Use of a modified autocorrelation method to study C-13 chemical shifts-chemical structure relationships for alkanes

The concept of the multifunctional autocorrelation method governing global description of molecules was modified, in order to generate new atomic envi ronments. Such atomic environments allow the generation of atomic descripto rs by means of a modified autocorrelation method. The principle of this app roach is explained through a case study dealing with the design of a model allowing the simulation of the carbon-13 nuclear magnetic spectra for alkan es. The data set contains 243 values of chemical shifts. Carbon atoms in al kanes are described by using as structural descriptor a vector correspondin g to only four components of the multifunctional autocorrelation method. Th e established model allows us the prediction of the C-13 chemical shift wit h high success. Results obtained by means of multiple regression analysis w as less satisfying than those obtained using neural network approach. The q uality of the neural network model was evaluated by means of the Cross vali dation technique, with a mean error between delta (exp) and delta (calc) of 0.60 ppm. (C) 2000 Academie des sciences / Editions scientifiques et medic ales Elsevier SAS.