Computational methods for structure-based drug design and macromolecular crystallization

Large data sets are common in crystallography and in structure-based drug design, and this thesis presents computational methods useful for recording and understanding them. For example, determining conditions suitable for crystallizing macromolecules often involves many iterative trials. The current standard procedure of manually designing experiments and recording observations is tedious, is susceptible to errors, makes organization of records and searches for patterns difficult, and is unsuited to automation. We have therefore written Etray, an open-source program obtainable over the Web, to automate design and to aid analysis of experiments. An example of the use of Etray is presented.;Structure-based drug design increasingly relies on discovering patterns or anomalies in ensembles of ligands or functional groups whose properties are difficult to describe when individually investigated. A general purpose molecular visualization program, EVAn, which incorporates visual data mining features has been developed to facilitate interpretation of this data, and its applications are presented. In one application of EVAn, Multiple Copy Simultaneous Search (MCSS) minima calculated on the force field projected by the Delta Antigen (DAg) were analyzed and used to design D-peptides to bind to the target.;We report the development of a novel class of synthetically accessible small molecules with few degrees of freedom called Spiders, which can present chemical moieties in the same orientations as they appear on an a helix. A computational method is presented for determining the optimal side chains and binding orientation for a Spider in the force field of a particular target. The method consists of an iterative algorithm that uses multiple independent simulated annealing calculations on a candidate ligand docked to the target protein for two purposes. First, the simulations determine a low energy binding location and conformation. Second, the data are used to optimize the sequence, based on the hypothesis that candidate side chains whose positions vary the most in the low energy minima are the least likely to make significant energetic contributions. The automated simulated annealing algorithm outperformed Dead-end elimination [Desmet et al., Nature, 1992] at guiding the choice of side chains to be used.