|
Internet Electronic Journal of Molecular Design - IEJMD, ISSN 1538-6414, CODEN IEJMAT
| ABSTRACT - Internet Electron. J. Mol. Des. April 2006, Volume 5, Number 4, 213-223 |
Diterpene Skeletal Type Classification and Recognition using Self-Organizing Maps
Vicente de Paulo Emerenciano, Marcus Tullius Scotti, Ricardo Stefani, Sandra A. V. Alvarenga, Jean Marc Nuzillard, and Gilberto V. Rodrigues
Internet Electron. J. Mol. Des. 2006, 5, 213-223
|
Abstract:
Kohonen Self-Organizing Feature Map (SOM Kohonen map) is a
technique used for pattern classification. The method can be
applied to classify different classes of organic compounds based on
13C NMR chemical shift data. This can be a very useful tool in
structure validation, which is one of the steps of automated
structure elucidation process. In this paper we present the use of
Kohonen ANN to predict and classify different skeletal types of
diterpenes. The Kohonen neural network was trained using Matlab
version 6.5 with the package Somtoolbox 2.0. A total of 957 cases
belonging to 12 different skeletal types of diterpenes were used to
train the network. During the training phase, 91.12% of the
patterns were highly correctly classified, while for the testing
phase, 75.22% of the input data were correctly classified by the
Kohonen neural network. As demonstrated by these results, SOM
Kohonen neural network can be a reliable tool to predict diterpene
skeletal types from 13C NMR spectrum data.
|