Re: Using ABF to explore the conformational space of a spin label attached to a membrane protein

From: JťrŰme Hťnin (
Date: Mon Jan 17 2011 - 11:44:24 CST

Hi Ajasja,

On 17 January 2011 16:56, Ajasja Ljubetińć <> wrote:
> Dear J√©r√īme,
> I have mediated upon your advice:
>>The ideal situation would be for you to find a 2 or 3D space that might be
>>a good enough descriptor. For a flexible chain, I would start with
>>some RMSD and radius of gyration.
> And I think I have found two or three colvars that would be the most
> relevant.
> Those would be the polar coordinates (theta, phi, r) of the Ca-O1
> vector. (Where O1 is the last atom in the side-chain of my spin-label).

Two questions on this:
1) in what frame of reference do you need these to be expressed? I
assume it would be a local frame, based on nearby atoms? You'll
probably need to enable rotateReference, and related options.
2) if polar coordinates are okay, wouldn't Cartesian coordinates work
just as well? It would be a lot simpler...

> I am aware that using these components can only be used  with metadynamics.
> Don't these components (for example angle)
> fulfill the conditions of colvars for use in ABF? Is it just
> the availability of system forces or some orthogonality reason?

For the angle, it is just a matter of doing some math and finding the
Jacobian term in a usable form. I never needed it badly enough to do
it. If you feel like trying and send me your result, I will gladly
implement it.

> I could also just sample along all three projections of the vector and
> convert this to polar coordinates...

Indeed, as a post-processing. It sounds like the simplest solution by far.

> PS: I have noticed that the tilt and spinAngle components are not documented
> in the NAMD user guide 2.7.(they are just listed as available without a
> detailed explanation).

True, these are still considered experimental, but we should document
them at some point. The "tilt" component might be of some use in this
case, but if you define the other polar coordinates as you said, then
the "tilt" will only give you an approximate value of the missing


> On Wed, Dec 8, 2010 at 23:16, J√©r√īme H√©nin <> wrote:
>> Dear Ajasja,
>> On 8 December 2010 17:31, Ajasja Ljubetińć <>
>> wrote:
>> > Dear all,
>> > I would like to explore the conformational space (ie the free energy
>> > landscape) of the side-chain of a spin label (MTSSL) in water and in the
>> > membrane. The side chain has five dihedral angles (chi1 to chi5). I'm
>> > thinking of using ABF for this purpose, but the following questions
>> > arise:
>> >
>> > What would be a sensible value for the fullSamples value? (Since lipid
>> > tails
>> > may relax slowly, probably something in the range of 500-1000).
>> That is, I think, a point of detail compared to the other questions.
>> The key question is choosing the right reaction coordinate space, as
>> discussed below.
>> > The first dihedral angle is composed of two backbone atoms. If I
>> > restrain
>> > the backbone, does the first chi angle still count
>> > as unconstrained (does it
>> > fulfill the requirements for a well defined ABF colvar)?
>> Yes. Restraints do not interfere with ABF (even if they are called
>> "constraints" in NAMD for historical reasons).
>> > Would metadynamics be a better choice? What would be a good hillHeight?
>> > (about 1/tenth of the highest torsional energy barrier).
>> Metadynamics might be a better choice for a high-dimensional
>> landscape. One classic metadynamics study computed a 6D map, although
>> there was not much hope of it being converged. That study went on with
>> a more quantitative one-dimensional calculation:
>> Dealing with the 6D PMF probably presented the author with interesting
>> technical hurdles, chief among which visualizing the data!
>> > And some more philosophical ones:
>> These are good. They are the right ones to start with :-)
>> > What happens in the conformations where the side chain overlaps the
>> > backbone? The free energy there is really high due to VDW contacts. Does
>> > ABF
>> > push the sidechain through the backbone?
>> I would not expect that to happen. In practice, because of finite grid
>> resolution, ABF biases are rarely steep enough to push atoms through
>> walls, especially in multidimensional simulations, where alternate
>> pathways around the obstacle are available.
>> > Is it even feasible to explore a 5D space using ABF? I'm thinking of
>> > using a
>> > bin width of either 15 or 30  degrees (~250k or 8M bins).
>> I would recommend against it. On top of the impracticality of handling
>> the 5D data, getting adequate sampling would be a nightmare. ABF is
>> not very resilient to poor sampling, probably less so than
>> metadynamics, which offers the option of a quick non-equilibrium
>> exploration of a large landscape.
>> The question you should consider is: do you absolutely need to work in
>> this 5D space? Are all these degrees of freedom equally significant?
>> Is the chain fully flexible? Are there intrinsic conformational
>> restrictions? Do you have prior knowledge of key conformations? The
>> ideal situation would be for you to find a 2 or 3D space that might be
>> a good enough descriptor. For a flexible chain, I would start with
>> some RMSD and radius of gyration. You can first experiment with very
>> quick and inexpensive vacuum simulations, just to get a better idea of
>> what coordinates will make the molecule explore its conformational
>> space.
>> Best,
>> Jerome

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:56:32 CST