RAPID AND ACCURATE METHOD FOR ESTIMATING MOLECULAR-WEIGHTS OF ORGANIC-COMPOUNDS FROM LOW RESOLUTION MASS-SPECTRA

A fast improved method of estimating the molecular weights of volatile organic compounds from their mass spectra has been developed and implemented with a pattern recognition/expert system. A pattern recognition-based classifier is followed by molecular weight estimators for six classes. The classes, verified by SIMCA class modeling, are: nonhalobenzenes; chlorobenzenes: bromo- and bromochloroalkanes/alkenes; mono- and dichloroalkanes/alkenes; tri-, tetra- and pentachloroalkanes/alkenes; and unknowns. For the first five classes empirical linear corrections, determined by the expert system, are iteratively applied to MAXMASS, a mass spectral feature, to yield molecular weight estimates. Two iterations are used except for the very large and diverse unknown class. For this class only one iteration is used together with a combination of MAXMASS and HIMAX1, another spectral feature. As far as possible, the system uses ternary encoded intensities in the mass spectra and high intensity masses. New rules for the unknown class were derived from 32 nontarget National Institute of Standards and Technology reference spectra. Extensive testing of the system was conducted vs. the self training interpretive retrieval system (STIRS) with reference and field gas chromatography-mass spectrometry (GC-MS) spectra. The probable errors, determined with a robust statistic, were 0.5 Da for training and spectrally similar test reference spectra and 5.0 and 1.3 Da for randomly selected and field GC-MS spectra. The test results show that the domain of expertise of this present system greatly exceeds the original training domain. The system will run on DOS-based personal computers.