Re: [IMP-users] getting the MCCGsampler to work

[Ben Webb <>]

On 7/11/14, 5:12 AM, Josh Bullock wrote:
> So I have the same overall problem as before - creating an ensemble of a
> 4-subunit complex using MSconnectivity restraints. Having visualised the
> output (via RMF - thanks Barak :¬) , it's clear that no matter how many
> steps of CG or MC I put, the models do not change from their initial
> random placement. I know that the restraints are present because the
> models are evaluated and scored appropriately.

A sampler that ignores the scoring function would be pretty useless! 
Obviously MCCG doesn't do that. So my guess would be that you didn't add 
the restraints to the scoring function, the thresholds are all wrong (so 
that no MC moves are ever accepted), you have terrible starting 
conditions (so that the optimizer can never improve the score) or you 
have a poor choice of MC move set. I don't see you actually adding any 
Movers to your sampler in the script, so I'd suspect the last one. The 
underlying MonteCarlo class keeps some statistics (e.g. number of 
accepted/rejected moves) which might shed some light on what's going on.

> So I saw on an old (2011) nup84 example that MCCG can't handle rigid
> bodies, is this still the case ?

No. You just need to add a suitable RigidBodyMover.

> If so, should I switch to the DOMINO
> sampler ?

DOMINO enumerates in a discrete space, so is a rather different kind of 
optimizer. It is probably not the best choice in your case.

	Ben
--
                      http://salilab.org/~ben/
"It is a capital mistake to theorize before one has data."
	- Sir Arthur Conan Doyle