Prediction of ¹H NMR Chemical Shifts Using Neural Networks

Abstract

Counterpropagation neural networks were applied to the fast prediction of 1H NMR chemical shifts of CHn groups in organic compounds. The training set consisted of 744 examples of protons that were represented by physicochemical, topological, and geometric descriptors. The selection of descriptors was performed by genetic algorithms, and the models obtained were compared to those containing all the descriptors. The best models yielded very good predictions for an independent prediction set of 259 cases (mean absolute error for whole set, 0.25 ppm; mean absolute error for 90% of cases, 0.19 ppm) and for application cases consisting of four natural products recently described. Some stereochemical effects could be correctly predicted. A useful feature of the system resides in its ability to be retrained with a specific data set of compounds if improved predictions for related structures are required.