Prediction of infrared spectra from chemical structures of organic compounds using neural networks

Abstract

Chemical constitutions of organic compounds are encoded using a previously described graph-theoretical approach (Clerc and Terkovics, Analytica Chimica Acta, 235 (1990) 93–102). Neural networks are used to map the correlations between the IR spectral properties and the structure descriptors. The correlations thus established are then used to predict the IR spectral properties from the chemical structure. For a test set with members from many different compound classes about 80% of the predictions were satisfactory.