EVALUATION OF AUTOMATED METHODS FOR THE SELECTION OF MODELS FOR SIMULATION OF C-13 NUCLEAR-MAGNETIC-RESONANCE SPECTRA OF KETO STEROIDS

Computer-aided methods for determining the suitability of stored predictive chemical shift models for use in simulating C-13 NMR spectra were characterized and further evaluated. These model selection techniques were applied by using a newly constructed spectral simulation data base containing 55 stored equations developed for a broad assortment of compounds examined in previous spectral simulation studies. Appropriate models were selected for the chemical shift predictions for a set of four keto steroid molecules. Highly accurate simulated spectra were generated by using these equations, demonstrating the utility of the current model selection approach. New predictive models, incorporating several newly developed structural descriptors, were also generated for a collection of 24 keto steroids. Models developed specifically for keto steroids, in general, outperformed selected models, created originally for non-keto steroids, but did not allow for complete spectral simulations in all cases. For many carbon atoms, the chemical shift predictions were more accurate when selected models were used instead of equations developed specifically for keto steroids. Alternative means of characterizing atomic structural similarity were investigated for use in enhancing model selection methods. The results of these experiments highlighted limitations regarding the model selection process and suggested new methods for use in future work.