Prediction of Substituent Types and Positions on Skeleton of Myrcane-Type Monoterpenoids using Generalized Regression Neural Network

Abstract

Aim: To explore the ability of GRNN as a tool of structural elucidation in predicting the substituent types on myrcane, one of the representative skeletons of monoterpenoids. Methodology: Generalized regression neural network (GRNN) was used in the study. Carbon-13 (13C) NMR chemical shift values of skeletons of 104 myrcane monoterpenoids were used as the input data used for the network. Each substituent type on the skeleton of the different compounds were coded and used as the output data for the network. These data were used to train the network while the spread constant of the GRNN was varied.  After training, the network was simulated using 15 test compounds. Results: GRNN at a spread constant of 1.0 gave the best result. The network had between 80 to 90% recognition rates in 14 of the 15 test compounds. The network could not predict correctly the substitution pattern on ‘compound 11’ as all the positions was predicted to be unsubstituted. This could be due to the non-existence of precise rules for the compound. Conclusion: GRNN, one of the architectures of Artificial Neural Networks (ANNs), could be a powerful aid in the structural elucidation of organic compounds.