Re: [IMP-users] IMP.pmi.restraints.em.GaussianEMRestraint vs IMP.em.EnvelopeFitRestraint

[Ben Webb <>]

On 10/14/16 2:00 PM, Julian wrote:
> Hi, I was wondering what the difference is between the different EM
> restraints.  The doc says EnvelopeFitRestraint may be faster for low-res
> maps but that's about it.

There have been many ways to score an EM fit in IMP over the years, but 
basically they boil down to calculating the cross correlation between 
the experimental map and the model map either by a double sum over the 
actual voxels, or by comparing some approximate representation (envelope 
or Gaussians) of the two maps. GaussianEMRestraint is the most recent 
iteration of the latter approach (IMP.em.FitRestraint uses the former) 
and is the one I'd recommend. EnvelopeFitRestraint is probably 
comparable to GaussianEMRestraint when the latter is used with a very 
small number of Gaussians.

> The only place I could find using it is the pde6 exampe.

pde6 is an actual published application, so was optimized for results, 
not for being an example. We keep it around because we feel that all of 
our published work should be reproducible. You're of course welcome to 
look through any of these applications, but some of them aren't that 
tidy, and use technology of the time, not the latest methods. At the 
time GaussianEMRestraint wasn't available, which is why 
EnvelopeFitRestraint was used there. I'll update the documentation to 
point this out.

> I guess I don't really
> understand what kind of particles are expected in
>
> IMP.em.EnvelopeFitRestraint(particles, map, 11.0, 5.0)

This is an atomic restraint, so the particles are expected to be atoms 
(technically they could be anything with XYZ coordinates, but the 
assumption is that they're not diffuse beads).

> Would this restraint work with the ReplicaExchange anyway?

I don't see why not, although you'd have to ensure that the restraint is 
only applied to your atomic representation, and if you wanted to use it 
in conjunction with PMI you'd probably find it more convenient to write 
a small PMI wrapper class for it.

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