[Post-print] Discrimination of constructed air samples using multivariate analysis of full scan membrane introduction mass spectrometry (MIMS) data

Abstract

RATIONALE: Volatile and semi-volatile organic compounds (S/VOCs) are important atmospheric pollutants affecting both human and environmental health. They are directly measured as an unresolved mixture using membrane introduction mass spectrometry (MIMS). We apply chemometric techniques to discriminate, classify, and apportion air samples from a variety of sources. METHODS: Full scan mass spectra of lab constructed air samples were obtained using a polydimethylsiloxane membrane interface and an electron ionization ion trap mass spectrometer. Normalized full scan spectra were analyzed using Principal Component Analysis (PCA), cluster analysis, and k-nearest neighbours (kNN) for sample discrimination and classification. Multivariate curve resolution was used to extract pure component contributions. Similar techniques were applied to VOC mixtures sampled from different woodsmoke emissions and from the headspace above aqueous hydrocarbon solutions. RESULTS: PCA successfully discriminated 32 constructed VOC mixtures from nearly 300 air samples, with cluster analysis showing similar results. Further, kNN classification (k=1) correctly classified all but one test set sample, and MCR successfully identified the pure compounds used to construct the VOC mixtures. Real-world samples resulting from the combustion of different wood species and those associated with water contaminated with different commercial hydrocarbon products were similarly discriminated by PCA. CONCLUSIONS: Chemometric techniques have been evaluated using full scan MIMS spectra with a series of VOC mixtures of known composition containing known compounds, and successfully applied to samples with known sources, but unknown molecular composition. These techniques have application to source identification and apportionment in real-world environmental samples impacted by atmospheric pollutants.