Positive matrix factorization: A data preprocessing strategy for direct mass spectrometry-based breath analysis

Abstract

Interest in exhaled breath has grown considerably in recent years, as breath biosampling has shown promise for non-invasive disease diagnosis, therapeutic drug monitoring, and environmental exposure. Real time breath analysis can be accomplished via direct online mass spectrometry (MS)-based methods, which can provide more accurate and detailed data and an enhanced understanding of the temporal evolution of exhaled VOCs in the breath; however, the complicated chemical composition and large raw datasets involved in breath analysis have hindered the discovery of sources contributing to the exhaled VOCs. The positive matrix factorization (PMF) receptor model has been widely used for source apportionment in atmospheric studies. Since the exhaled VOCs contain compounds from various sources, such as alveolar air, mouth air and respiratory dead-space air, PMF may be also helpful for source apportionment of exhaled VOCs in the breath. Thus, this study explores the application of PMF in the pretreatment of direct breath measurement data. The results indicate that (i) endogenous compounds and background contaminants sources can be readily distinguished by PMF in data obtained from replicate measurements of human exhaled breath at single time points (~30 s/measurement), which may benefit both exhalome investigations and the identification of exposure biomarkers; (ii) sources resolved from online measurement data collected over longer periods (1.5 h) can be used to isolate the evolution of exhaled VOCs and investigate processes such as the pharmacokinetics of ketamine and its major metabolites. Therefore, PMF has shown promise for both data processing and subsequent data mining for the ambient MS-based breath analysis.