Autocorrelation method adapted to generate new atomic environments: application for the prediction of 13-C chemical shifts of alkanes.

The concept of the multifunctional autocorrelation method governing global description of molecules was changed in order to take into account the structural environment of each atom. New atomic environments are generated as possible descriptors in QSARs and can be useful for database characterization. The principles of this approach are widely explained through a case study dealing with the design of a model allowing the simulation of the carbon-13 nuclear magnetic spectra for alkanes. Carbon atoms in alkanes are described by using as structural descriptors a vector corresponding to only four components vectors of the multifunctional autocorrelation method. The statistical method used for deriving the model was a classical three-layer feedforward neural network trained by the back-propagation algorithm and multilinear regression (MLR). The predictive ability of the ANN model was tested by -10%-out(L10%O) cross-validation method, demonstrating the superior quality of the neural model. The established model allows us the prediction of the 13-C chemical shifts with success because since all types of carbons are taken into account without distinction of connectivity. The neural network possessed a 4:7:1 architecture with a sigmoid shape as a activation function. The model produced a cross-validation standard coefficient r between delta(exp) and delta(calc) about 0.99, while the cross-validation standard s and the mean error are equal to 0.87 and 0.60 ppm, respectively.