April 2011 - IMP-dev - salilab.org

Anchoring part of a model
by Javier Velazquez 28 Apr '11

28 Apr '11

I would like to anchor a component of a complex while the others are optimized. With the MonteCarlo sampler is easy, just not adding a mover to the component. But ConjugateGradients moves the position anyway. Is there any way to nicely do this, or do I have to go to each particle of the component and set_is_optimized as False for all the x,y,z keys? -- Javier Velazquez Postdoc at Salilab, UCSF 1700 4th st. Byers Hall, office 503. 94158 San Francsico

2 1

Recovering of restraints from a model
by Javier Velazquez 27 Apr '11

27 Apr '11

Recovering restraints (after optimization) from a model in python has an unexpected behaviour. This code fails: for i in range(model.get_number_of_restraints()): r = model.get_restraint(i) print "checking restraint ...",r.get_name() pairs = r.get_connected_pairs() All the restraints in the model are connectivity restraints, but I obtain: raise AttributeError(name) AttributeError: get_connected_pairs Same thing if the restraint is EM2DRestraint when accessing non-inherited members. Is there another way to access the restraints when the optimization has finished? -- Javier Velazquez Postdoc at Salilab, UCSF 1700 4th st. Byers Hall, office 503

2 1

IMP on the cluster
by Daniel Russel 25 Apr '11

25 Apr '11

I have built various libraries that are helpful for running IMP on the cluster and put them in my home directly there (~drussel/production). The libraries available are - ~drussel/production/hdf5: the version of hdf5 installed on the cluster is old and has several important differences from more recent versions. Backporting the hdf5 code looks to be a bit painful and messy and hdf5 is very easy to build. - ~drussel/production/bullet: the physics based optimizer, it is not installed in the cluster - ~drussel/production/CGAL: the geometry library, it is not installed in the cluster If you want to use any of those, I suggest you copy them to your own directory somewhere and then add that directory to you imp config paths (libpath, includepath and ldlibpath). That way you won't have problems if I mess with my copies. None of the libraries have data and so they all work fine when relocated. Ultimately, the nicest solution might be to have IMP include them or fetch them when needed itself. But I don't feel like setting that up now.

2 7

recently in IMP
by Daniel Russel 25 Apr '11

25 Apr '11

- IMP.hdf5 was renamed to IMP.rmf, the proposed name of the file format that it supports. That also better separates the hdf5 support from the file format support. - The old I/O functions have been marked as deprecated in favor of IMP.hdf5. You may need to compile with deprecated="yes" if you use them. - The undocumented show_dotty() method on python graphs has been renamed to the still undocumented show_with_altgraph(), but at least it is clearly different from show_graphviz(). - The python graph interface now has a remove_vertex method. - One can now access the model restraint evaluation statistics programmatically using IMP::Model::get_restaint_statistics(). Restraint statistics can be logged to a file during optimization using IMP::core::WriteRestraintScoreStatisticsOptimizerState. - IMP::domino::DominoSampler now allows you to specify the tree on which merges occur directly. Also, you now need to specify the merge tree rather than the JunctionTree if you want to do non-loopy inference and specify something about how it is done. Sorry, it makes the code a lot easier. - Preliminary symmetry support has been added via the IMP::core::TransformationSymmetry and IMP::core::TransformationAndReflectionSymmetry classes.

1 0

IMP I/O
by Daniel Russel 21 Apr '11

21 Apr '11

As I'd mentioned before, I've been very happy with the I/O support provided in the IMP.hdf5 module and would like to simplify the various IMP I/O options around this. While I think the IMP.hdf5 support is generally better for almost every problem, the module does not provide direct one to one replacements for much of the existing functionality, so there may be usages which are more annoying with the new structure. So please try it out and provide feedback. To simplify things, I would propose - deprecating and then removing the kernel binary I/O. This would remove IMP.read_particles_binary, IMP.write_particles_binary, WriteBinaryOptimizerState, IMP.write_configuration_set and IMP.read_configuration_set. This then removes the use of NetCDF, dropping the number of dependencies by one. - deprecating and then removing the kernel text I/O. This would remove IMP.read_particles, IMP.write_particles. Particular areas where IMP.hdf5 differs from what existed previously include - there is nothing quite so space efficient as IMP.write_particles_binary. I haven't directly compared the two, but the hdf5 files are not too bad in terms of size - in contrast to IMP.write_particles_binary, you can't/don't have to specify which attributes are written, with IMP.hdf5 all are written. For now, please be careful about using the hdf5 files as an archive format for long term storage.

1 0

MinimumRestraint
by Pia Unverdorben 13 Apr '11

13 Apr '11

Dear all, I have a question concerning the MinimumRestraint Class. My problem is, that I don't know exactly how to handle it, so that I can use the modified assembly further, in my case for an optimization with the restraints still switched off. When I read in an assembly (model and restraints), evaluate it before and after I applied MinimumRestraints, I obtain the same value. But with MinimumRestraint.unprotected_evaluate it gives back a smaller score. Here is the specific part (inactivate_pct is the percentage of restraints I want to inactivate): E_ini=self.ass.evaluate(False) res=self.ass.restraints.restraint_sets['rigid_bodies'].get_restraints() l=len(res) num=int((1-inactivate_pct)*l) minres=IMP.core.MinimumRestraint(num, res, 'MinimumRestraint') minres.set_model(self.ass.model) E_minres=minres.unprotected_evaluate(IMP.DerivativeAccumulator(1.0)) E_end=self.ass.evaluate(False) So E_ini = E_end > E_minres. How can I access the changed model?? Thanks a lot in advance! Pia

2 2