Comparison of SIMCA pattern recognition and library search identification of hazardous compounds from mass spectra (original) (raw)

SIMCA pattern recogmtlon methods have been apphed to mass spectra1 data from a target list of hazardous chenucals A scheme has been proposed for classification and ldentlfication of fnre classes of compounds including aromatics, chlorocarbons, bromocarbons, hydrocarbons and polychlonnated hphenyls (PCBs) In addrtlon, partial least squares regressIon has been used to predict the number of chlorine atoms present m the PCBs A trammg set of 429 compounds was used Classlficatlon models were derived by usmg pnnapal components analysis of the autocorrelatIon transformed spectra and were apphed to two gas chromatography-mass spectrometry ambient zur field samples and one hazardous waste sample An optunal number of three prmclpal components was determined For ldentlficatron the mass spectra of the unknowns were compared wrath the training set mass spectra m the predicted class The overall classlficat~on and rdentlficatlon rates for trammg spectra were 89% and for the unknowns were 92% and 82%, respcctwely Identlficahon results of the SIMCA scheme for an ambient ar field sample also was compared to that of probablhty based matching (PBM) Correct tdentlfication for the spectra m this sample by SIMCA was 71/84 (85%) as opposed to 35/84 (42%) by PBM There are many aspects of a pattern recogmtlon analysis of chemxal data One must first determtne the level of mformatlon required m order to solve the given problem [I] In the case of the ldentlflcatlon of hazardous compounds ' Present address