Abstract. The design of systems for spectral data interpretation requires clustering of chemical compounds based on their spectral characteristics. Kohonen networks have been shown to be efficient tools to achieve this clustering. These auto-organising systems perform a mapping between a high-dimensional variable space and a two-dimensional one. An application to infrared spectra of organic compounds is presented here. The non-supervised learning algorithm used allows classification of compounds by spectral characteristics without a priori knowledge. An analysis of the distribution of spectra on the resulting maps is used to build models for predicting the presence or absence of specific structural features. The performance of the models in recognising structural features is discussed and compared with the prediction of a multilayered feed forward network (MLFFN). Localisation of compounds wrongly classified by the MLFFN on the Kohonen maps allows to establish a link between the supervised and the unsupervised approaches.
In the last few years, intensive research by several groups has shown that neural networks can be used to analyse spectral data for structural elucidation, and that their performance approaches that of an expert in the field. The construction of such networks, their training and evaluation, requires large structural and spectral databases and significant computational resources and time. However, once the network has been completed it can be used very effectively for practical applications on an ordinary desktop computer. In this article we describe the methodology for creating such a network for infrared and mass spectra, and present a program for use on a personal computer, either connected to a spectrometer or independently. The program accepts data in ASCII format, both for the network description and for the spectral information. This approach permits the use of neural networks in an analytical laboratory with limited computational resources. Keywords: neural networks, infrared spectroscopy, mass spectroscopy, structure determination.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.