Generation of Simulated Mass Spectra of Small Organic Molecules by Computerized Pattern Recognition Techniques

An empirical method employing computerized pattern recognition techniques has been applied to the generation of simulated mass spectra of small organic molecules. Molecular structures are represented in computer-compatible form through the use of a fragmentation code which assigns code designations to specific groups of atoms and/or bonds within the molecules. Using such descriptions of molecules, pattern classifiers have been developed to predict the presence or absence of mass spectral peaks in each of 60 nominal m/e positions and to give a measure of the intensity of peaks in 11 of these. Information in the molecular descriptor lists which correlates with the appearance of specific peaks is shown to be present in relatively few of the descriptors developed. To test the complete system, a number of entire mass spectra were developed; in this test, 93% of the classifications were made correctly.