PREDICTION OF GAS-CHROMATOGRAPHIC RETENTION INDEX DATA BY NEURAL NETWORKS

Neural networks using the backpropagation algorithm can be applied to quantitative structure-physical property relationship studies. Neural networks can be trained with electrotopological indexes of monofunctional compounds to predict the corresponding retention index data. These networks can also be applied to the prediction of retention index data of acyclic and cyclic monoterpenes and a mixed set of monosubstituted and terpene compounds. Predictions by neural networks are generally in good agreement with predictions done by multilinear regression techniques. In the case of predicting retention index data of compounds from a class not represented in the training data, neural networks show strong deficiencies in comparison with multilinear regression methods.