[gmx-users] Obtaining PMF for change in domain position

[Justin Lemkul]

On 12/29/14 11:34 AM, Abhi Acharya wrote:
> On Mon, Dec 29, 2014 at 8:59 PM, Justin Lemkul <jalemkul at vt.edu> wrote:
>
>>
>>
>> On 12/29/14 6:57 AM, Abhi Acharya wrote:
>>
>>> Hello GROMACS Users,
>>>
>>> This is a problem I am facing for the first time. Kindly guide be to the
>>> best options.
>>>
>>> I have a protein which has two large domains connected by a flexible
>>> linker
>>> peptide (~10 aa). The two domains seem to interact with each other and
>>> have
>>> been crystallized in three different conformations. I want to calculate
>>> the
>>> change in binding energy of the two domains wrt change in their relative
>>> position, i.e. keeping position of one domain constant what is change in
>>> binding energy as the other domain moves from conformation 1, through
>>> conformation 2 to finally, conformation 3. What is the best way to do so?
>>>
>>>
>> You need to describe what these three conformations are.  Do they involve
>> rotations of the domains with respect to one another, or is it a simple
>> linear distance that varies?
>
>
> Yes, there are significant rotations+translations along different axes
> involved which makes it challenging. Is there a way to model such movements
> using the pull code?
>

Perhaps the enforced rotation options can be useful here.

>>
>> At first, I considered using umbrella sampling to address the problem. Here
>>> too, I was not sure how to write a pull code for such a complex reaction
>>> coordinate. Also, even if I can generate the conformations, what would be
>>> the ideal number of windows that would be needed to correctly generate a
>>> PMF ? My feeling is it would be a large number, but again it just a guess.
>>> Is there is a better way than this?
>>>
>>>
>> Hard to know a priori.  You may need some trial and error here, but with
>> umbrella sampling it's trivial to just go back and add windows, since the
>> simulations are still all independent of one another.
>>
>
> My concern is that since the domain movement from initial to final
> conformation is of about 10 nm, the number of windows required maybe too
> large. I read that the windows need to be spaced close enough so that each
> one samples some part of the next window. My system is also very large
> (200K atoms), so it seems to be computationally very expensive.
>

The number of windows needed is largely a function of the force constant, e.g. 
how strong the biasing potential is.  But that's also a function of how strong 
the interactions are that are intrinsic to the system.

> I was just now thinking to reduce the problem to  conducting umbrella
> sampling for each conformation, simply using COM distance of the two
> domains as the reaction coordinate to obtain the PMF in each case. From the
> PMF, the binding energy of the domains in each conformation can be obtain.
> This will allow me to circumvent the complications introduced by domain
> rotations and maybe reduce the number of simulations required a bit . Will
> this be a correct approach?
>

Possibly, but COM distance can be a degenerate measure in the case of rotation. 
  The PLUMED plugin may be useful here rather than standard Gromacs options, 
otherwise try to find an example in the literature and follow a known procedure.

-Justin

-- 
==================================================

Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul

==================================================