Technically aMD /is/ a form of umbrella sampling. However, the latter
has now come to mean something very different from its original coinage
and is almost always meant to imply a series of harmonic biases along a
specific collective variable (or variables) in order to stratify a
transition (usually chemical or conformational). This is often, but not
necessarily, then used to compute a PMF.
However, even this is not strictly correct and mostly derives from
assumptions used in deriving the weighted histogram analysis method
(which originally assumed a density of states in the collective
variable, which could also be taken as the potential energy).
Contemporary derivations of WHAM (MBAR, UWHAM, dTRAM, etc.) now dispense
with the computation of the PMF as part of the initial set of equations
and do not require it to be calculated for a valid solution.
My first point is largely semantics, the rest is something of a 30 year
history. I'll stop now for fear of coming across as too pedantic to
those who already know this quite well.
In summary, you can always choose a bias potential to hopefully increase
sampling globally (as in aMD) or to concertedly localize sampling along
some coordinate (what is now referred to as umbrella sampling). You can
also use temperature based methods like SGLD and its variants, but this
is not as popular. All of these methods are most effective when you have
a good idea of what it is you want to see or calculate from the
simulation. There is not, to my knowledge, a generally applicable "best"
simulation method for exhausting phase space beyond running a whole
bunch of long timescale MD, perhaps with replicates.
A general tactic now seems to be to either 1) monopolize a whole bunch
of GPUs or 2) get time on one of the Anton machines and see if something
interesting happens and then follow up with more detailed simulations to
extract interesting quantities (e.g. free energies). The former seems
like the easier and cheaper approach favored by the AMBER community.
Regards,
Brian
On 11/11/2015 07:10 AM, Jason Swails wrote:
> On Wed, Nov 11, 2015 at 6:13 AM, Ofir Tal <pantufel.hotmail.com> wrote:
>
>> Dear members,
>>
>> 1.
>>
>> I simulate large system (~250 thousand atoms) of membrane protein dimer
>> embedded inside a POPC membrane on 2X gtx980 GPUs.
>> The system and its variant ran 200ns. The systems are stable in terms of
>> energy and area-pre-lipid. RMSD seems to keep constantly raising along the
>> simulation, although less in comparison with the first 50ns. RMSD increased
>> by ~1A during the last 150ns.
>> I want to enhance the systems sampling sufficiently, so I think it need to
>> be sampled in one of the methods - aMD or umbrella.
>>
>> Which of the methods would be the best choice for sampling a big membrane
>> system?
>>
> They are completely different and used for completely different purposes.
> Umbrella sampling gives you a PMF -- which is effectively a free energy
> along a *particular* degree of freedom (or multiple degrees, if you do
> multi-dimensional umbrella sampling). To do this, you need to know, in
> advance, what degree of freedom you want to enhance sampling along.
>
> aMD is used when you *don't* have a particular degree of freedom you want
> to enhance sampling for. It simply boosts the sampling of many degrees of
> freedom "blindly".
>
> These two methods are rarely (if ever) relevant to the exact same problem
> (so the choice as to which is "better" for a particular application is
> almost always obvious).
>
> Is it possible or sufficient to run the simulations in triplicates with
>> different random seeds (200ns each) instead?
>>
> "Sufficient" is hard to define, and depends on what you're trying to
> learn. This is sufficient for *some* things, but not others I am sure.
>
> 2.
>> As I understand from the literature, MMPBSA.py in general is not suitable
>> for membrane systems. Is it possible to calculate binding energy values (of
>> the membrane complex) using just the 'Decomposition' method of MMPBSA,
>> directed to the interaction area of the two sub-units (only between
>> residues which participate the interaction of sub-unit-A and sub-unit-B)?
>>
> I wouldn't do that. You can run an end-state free energy type of analysis
> "by hand" using PBSA, since it supports membrane calculations (but in this
> case, the membrane is implicit, not explicit). But MMPBSA.py will not
> realistically be able to handle this system.
>
> HTH,
> Jason
>
--
Brian Radak
Postdoctoral Scholar
Gordon Center for Integrative Science, W323A
Department of Biochemistry & Molecular Biology
University of Chicago
929 E. 57th St.
Chicago, IL 60637-1454
Tel: 773/834-2812
email: radak.uchicago.edu
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Nov 11 2015 - 08:00:04 PST
