AAAI Publications, Workshops at the Twenty-Seventh AAAI Conference on Artificial Intelligence

Font Size: 
Packing Models for Multi-Domain Biomolecular Structures in Crystals with P212121 Space-Group Symmetry
Yan Yan, Gregory S. Chirikjian

Last modified: 2013-06-29

Abstract


In the context of X-ray crystallography, the molecular replacement (MR) method is frequently used to obtain phase information for a crystallographic unit cell packed with amacromolecule of unknown conformation. This is important because an X-ray diffraction experiment on its owndoes not provide full structural information. The shape and symmetry of the unit cell is determined by the space groupsymmetry. The most common space group for biological macromolecules is P212121.The goal of MR searches is to place a homologous/similar molecule in the unitcell so as to maximize the correlation with X-ray diffraction data, and then to use the model toadd the unknown phase information to the experimental data. MR software packages typically perform rotationand translation searches separately. This works quite well for single-domain proteins that can be treatedas rigid bodies. However, for multi-domainstructures and complexes, computational requirements can become prohibitive and the desired peaks can becomehidden in a noisy landscape. The main contribution of our approach is that computationally expensive MR searches incontinuous configuration space are replaced by a search on a relatively small discrete set of candidate packingarrangements of a multi-rigid-body model. First, candidate arrangements are generated by collision detections on a coarse grid in the configuration space. This list of feasible arrangements is short because collision-free packing requirement together with unit cell symmetry and geometry impose strong constraints. After computing Patterson correlations of the collision-free arrangements, an even shorter list can be obtained using the 10 candidates with highest correlations. In numerical trials, we found that a candidate from the feasible set is usuallysimilar to the arrangement of the target structure within the unit cell. To further improve the accuracy, a rapidly-exploring random tree (RRT) can be applied in the neighborhood of this packing arrangement. Our approach is demonstrated with multi-domainmodels in silico for 3D, with ellipsoids representing both the domains of the model and targetstructures. Our results show that an approximate phase can be found with mean absolute error less than 5 degrees and search speed is effectively improved.

Full Text: PDF