| With the advent of structural genomics and the rapid growth of the protein structure database, computational studies of protein structures have grown increasingly important as a means to investigate the relationship between structure and function. This dissertation contributes to the study of protein structure and function by presenting computational models for the comparison and classification of protein structures and the analysis of protein-ligand interactions.; We present two different algorithms for quantitatively comparing protein structures by computing an optimal correspondence between atomic coordinates from two structures. Our first technique, LOCK, computes accurate structural alignments by using secondary structure vectors to generate an initial alignment and then performing a detailed alignment of core C-α atoms. Our second technique, 3dSearch, is optimized for speed and computes approximate secondary structure alignments using geometric hashing. Our evaluation results show that LOCK is able to identify global and local structural similarities with high sensitivity and specificity and is an order of magnitude faster than other C-α level alignment techniques.; We use the alignment scores generated by LOCK to hierarchically classify a representative subset of protein structures. Our classification scheme, PClass, improves on other automated structural classifications by generating annotations at each internal node of the tree. Our evaluation results show that PClass is able to capture approximately 80% of the structural relationships in the scop tree.; We present a novel motion planning approach to efficiently model flexible ligand interactions at the catalytic site of a protein. Our roadmap-based motion planning model examines the distribution of all energetically favorable paths to the catalytic site and computes a weight for each path. We show that our model is able to detect the presence of a distinct energy barrier around the binding site and that this barrier can be used to distinguish the catalytic site from other potential binding sites. We also present a comprehensive evaluation system for empirically measuring the accuracy of roadmap-based models of molecular motion. We use this system to perform a series of statistical experiments to study the effect of seven specific roadmap parameters on the accuracy of the roadmap. |
