Re: Accelerated MD in NAMD

From: Jeff Wereszczynski (
Date: Thu Nov 07 2013 - 08:19:53 CST

I just have a couple things to add to this discussion:

1. The energy thresholds you use for deciding on your aMD parameters
should be based on the potential energy and not the total energy (potential
+ kinetic). In aMD you are modifying the potential energy surface of the
system, so you shouldn't use the kinetic energy when setting the aMD

2. In Irina's paper that you mention, she used a version of Desmond with
an aMD plugin I wrote that allowed for selectively boosting different parts
of the system. Not only is that functionality not in NAMD, but the plugin
doesn't work in any version of Desmond past 2.2 so its effectively died
off. So unless you want to do some coding, you're going to have to boost
the entire system.

3. You might also find this paper interesting:
There is some discussion in the SI about choosing aMD parameters for a
dual-boost setup for a different GPCR that may be helpful.

4. How you present your data depends on what you are trying to present.
 If you are trying to show that your system can go between two x-ray states
then you may want to project into a PC space defined by your x-ray
structures (assuming you have a decent number of x-ray structures). If you
are trying to show the conformational space accessible to your simulations,
you may want to define your PC space based upon your simulation structures.
 It really depends on what you are looking for in your analysis.

Good luck!

Jeff Wereszczynski
Assistant Professor of Physics
Illinois Institute of Technology

On Thu, Nov 7, 2013 at 6:48 AM, James Starlight <>wrote:

> Thomas,
> thanks for suggestions!
> 1) I also would rise question about precise values of the <Udihe>. As a
> test I decide to simulate the same b2a receptor in the 120 lipids bilayer
> in cMD (npt conditions, 2ps damping, 310K. all parameters have been taken
> from charm-gui). After short simulation I've obtained total <Udihe>=10000
> Kcal/mol (2000 for protein and 8000 for lipids).
> So my boost should be based onto the E=10000+4*(340+120) <! here I'm not
> sure how include lipids so I include just number of lipids > ~ 12000 and
> alpha should be ~ 400-500.
> In comparison In Tikhonova paper I found the same values of <Udihe> ~
> 1300 kcal/mol for protein and 900 (!) for lipids. Does it meas that my
> system in need not for the boost potential at all ? :)
> 2) About projection. In my example I have 80% variance along 1 PC (here
> all structures have been grouped on R and R*) and 20% along other 10 modes
> (total 10 x-ray structures= 10 modes). On other hand on MD trajectory I
> have only 40% variance of 1 PC ( I didnt notice large-scale conformation
> changes on this mode corresponded to the R-> R* transition). Ther fore I
> should to project MD ensemble onto PC calculated for the X-ray dataset,
> shouldnt it? :)
> James
> 2013/11/7 Thomas Evangelidis <>
>> On 7 November 2013 09:09, James Starlight <> wrote:
>>> Professor Wereszczynski,
>>> also I've seen the implementation of the aMD to the GPCR system (work of
>>> I.Tikhonova) where only dihedral boost have been applied to the protein and
>>> lipids as two separate terms.
>> The authors state that the boost dihedral potential was applied on the
>> receptor alone as well as the receptor and lipids together (confusing).
>> They also did a lot of experimentation with E threshold and alpha
>> parameters (see Supporting Info) to find the optimal combination that
>> doesn't distort the secondary structure of the protein too much and does
>> not perturb the POPC membrane excessively (area per lipid headgroup, lipid
>> bilayer thickness, etc.)
>>> Does it means that the transition to the active state of the receptor is
>>> primarily structural dependent event? (this time you have not used total E
>>> boost which are influence on the diffusion rates). Doest it takes from the
>>> assumption that the receptor is fixed in the membrane so we might not take
>>> into account diffusion ?
>>> some off-topic- in this paper I've seen interesting methodology when you
>>> compare MD observations with the distribution of the X-ray structures of
>>> this receptor (solved in active and inactive states as such reference
>>> points). My question is in methodology- will it more correct to project
>>> C-alpha coordinates of X-ray structures onto the Principal modes calculated
>>> for the MD trajectory or alternatively project MD snapshots onto the
>>> Principal modes calculated from the ensemble of the X-ray structure ?(!) In
>>> last case we have robust assumption that X-ray ensemble is the trajectory
>>> capturing R->R* transition seen in experiment (!) So lowest frequency modes
>>> gives evidence about possible path of this transition if we have number of
>>> intermediates. Alternatively in the first case (as has been done in the
>>> Tikhonova paper) we have assumption that accelerated simulation with the
>>> artificial boost can be the reference for the monitoring some biological
>>> event (R->R* transition) so in this case we test X-ray structures (not the
>>> simulation setup). Which statement would be most probably? Does somebody
>>> see the implementation of such methodology (projections of the X-ray
>>> structures onto MD and vice versa) with other proteins ?
>>> The eigenspace derived from the MD trajectory will be much larger than
>> that derived from 3 or so X-ray structures. Therefore projecting the the
>> X-ray coordinates onto the PCs of the MD trajectory makes more sense than
>> the reverse. The other important detail you have to consider when you
>> present that heatmap is how much of the variance of the atomic fluctuations
>> the first two PCs explain. If for example they explain just 30% then it
>> doesn't make much sense to present it.
>>> --
>> ======================================================================
>> Thomas Evangelidis
>> PhD student
>> University of Athens
>> Faculty of Pharmacy
>> Department of Pharmaceutical Chemistry
>> Panepistimioupoli-Zografou
>> 157 71 Athens
>> email:
>> website:

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:23:57 CST