AUTOMATED MODEL SELECTION FOR THE SIMULATION OF C-13 NUCLEAR-MAGNETIC-RESONANCE SPECTRA OF CYCLOPENTANONES AND CYCLOHEPTANONES

Linear equations capable of simulating the carbon-13 nuclear magnetic resonance spectra of cyclopentanones and cycloheptanones are developed. The equations relate structural descriptors to C-13 NMR chemical Shifts. These descriptors encode the topological, geometrical, and electronic environment surrounding each carbon atom. Due to the conformational flexibility of these compounds, the significance of the geometrical descriptors, in particular, was evaluated. When used together, the models can simulate the chemical shifts of the 36-compound data set with an average spectral error of 1.12 ppm. The models are now included in a spectral simulation database containing 71 predictive equations. This study also involves enhancing the existing model database and the automated model selection procedure used to probe the database. The new model selection scheme determines model suitability based on a similarity measure between a query carbon and all of the carbon atoms used to develop the database models in previous studies. This enhancement dramatically improves the prediction results for an eight-compound external prediction set containing cyclopentanones and cycloheptanones.