Novel computational methods for drug design

The advances and applications of computational methods in drug design have made a tremendous impact on the pharmaceutical industry. Modern approaches to computer-aided molecular design fall into two general categories. The first includes ligand-based methods which characterize structural or physico-chemical properties of ligand molecules. The derivation of empirical quantitative structure-activity relationships (QSAR) furnishes a theoretical basis for future lead optimization. The second includes structure-based methods which utilize the three-dimensional (3D) structure of the ligand-bound receptor. Many innovative algorithms have been implemented to construct de novo ligands that fit the receptor binding site in a complementary manner.;The first part of the thesis is focused on ligand-based design methods. In Chapter two, we introduce a new QSAR tool, the genetic neural network (GNN), for analyzing high dimensional data sets. GNN utilizes a genetic algorithm to select descriptors and a neural network to correlate the chosen descriptors with activity. In Chapter three, the GNN method is applied to a set of anxiolytic benzodiazepines (BZ), where the results show significant improvements over previous benchmarks. Its application in pharmacological design is illustrated by a minimum perturbation approach to predict new BZ candidates. Chapters four and five describe the utility of GNN to obtain 3D QSAR from molecular similarity matrices. Chapter four emphasizes the new approach and its validation based on a standard data set. Chapter five reports its application to eight additional data series. The results indicate that the method is a general and efficient way to obtain predictive structure-activity relationships for a broad range of chemical classes.;The second part of the thesis is devoted to structure-based ligand design. Chapter six describes the design and implementation of a hybrid prediction system to predict the binding affinity of a ligand to glycogen phosphorylase (GP), a regulatory enzyme for glycogen degradation and a potential therapeutic target for diabetes. Chapter seven reports the utility of this system to characterize novel GP inhibitors that have been constructed using an array of computational ligand design strategies. The role of computational methods in the overall strategy of drug development is discussed in the concluding chapter.