On Sep 20, 2011, at 7:50 AM, Benjamin SCHWARZ wrote:
Thanks for your answer Daniel,
Till now I just set the bbox so that it would fit approximately twice the expected diameter of the whole complex, but your explanation is appealing, this parameter too might influence the result... I'll have a look at that.
You can also more directly influence by adding DiameterRestraints for the proteins.
By the way, trying to investigate the effect of the various parameters, I was wondering if there was any fancy debug mode that would store every intermediate configurations corresponding to the intermediate states occurring between the various MC and CG steps of the sampler procedure ? If not, I think I could just rewrite MonteCarloWithLocalOptimization::do_step() to insert intermediate configurations in a ConfigurationSet…
There are various such OptimizerStates that log every accepted step along the way. If you want to look at every evaluated step, you can use a misc.StateAdaptor to stick in OptimizerStates as ScoreStates (so they get called every time evaluate is called).
My guess is that the molecules end up straight because they are initially randomized within some larger box, the connectivity restraint brings them together away from the other molecules and then they are pulled to the other molecules to satisfy their one restraint connecting them to other things. The pulling only occurs on one bead of the molecule, so the whole thing gets kind of strung out. Once that restraint is satisfied, all the scores are 0 and so the sampler decides things are good enough.
Hi guys,
I am currently interested in the coarse modeling of a small (10 subunits) system for which I have nothing but a few interactomics data as well as some (parts of) structures to help me to assess the size and number of beads to represent some of the subunits (details on my OS as well as on the modeled system can be found below.).
Thanks to the nup84 example, I was rapidly able to generate a large number of configurations and to visualize it (thanks for the MCCG sampler and the IMP.display.PymolWriter). I am however puzzled by the look of the obtained configurations : having represented my subunits as a succession of beads (I explicitly defined my restraints between successive balls), I expected my subunits to accommodate various types of shapes whereas they seem to have a tendency to adopt preferentially a straight aligned rod-like shape ( An example of a pymol file containing 1000 configurations can be downloaded for visualization at alnitak.u-strasbg.fr/~schwarz/Perso/1000-configs.pym).
I am now wondering if these configurations are really random generated, or if -as I fear- I have a bias in my sampling scheme, and how I can circumvent the trouble.
Here is one explanations I could find to this apparent behavior : I don't know how the excluded volume is handled by the MCCG sampler, but my guess is that the CG step helps a lot in "locally" optimizing this particular restraint. Now, if all balls are considered in the computation of derivatives (and not only those that really present a common intersection), it is intuitively quite probable that the system will ultimately push successive balls the furthest away from each other, hence favoring rod-like shapes upon epileptic-snake-like ones. Does this assumption makes any sense ?
I am running IMP kernel SVN 10177 with Boost.FileSystem on Mac OS
My system is composed of
- 10 subunits (divided in 30 beads to represent specific domains)
- 29 restraints to enforce pair connectivity between domains of a same subunits, or domains of different subunits for which a connectivity is known
- 2 ambiguous connections between several domains or subunits
- 1 excluded volume restraint between every 30 particles in the system.
--Ben (S)
_______________________________________________ IMP-users mailing list "> https://salilab.org/mailman/listinfo/imp-users