AMBER Archive (2007)

Subject: Re: AMBER: How precise is a MD simulation?

From: David Mobley (dmobley_at_gmail.com)
Date: Tue Feb 13 2007 - 14:11:17 CST

Yong, Sonya,

> David said it well. Just add that MD is a tool to study thermodynamics. If
> you want a representative structure, as in the case of x-ray crystallography
> or the average NMR structure, you may try to obtain an average structure.
> Single snapshot is not very meaningful, even if it is the snapshot with the
> lowest energy.

I'd add to this a bit and say that you have to be careful even with
average structures -- for example, a X-ray structure is not really an
average structure in the same sense as an average MD structure would
be. An average MD structure in principle would average over a
Boltzmann-weighted distribution of conformations; an X-ray structure
is usually taken at low temperature, and a *single* structure is fit
to the observed electron density; even if the electron density
represented a Boltzmann-weighted distribution of conformations (which
is debatable) the distribution might be from a different temperature
(since diffraction data is typically taken at low T), and the fact
that a single structure is fit to the electron density would mean this
distribution data is missing unless you go back to the electron
density.

Average structures also have known artifacts -- for example, if you
run an MD simulation where a sidechain spends 1/2 time in conformation
A and 1/2 the time in conformation B, the apparent conformation in the
average structure is NEITHER of the conformations, and may be
unphysical.

One thing to look at may be clustering of different sorts. Sometimes
you can use clustering to give you groups of dominant conformations
from MD... If one set of simulations acesses substantially different
conformations from the other, you might be able to pick up on this
from clustering.

David

>
> yong
>
> -----Original Message-----
> From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf Of
> David Mobley
> Sent: Tuesday, February 13, 2007 11:36 AM
> To: amber_at_scripps.edu
> Subject: Re: AMBER: How precise is a MD simulation?
>
>
> Sonya,
>
> > I have done molecular dynamics, explicit simulation
> > of a short (8aa) alpha helix. I followed methods
> > given by the tutorial. I am going to be comparing
> > precise locations of backbone atoms in two such
> > simulations. So, my concern is, how precise is the
> > location of atoms in the simulation. I'm guessing
> > that can't be answered w ith a simple number, but
> > maybe you can tell me how I can figure out the
> > answer from my simulation. Basically, i'm curious
> > to what is the resolution of the result of my
> > modeling.
>
> If I understand right, you're saying you want to compare a particular
> snapshot from simulation 1 with a particular snapshot from simulation
> 2, and draw conclusions based on the differences you see? This is
> probably a very bad idea. MD does not give you a single structure --
> not even in principle. Rather, you simulate the motion of the system;
> in principle, if you simulate long enough, you'll sample from the
> correct Boltzmann distribution of conformations. Maybe what you're
> saying is that you want to compare the low energy conformations or low
> free energy states from simulation (1) from those with simulation (2)?
>
> I would resist the temptation to think of the end structure of one of
> your MD simulations as the "result". It isn't. It's the final
> structure (final snapshot), and is no more meaningful than any of the
> other snapshots (after equilibration).
>
> David Mobley
> UCSF
>
>
> On 2/12/07, Fenghui Fan <fenghui_fan_at_yahoo.com> wrote:
> > In my opinion, maybe you cannot get a stable
> > conformation. You can watch your simulation with VMD,
> > and you can see the conformations change. From both
> > the VMD and RMSD calculation, you can see there maybe
> > several or even more conformations, and you can have
> > each one for analysis.
> >
> > For analysis, will you please take a structure check
> > tool for example Procheck? This will be very important
> > to analysis your structure.
> >
> > I think these will be helpful.
> >
> > Best regards.
> >
> > Fenghui Fan
> > --- "Dave, Sonya" <sonya.dave_at_vanderbilt.edu> wrote:
> >
> > > Hello,
> > >
> > > I have done molecular dynamics, explicit simulation
> > > of a short (8aa) alpha helix. I followed methods
> > > given by the tutorial. I am going to be comparing
> > > precise locations of backbone atoms in two such
> > > simulations. So, my concern is, how precise is the
> > > location of atoms in the simulation. I'm guessing
> > > that can't be answered w ith a simple number, but
> > > maybe you can tell me how I can figure out the
> > > answer from my simulation. Basically, i'm curious
> > > to what is the resolution of the result of my
> > > modeling.
> > >
> > > My guess is that I plot RMSd of the backbone atoms.
> > > I usually do that just to make sure it has
> > > stabilized. But, would the value of RMSd represent
> > > the resolution?
> > >
> > > Another thing, I ran the simulation, and the final
> > > product didn't really look like an alpha helix, it
> > > was somewhat compressed/distorted. However, I
> > > extracted the structure of the lowest potential
> > > energy state (lowest EPTOT that was ever experienced
> > > in the MD). That resembled a helix much more - But
> > > still not perfect. So, obviously, in different
> > > steps there is significant differences in geometry.
> > > This concerns me, because I am concerned about small
> > > distances here. So, how can I be fairly confident
> > > my structure is the best structure?
> > >
> > > Thank you,
> > > Sonya
> > >
> >
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu