Re: PMF with ABF for ion channels in a spherical molecule

From: Aron Broom (broomsday_at_gmail.com)
Date: Fri Jul 01 2016 - 18:42:21 CDT

Hi Olga,

For ABF everything will be done in a single simulation. You could also try
metadynamics to do the same thing. The process you are describing, of
having multiple different windows and different initial configurations is
essentially what I would normally call Umbrella Sampling. Fortunately, ABF
and metadynamics make things quite simple.

So in theory you could start with the ion anywhere and ABF would run,
generating a more and more accurate PMF the longer it went. As this
happens the ion itself will get forced back and forth along the channel.
In practice, however, some problems might occur.

The first as you suggested is that if you have other ions in the channel
then maybe the PMF will be altered, so you might have to keep them away,
I'm not sure offhand what the best way to do this would be, but I suspect
the solution is trivial.

The second problem, that is much less trivial, is that the ion could get
stuck to the side of the channel entrance, and ABF will keep trying to push
it along Z, estimating an unrealistically high energy barrier to entering
the channel (as it attempts to force the ion through the atoms that make up
your spherical thing). One solution to this would be to also bias along
another dimension, but this will make the simulation take exponentially
longer to estimate a reasonable PMF. The more typical solution to this
problem is to restrain the ion such that when it has left the channel on
either end, it can't float too far off to the side. This can be done using
a cylindrical restraint, which I think is called something like distanceXY
in the colvars module. This would restrain the two axes that are not Z to
be say at most 2 angstroms from 0, that way the ion is only able to explore
a 2 angstrom radius cylinder that extends along your Z axis. The actual
radius you choose should depend on the width of your channel. You want to
give the ion enough freedom to interact with all sides of the channel and
all important parts of the ends of the channel, but not enough that it can
wander off when it exits on either end. In the end there are no perfect
solutions to this last problem, but this one is at least workable, other
people might have different techniques they use to overcome issues like
this. Of course, if you only care about the PMF withing the channel, and
not how it leaves solution or becomes solvated on the other side, you could
restrain it and not worry about this, but it sounded from your previous
messages like this wasn't the case.

Aron

On Fri, Jul 1, 2016 at 4:50 PM, Olya Kravchenko <ovkrav_at_gmail.com> wrote:

> Hi Aron,
>
> thanks a lot, this is very helpful!
>
> I ran my simulation a few nanoseconds longer and saw ion exiting the
> channel, but only in cases where I already had one or two ions in the
> channel to start with and only after more ions entered the channel. It
> took ~14ns.
>
> Now I would like to do single ion PMF with Z as reaction coordinate
> and for that I am setting up another simulation where one of the
> channels is aligned with Z.
>
> Here is what I don't understand. For single ion PMF using ABF method,
> do I have to run a really long trajectory that shows ion passing all
> the way through the channel (with the purpose of using the snapshots
> to define subsequent ABF windows)? Or should I equilibrate my system,
> then physically place ion in the next subsequent window, equilibrate
> it, then use result as initial configuration for the next window and
> so on? I am not sure I understand the process correctly.
>
> And another thing, if I understand it correctly, there should be only
> one ion in the channel and no ions within certain distance from the
> ion that I follow. What if other ions approach during simulation? Is
> there a way to keep them at certain distance?
>
> Olga
>
>
>
>
>
>
>
> On Tue, Jun 28, 2016 at 5:37 PM, Aron Broom <broomsday_at_gmail.com> wrote:
> > Yes, the simple equilibrium simulation wouldn't be usable unless you had
> > complete transit, ideally many times, such that you could construct a
> > reliable histogram of the population density at each position along your
> > reaction coordinate.
> >
> > For ABF, yes this is certainly possible. If you just use the distance
> from
> > the center of the ferritin, you wouldn't need the orientation restraint,
> and
> > you could try the whole thing without any other restraints. You might
> also
> > think of trying meta-dynamics or well-tempered metadynamics, which are
> both
> > available within NAMD.
> >
> > In general, I would suggest the following two things to look out for:
> >
> > 1) Do the biasing simulation twice starting from some different initial
> > snapshot, and see how reproducible the PMF is.
> >
> > 2) If the energy values along the PMF seem extremely high (like energy
> > barriers of hundreds of kcal/mol) it is likely that the simulation needs
> to
> > be run for longer in order to better equilibrate (what that means in
> terms
> > of the parameters for a given biasing method is different). I think
> > realistically you will need to have at least 100 ns of biased simulation
> > time to get a reliable answer, and maybe more (maybe even much more).
> >
> > On Tue, Jun 28, 2016 at 5:20 PM, Olya Kravchenko <ovkrav_at_gmail.com>
> wrote:
> >>
> >> Hi Aron,
> >>
> >> thanks a lot! I want to do both: set up a new simulation to obtain PMF
> >> via ABF as well as analyze the one I already have. I just was not sure
> >> if it is at all usable for PMF, particularly because I do not observe
> >> ions leaving the channel (but I see progression through other binding
> >> sites, they just get stuck at the last one).
> >>
> >> Olga
> >>
> >>
> >>
> >>
> >>
> >> On Tue, Jun 28, 2016 at 4:57 PM, Aron Broom <broomsday_at_gmail.com>
> wrote:
> >> > Hi Olga,
> >> >
> >> > For the orientation restraint, the protein will rotate just because of
> >> > random thermal motion and interaction with the solvent.
> >> >
> >> > I had thought you were thinking of setting up some kind of biased
> >> > simulation, like umbrella sampling, adaptive bias force, or something
> >> > else.
> >> > But it sounds like you just want to do the analysis of an existing
> >> > simulation. In that case no restraints are needed.
> >> >
> >> > If you think that you have something useful in your 20 ns simulation
> and
> >> > you
> >> > are hoping to just analyze it along a single reaction coordinate to
> >> > generate
> >> > a PMF, then I think you could just use the distance from the ion to
> the
> >> > center of mass of the whole ferritin.
> >> >
> >> > I have never done ion transport simulations, but I would be surprised
> if
> >> > 20
> >> > ns was long enough to see enough transits to get good statistics for a
> >> > PMF.
> >> >
> >> > ~Aron
> >> >
> >> >
> >> > On Tue, Jun 28, 2016 at 4:35 PM, Olya Kravchenko <ovkrav_at_gmail.com>
> >> > wrote:
> >> >>
> >> >> Hi Aron,
> >> >>
> >> >> thank you for replying!
> >> >>
> >> >> My system is ferritin, it is a spherical molecule that is empty
> inside
> >> >> (such as soccer ball). The ions go inside through channels in the
> >> >> surface, there are 8 channels, each consists of three chains. I am
> >> >> trying to implement your suggestion now, i.e. align one of the
> >> >> channels with Z axis.
> >> >>
> >> >> For the harmonic restraint, what exactly should be restrained? I read
> >> >> the manual about orientation variable, but I don't quite understand
> >> >> what happens if I do not use it. The manual says it is needed to stop
> >> >> the rotation of the protein. Why would it rotate?
> >> >>
> >> >> Another thing: I recently ran a 20 ns simulation (I set it up without
> >> >> PMF in mind at the time), so I now have a trajectory and roughly
> >> >> identified ion pathway, but none of the channels are aligned with Z
> >> >> axis. Do I understand it correctly that I cannot use the results for
> >> >> obtaining PMF along the axis of the channel? Is there a conversion
> >> >> that would allow me to utilize the current trajectory? (this is why I
> >> >> thought about abscissa first).
> >> >>
> >> >> Thank you,
> >> >>
> >> >> Olga
> >> >>
> >> >>
> >> >>
> >> >>
> >> >>
> >> >>
> >> >> On Sun, Jun 26, 2016 at 4:13 PM, Aron Broom <broomsday_at_gmail.com>
> >> >> wrote:
> >> >> > I don't think I fully understand what your system looks like.
> >> >> >
> >> >> > But assuming it is some kind of continuous channel or pore, you can
> >> >> > use
> >> >> > the
> >> >> > DistanceZ collective variable, and align the channel along Z. Then
> >> >> > you
> >> >> > would also use the Orientation collective variable with a harmonic
> >> >> > restraint
> >> >> > to keep your spherical molecule's channel aligned with the Z-axis.
> >> >> >
> >> >> > In terms of having a longer distance, that would depend. If you
> only
> >> >> > cared
> >> >> > about being longer in one direction, you could define the reaction
> >> >> > coordinate as DistanceZ between the ion and the center of mass of
> >> >> > atoms
> >> >> > at
> >> >> > the channel entrance. If you wanted to get extra PMF information
> in
> >> >> > both
> >> >> > directions, then I think you would need to use the dummy atom
> >> >> > functionality
> >> >> > of the collective variable module to define a dummy atom at some
> >> >> > particular
> >> >> > Z coordinate that would be before the channel entrance.
> >> >> >
> >> >> > On Fri, Jun 24, 2016 at 6:01 PM, Olya Kravchenko <ovkrav_at_gmail.com
> >
> >> >> > wrote:
> >> >> >>
> >> >> >> Hi all,
> >> >> >>
> >> >> >> I am trying to figure out how to calculate PMF for ion channels
> in a
> >> >> >> spherical molecule. I would like to use position of the ion along
> >> >> >> the
> >> >> >> channel axis as the reaction coordinate. How does one define the
> >> >> >> reaction coordinate for a radial channel in a sphere?
> >> >> >>
> >> >> >> At this moment I can only think of defining it via abscissa and
> use
> >> >> >> centers of mass of the atoms at the entrance and exit of the
> >> >> >> channel.
> >> >> >> But then residues there move, would that affect the reaction
> >> >> >> coordinate? Also, if I want the distance to be longer than the
> >> >> >> length
> >> >> >> of the channel, how would I define that within abscissa?
> >> >> >>
> >> >> >> I would appreciate your input.
> >> >> >>
> >> >> >> Olga
> >> >> >>
> >> >> >
> >> >> >
> >> >> >
> >> >> > --
> >> >> > Aron Broom M.Sc
> >> >> > PhD Student
> >> >> > Department of Chemistry
> >> >> > University of Waterloo
> >> >
> >> >
> >> >
> >> >
> >> > --
> >> > Aron Broom M.Sc
> >> > PhD Student
> >> > Department of Chemistry
> >> > University of Waterloo
> >
> >
> >
> >
> > --
> > Aron Broom M.Sc
> > PhD Student
> > Department of Chemistry
> > University of Waterloo
>

-- 
Aron Broom M.Sc
PhD Student
Department of Chemistry
University of Waterloo

This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:22:18 CST