ARTIFICIAL NEURAL NETWORKS APPLICATIONS. PART 6 USE OF NON-BONDED VAN DER WAALS AND ELECTROSTATIC INTRAMOLECULAR ENERGIES IN THE ESTIMATION OF ~(13)C-NMR CHEMICAL SHIFTS IN SATURATED HYDROCARBONS

摘要

The ~(13)C-NMR chemical shifts of alkanes and cycloalkanes were estimated with multi-layer feed forward neural networks, using as structural descriptors the degree of the carbon atom, the non-bonded van der Waals intramolecular energy, and the local electrostatic energy. The neural network Quantitative Structure Property Relationship (QSPR) model provides better results than the multilinear regression one. The predictive ability of the neural networks was tested with leave-one-out method. Neural networks with a bell-shape hidden activation function or with linear output activation function were tested, with good results.