential. Constant velocity pulling (used via
jar=1 in the input file) automatically moves the center of the potential
with a constant velocity. You implement constant force by using NMR
restraints and making sure that the potential is linear over the range of
the RC you want to pull along.

> 3. I’ve set rk2 = 500. Is this force acceptable?

​It might be OK, but it is awfully stiff. It might be worth seeing what
other studies have used and emulate them, or just try it out and see if it
causes problems (like integration errors for the time step you're using).


Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Fri Oct 17 2014 - 15:00:02 PDT
Custom Search