EXPERT SYSTEM FOR ESTIMATES OF MOLECULAR-WEIGHTS OF VOLATILE ORGANIC-COMPOUNDS FROM LOW-RESOLUTION MASS-SPECTRA

MAXMASS, the highest mass with an intensity of 5% of the base peak in a low-resolution mass spectrum, has been found to be linearly correlated with the true molecular weights of 400 randomly selected spectra, yielding a family of parallel lines. A simple expert system using MAXMASS has been developed to estimate molecular weights of unknown volatile compounds from their mass spectra. This is an empirical rule-based system which will run on a personal computer. It consists of a classification module chained to separate molecular weight prediction modules for six classes. These are benzenes, chlorobenzenes, bromoalkanes/alkenes, mono- and dichloroalkanes/alkenes, tri-, tetra- and pentachloroalkanes/alkenes, and unknown (other) classes. The rules were derived from National Institute of Standards and Technology (NIST) reference spectra of 75 target toxic compounds from the first five classes and 32 from the unknown class. Performance tests of molecular weight predictions were made with the expert system and the self training interpretive and retrieval system (STIRS) on NIST reference spectra of the 107 training compounds; on a set of 32 test NIST reference spectra of other compounds found in air samples; on a set of 100 randomly selected NIST reference spectra; and on 30 spectra taken from actual field gas chromatography-mass spectrometry (GC-MS) samples. The expert system performed much better than STIRS on training spectra, STIRS performed better on test and random spectra, and the expert system performed better with field data. The median absolute value deviation from the true results for the expert system was 1 and 3 mass units for training and field GC-MS spectra, respectively. Moreover, the expert system is much faster than STIRS.