Prediction of enzyme inhibition and receptor antagonist properties from molecular structure, and, Development of radial basis function neural networks for the analysis of inhibitor binding.

摘要

Two areas of computational chemistry are described in this thesis.; Methodology involved in QSAR is presented. Numerical descriptors are used to encode molecular structures. Specifically, development of classification algorithms using radial basis function neural networks is presented.; The first QSAR application involves the prediction of IC50 values for acyl-CoA: cholesterol O-acyltransferase (ACAT) inhibitors derived from N-chlorosulfonyl isocyanate. A CNN model is developed using eight descriptors that provides a root-mean-square error (RMSE) of 0.242 log units for an external prediction set. 27 exclusion compounds were predicted on the basis of the best model available, the CNN model.; The second QSAR application involves prediction of inhibitory concentration for inhibitors of Glycine site of N-methyl-D-aspartate (Glycine/NMDA) receptor antagonist. Predictive models are generated for varied set of hydroxyquinolins using multiple linear regression, and CNN techniques. A twelve-descriptor CNN model is developed for prediction of inhibitory concentrations for inhibitors of Glycine/NMDA antagonist that produces RMSE of 0.776 log units for compounds in the external prediction set.; The second part of this thesis presents work where k-nearest neighbors analysis, linear descriminant analysis and radial basis function neural networks are used to generate models to classify inhibitors of Protein Tyrosine Phosphatase 1B. Two types of models are generated: one type to classify compounds as inactive, moderately active, and active (three-class problem), and one type to classify compounds as inactive or active without considering the moderately active class (two-class problem).; The second classification application involves classification of HIV protease inhibitors on the basis of their antiviral potency. The effect of using majority of predictions was tested for the radial basis function neural network classifier, which led to significant increase in the classification rate for training and cross validation set however the external prediction set remained the same.; In addition, the newly developed RBFNN classifier was used to model the toxicity data of HIV Protease inhibitors. (Abstract shortened by UMI.)