Application of quantitative chemometric analysis techniques to direct sampling mass spectrometry.

This paper explores the use of direct sampling mass spectrometry coupled with multivariate chemometric analysis techniques for the analysis of sample mixtures containing analytes with similar mass spectra. Water samples containing varying mixtures of toluene, ethyl benzene, and cumene were analyzed by purge-and-trap/direct sampling mass spectrometry. Multivariate calibration models were built using partial least-squares regression (PLS), trilinear partial least-squares regression (tri-PLS), and parallel factor analysis (PARAFAC), with the latter two methods taking advantage of the differences in the temporal profiles of the analytes. The prediction errors for each model were compared to those obtained with simple univariate regression. Multivariate quantitative methods were found to be superior to univariate regression when a unique ion for quantitation could not be found. For prediction samples that contained unmodeled, interfering compounds, PARAFAC outperformed the other analysis methods. The uniqueness of the PARAFAC model allows for estimation of the mass spectra of the interfering compounds, which can be subsequently identified via visual inspection or a library search.