AUTOMATED SELECTION OF REGRESSION-MODELS USING NEURAL NETWORKS FOR C-13 NMR SPECTRAL PREDICTIONS

A model selection method utilizing neural networks has been developed to perform automated spectral predictions using a library of previously generated regression equations (models). The library contains models capable of simulating the C-13 NMR spectra for various classes of organic compounds. The 4018 carbon atoms used to develop the 75 models were utilized to train the network to relate the chemical environment surrounding each of the atoms to the models which they were used to develop. A neural network was trained to correctly select models for 98.2% of the carbon atoms in the training set and 95.5% of those in the cross-validation and test sets. This trained network selected models for a 30 compound external prediction set and simulated their spectra with a mean RMS spectral error of 1.11 ppm.