IMP-users August 2010

imp-users@salilab.org

4 participants
5 discussions

problems with a sample script (basic IMP optimization)
by Benjamin SCHWARZ 12 Aug '10

12 Aug '10

Hi list, As a first step towards a serious usage of IMP, I tried to write a sample code using rigid bodies, restraints and optimizers. With MonteCarlo optimizer, at the end of optimization the model seems to remains unchanged. With ConjugateGradients or SteepestDescent, when using too many steps in the optimizer (under 100 though !?) the script generally crashes with one of these error messages : File "/usr/local/lib/python2.6/site-packages/IMP/core/__init__.py", line 642, in optimize def optimize(self, *args): return _IMP_core.ConjugateGradients_optimize(self, *args) _IMP.UsageException: Attempting to construct a rotation from a non-quaternion value. The coefficient vector must have a length of 1. Got: 0.165858 -0.916284 0.55176 0.327904 gives 1.27905 File "/usr/local/lib/python2.6/site-packages/IMP/core/__init__.py", line 642, in optimize def optimize(self, *args): return _IMP_core.ConjugateGradients_optimize(self, *args) _IMP.UsageException: Attempting to construct a rotation from a non-quaternion value. The coefficient vector must have a length of 1. Got: 0.389486 -0.600546 0.553302 0.99031 gives 1.79921 Moreover when the script finishes, its outputs seem not relevant at all, though the score is better (smaller). maybe it is normal though, due to the choice of my restraints or to the lack of optimization steps… I think my understanding of IMP usage is still very far from perfect, will you help me sorting that out by having a look at my code ? You will fin the script attached to this message, its outlet is really basic : - a structure is loaded ( I choose 1RDT, a complex comprising two protein domains and two peptides) - a map is sampled at 10A for that structure - chains are extracted from the structure, rigidified and randomly moved in space - two restraints are added to the model in order to tale into account - the fit of all atoms to the sampled map - no steric clashes between rigid bodies - optimization thank you very much Dr. Benjamin SCHWARZ Biocomputing group Email : schwarz(a)igbmc.fr Voice : +33 (0)3 68 85 47 30 FAX : +33 (0)3 68 85 47 18 Structural Biology & Genomics Dept. - IGBMC 1 rue Laurent Fries BP 10142 F - 67404 Illkirch CEDEX

4 14

ExcludedVolumeRestraint : meaning of the score
by Benjamin SCHWARZ 09 Aug '10

09 Aug '10

Hi list, I was wondering about the exact meaning of the ExcludedVolumeRestraint score, and how this score was computed. So I took my virtual shovel and had a few hours digging in IMP code, but I still have questions. For what I understood, the score is computed as the sum of harmonic functions over a set of pairs of intersecting spheres (this score being 0.0 for non interacting spheres), but the way these pairs are built is still a little fuzzy for me. Here is the sample python code I wrote using that restraint : rigidChains = {} for chain in IMP.atom.Chains(IMP.atom.get_by_type(mh, IMP.atom.CHAIN_TYPE)) : rigidChains[chain.get_id()] = IMP.atom.setup_as_rigid_body(chain) ... lsc = IMP.container.ListSingletonContainer() for rb in rigidChains.values() : lsc.add_particle(rb.get_particle()) m.add_restraint( IMP.core.ExcludedVolumeRestraint( lsc, IMP.core.LeavesRefiner(refChains.values()[0].get_traits()) ) ) I intuitively expected IMP to consider all pairs of intersecting atom-spheres between any two different chains, but reading the code I could not convince me this is actually what is computed. I could not decide for instance if pairs of atoms in a same chain were considered or not, and in fact I could not even see where exactly atoms were extracted out of my rigid body hierarchies… Could you explain me in a few words if my sample code does what I expect from it, and if not, how to change it to obtain the desired behavior. Thanks a lot Dr. Benjamin SCHWARZ Biocomputing group Email : schwarz(a)igbmc.fr Voice : +33 (0)3 68 85 47 30 FAX : +33 (0)3 68 85 47 18 Structural Biology & Genomics Dept. - IGBMC 1 rue Laurent Fries BP 10142 F - 67404 Illkirch CEDEX

2 5

A couple of naïve questions concerning IMP.Model
by Benjamin SCHWARZ 06 Aug '10

06 Aug '10

Sometimes I load particles in a model that won't be used in the restraints (e.g. molecules loaded twice in order to have a reference for RMSD computations between an initial and an optimized state). I am wondering about the incidence of inserting such "non-useful" particles, is there any overhead ? Also, sometimes the Optimizer appears to need one to provide explicit movements for the particles (RigidBodyMover in the case of Monte Carlo optimizer), and sometimes it appears to know which particles are to be moved (CG). In this latter case, I guess particles that are involved in a Restraint are somehow tagged as such, and the optimizer do not bother with particles that are not tagged as "restraint involved". Am I correct ? Thanks --Ben

2 1

about rotations
by Benjamin SCHWARZ 06 Aug '10

06 Aug '10

Hello list, The information is probably mentioned somewhere but I happened to miss it, so just to be sure can you please confirm that : 1.) IMP rotations (Rotation3) are always origin centered 2.) a RigidBody's transformation is composed of a rotation (about the body center in internal coordinates) and a translation from origin to body center; thus body.get_coordinates() always provides the coordinates of the rigid body mass center. Thank you --Ben

3 2

basic map handling
by Benjamin SCHWARZ 03 Aug '10

03 Aug '10

Hi list, I could not find how to Generate a density map from a structure with a *fixed* dimension : When using IMP.em.SampledDensityMap(ps,resolution,apix) with a fixed size for apix, the grid size depends on the resolution. I would like to write a function that would allow me to generate maps with the same number of vertices (and the same space coverage) for different resolutions. I thus tried : def pdbFile2dmap (pdbStream,densityFile,resolution,apix,sel,bboxCoverage=1.2): ''' pdbStream : where to read the structure densityFile : name of the file where to drop the density map resolution : resolution of the density map apix : voxel size sel : selector for reading structure bboxCoverage : approximate bbox coverage of the density map ''' m= IMP.Model() # read protein mh = IMP.atom.read_pdb(pdbStream,m,sel) # add radius info to each atom IMP.atom.add_radii(mh) # compute bbox, and map size in voxels bbox = IMP.atom.get_bounding_box(mh) diag = bbox.get_corner(0) - bbox.get_corner(1) nx = int(bboxCoverage * diag[0] / apix) ny = int(bboxCoverage * diag[1] / apix) nz = int(bboxCoverage * diag[2] / apix) # compute origin origin = bbox.get_corner(0) + (1-bboxCoverage)/2 * diag dmap = IMP.em.SampledDensityMap() # dmap.CreateVoidMap(nx,ny,nz) ps = IMP.Particles(IMP.core.get_leaves(mh)) dmap.set_particles(ps) dmap.set_origin(origin) dmap.get_header_writable().set_number_of_voxels(nx,ny,nz) dmap.get_header_writable().set_resolution(resolution) dmap.update_voxel_size(apix) dmap.resample() dmap.calcRMS() # computes statistic stuff about the map and insert it in the header print dmap.get_header().show(),"\n" IMP.em.write_map(dmap,densityFile,IMP.em.MRCReaderWriter()) that resulted in : python(10802) malloc: *** mmap(size=18446744073709314048) failed (error code=12) *** error: can't allocate region *** set a breakpoint in malloc_error_break to debug Traceback (most recent call last): File "./pdb2density.py", line 181, in <module> pdbFile2dmap(pdbStream,densityFile,resolution,apix,sel,1.2) File "./pdb2density.py", line 75, in pdbFile2dmap dmap.update_voxel_size(apix) File "/usr/local/lib/python2.6/site-packages/IMP/em/__init__.py", line 428, in update_voxel_size def update_voxel_size(self, *args): return _IMP_em.DensityMap_update_voxel_size(self, *args) MemoryError: std::bad_alloc Can you tell me if my approach is OK, and how to solve my allocation problem ? I would also like to know if there are easy means in IMP to 1.) Strip a density map : out of an existing map, create a new map or restrict the existing one, so as to conserve only a central subset of voxels 2.) Re-sample a map : Conserve the density map "box" size, and change the number of voxels that cover it --Ben

3 7

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

IMP-users August 2010