Abstract View
Beyond Positive Matrix Factorization: The Strengths and Weaknesses of 3D Factor Analysis Methods
MICHAEL WALKER, Brent Williams, Washington University in St. Louis
Abstract Number: 482
Working Group: Instrumentation and Methods
Abstract
Advances in mass spectrometry have greatly expanded our understanding of aerosol chemistry and composition. With increasingly high time- and chemical-resolution measurements, a wealth of complex data necessitates more elegant analysis methods. Positive matrix factorization (PMF) has been widely applied within the aerosol science community to identify the key, physically-relevant features within datasets. When paired with mass spectrometry data, PMF generates factors of covarying mass-to-charge values and the corresponding time series that reconstruct the initial data. However, many of the newly developed and adopted analytical methods to probe aerosol composition feature an additional separation dimension that will not be considered in the two-dimensional PMF model. This separation dimension can be understood as retention time in chromatography measurements, temperature-based separation from thermograms, or size in particle-time-of-flight or mobility-based measurements. To enable a more detailed analysis of these types of datasets, the existing, Igor-based PMF Evaluation Tool that has utilized the PMF2 solver for PMF analysis has been enhanced to incorporate the Multilinear Engine 2 (ME-2) solver and three-dimensional factor analysis models. The PMF2 and ME-2 solvers offer various tradeoffs in terms of their computational efficiency, capabilities to handle large datasets, and ability to constrain solutions that should be considered when determining the proper data analysis approach. Furthermore, the specific model and solver choice needs to be directed by the scientific questions driving the study. Examples from a variety of laboratory and field measurements from different 3D mass spectrometry techniques will illustrate the importance of these decisions in the factor analysis process.