From: Jan Saam (saam_at_charite.de)
Date: Tue May 30 2006 - 11:01:29 CDT
Dear Jerome, dear other prospective ABF users,
the discussion is getting quite interesting.
here's what you'll experience playing with the example of unfolding a
decalanin helix accompanied with my explanation:
Jerome Henin wrote:
>>thanks for the clarification, I got the example working now!
>Just for the record, what do you think made your simulation work, among the
>points Chris and I mentioned? I am sure that other users are likely to run
>into the same problems, so we'd better know in advance.
The reaction coodinate (RC) is chosen to be the distance between the
first and the last C-alpha carbon. The range of interest is defined
between 12 (shorter that the helix) and 32 (extended chain) Angstrom.
The sampling wil first look very non-uniform since the biasing force
depends on the (local) estimate of the free energy rpofile along the RC.
This estimate slowly builds up with the number of samples in each bin.
Since the first estimates are based on very few samples the adaptive
biasing force as meaningless as the PMF estimate itself. Thus it is
recommended that you use the parameter *fullSamples *to specify the
number of samples that are taken in each bin before the ABF is switched
on for this bin. Suppose you start in a PMF minimum. The time that is
needed to escape the minimum depends on the slope of the walls. The
steeper the walls the more samples are needed to build up a proper
estimate that allows for compensation of the barrier. Hence the system
will spend quite some time sampling the minimum until it suddenly
escapes. Transition states will be sampled less since the system leaves
them quickly again.
It is important to note that uniform sampling is only achieved after
there is an estimate for the PMF of the complete RC!
So just be patient and wait until the system escapes from the current
minimum. It finally will, even if it first has to accrue a very large
number of samples in these bins.
For the decalanin example the number of samples in the initially
populated bins before the system evolves to new states is several
thousend per bin.
The example from Jerome's and Chris' paper was done with an unterminated
helix. I also tried decalanin with regular C and N-termini. Because the
opposite charges attract each other it is much harder to pull the ends
apart which results in a much larger free energy difference between the
folded and extended state (~80 vs. ~30 kcal/mol). Moreover you'll need a
lot more samples before the ABF will be able to move the ends apart.
>>May I bug you with one more question?
>>I'm studying the diffusion of oxygen along a channel in a protein. The
>>channel is not straight but follows despite some kinks roughly along a
>>certain direction from the active center to the protein surface. There
>>are also some "bays" or dead end roads along the tunnel that can be
>>explored by oxygen.
>>Thus, most of the time the gas molecule will not travel exactly along
>>the defined reaction coordinate.
>>Does that mean the following?:
>>a) If distance active center -- O2 is the RC:
>>Oxygen could be temorarily trapped in a dead end road, its hammering
>>against the wall and the adapdive biasing force is cranked up until
>>oxygen breaks through that wall? And even when it's travelling back,
>>finding the right way, the time spent in the side channel will distort
>That is, to a certain extent, true. In such cases where the pathway is roughly
>described by the RC, but some details are not resolved, it usually boils down
>to a problem of timescales. If oxygen is able to diffuse fast enough between
>the main route and small dead ends surrounding it, then it will find its wasy
>and the PMF should not be distorted, because the time spent on the right
>pathway will dominate the average. The freedom to move back and forth along
>the way is crucial there. In an SMD simulation, it would be extremely hard
>for a system to return to the right track.
>If on the contrary, diffusion is a bit too slow, then yes, the system may
>become trapped in the dead end, and the adaptive bias will then grow to
>unrealistic values to try to pull oxygen through the wall.
>One parameter that may determine which of the two cases prevails is
>fullSamples. You want the system to find the right track *before* the bias is
>fully applied. That's where things become really system-dependent.
>>b) If abscissa along main direction is used as RC:
>>When I'm defining a direction of the RC there will always be a certain
>>angle (say between 0 and 50 deg.) between the real channel V and the RC
>>due to the crookedness of the channel. I.e. the adaptive biasing force
>>is pushing with that angle against the channel wall, which will also
>>distort my PMF. Correct?
>I would say, no, at least not if the simulation is in near-equilibrium. Again,
>fullSamples will play a role on that. What matters is that there is a
>one-to-one mapping between your RC and the actual progression of the system
>along the pathway, that's enough to makes the RC relevant. And that's also
>why alternate pathways or dead-ends are very hard to take into account.
>Another problem with this, though, is to define an absolute direction. Is your
>protein able to tumble or rotate in some way?
>>c) If I'm partitioning the channel into several shorter, almost straight
>>portions, I can stitch the PMF together and probably improve on these
>>problems. But I'll have to know the RC very well before and it is not
>>easy to define the directions in a system where everything moves.
>I would say the only useful coordinates would then be defined relative to the
>protein itself. The available RCs are probably not adapted for that, but we
>can talk about what would suit your needs. It would probably be helpful to
>many other people, too.
-- --------------------------- Jan Saam Institute of Biochemistry Charite Berlin Monbijoustr. 2 10117 Berlin Germany +49 30 450-528-446 saam_at_charite.de
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:42:06 CST