International Journal of Scientific & Engineering Research, Volume 5, Issue 12, December-2014 ISSN 2229-5518 IJSER © 2014 http://www.ijser.org Computer-Assisted Structure Verification of Eudesmane-type Sesquiterpenes using Generalized Regression Neural Network (GRNN) Taye Temitope Alawode*, Kehinde Olukunmi Alawode Abstract— This work describes procedures utilizing GRNN in the verification of structures of Eudesmane sesquiterpenes from 13 C NMR chemical shift values. In the first procedure, the substituent types on skeletons of 291 Eudesmane sesquiterpenes were coded and used as input data for the network. The 13 C NMR chemical shift values on the skeleton of the compounds were used as output data. After training, the network was simulated using thirty-four test compounds. Average and standard deviations were used to measure the accuracy of the predictions of the network. The procedure has a high potential to identify the Eudesmane skeleton as a substructure in the test compounds. A related procedure utilizing a GRNN trained employing 13 C NMR and coded substituents as input and output data respectively, was able to predict the substituents attached to various sites of the Eudesmane skeleton. Index Terms— 13 C NMR, Eudesmane skeleton, GRNN, Sesquiterpenes, substituents, verification, prediction ——————————  —————————— 1 INTRODUCTION rganic chemists are constantly faced with the challenge of either verifying chemical structures or elucidating the chemical structures of unknowns. Both processes involve the acquisition and analysis of an array of spectral data and both processes are known to be amenable to algorithmic solu- tions. Chemists frequently propose chemical structure(s) based on sample origin or knowledge of the potential prod- uct(s) of a particular chemical reaction. With this fore- knowledge, the approach generally adopted is to acquire and then examine the spectral data in terms of consistency be- tween spectroscopic expectations from the proposed structure and experimental data. This workflow requires experience in spectral interpretation, experimental access to the necessary data and, where appropriate, access to software tools for spec- tral prediction and comparison [1]. Methods for 1D NMR spectral prediction include rule-based approach for particular classes of compounds or as a suite of software tools covering one or more NMR active nuclei. Agreement between the rec- orded and predicted NMR spectral data is the primary tool used to identify the most probable structure in a set of sug- gested structures. Numerous studies devoted to NMR chemi- cal shift calculation have been reported [2-4]. The structure verification process compares a calculated chemical shift or spectrum; either 1D or 2D with the corre- sponding experimental data and the structural hypothesis is either accepted, rejected, or can be revised on the basis of vis- ual inspection or calculated mean or standard deviations [5]. Alternatively, it may be obvious from this analysis that addi- tional homo- or heteronuclear correlation data must be ac- quired to verify the structure or to test a revised structure. The two most widely used procedures for predicting NMR spectra are the construction of empirical models[6-8] and the applica- tion of prediction algorithms extracted from data collected within spectral databases[9-10]. Certain applications use both approaches simultaneously [11]. Prediction of 13 C NMR chem- ical shifts using artificial neural networks (ANN) has also been reported[12-15]. In a previous work[16], we have shown that GRNN, an ar- chitecture of ANN, could identify the substituents on the skel- eton of Eudesmane compounds when 13 C NMR chemical shift values at the various positions on their skeletons were used as inputs for the system. (Figs. 1 and 2 shows a single neuron model and the general structure of GRNN) [17]. In this work, we show that GRNN can predict the 13 C NMR chemical shift values at the various positions on skeletons of Eudesmane sesquiterpenes. We also demonstrate how these procedures may be used as complementary tools for structure verification and revision of selected Eudesmane sesquiterpenes. Fig. 1. Single Neuron Model [17] O ———————————————— x Kehinde Olukunmi Alawode is a lecturer with the Department of Electrical and Electronics Engineering, Osun State University, Osogbo, Osun State, Nigeria. Email: kenyem1@yahoo.com x *Corresponding Author: Taye Temitope Alawode is a lecturer with the Department of Chemical Sciences, Federal University Otuoke, Bayelsa State, Nigeria. Email: onatop2003@yahoo.com 839 IJSER