Application of the ART-2a Algorithm to Laser Ablation Aerosol Mass Spectrometry of Particle Standards

Abstract

Single-particle mass spectrometers are now commonly used to analyze atmospheric particles and generate tens of thousands of spectra from typical measurement campaigns. The ART-2a spectrum algorithm has been used to classify these spectra. In this work, we generate a range of particles that are models of those that are common in the atmosphere. A single-particle mass spectrometer is used to analyze these known particles, and the spectra are classified using ART-2a. The optimum vigilance parameter is approximately 0.5 while the optimum learning rate is approximately 0.05. The classifications elucidate limitations in generation of test particles, their analysis by single-particle techniques, and their classification by ART-2a.

Bioaerosol Mass Spectrometry for Rapid Detection of Individual Airborne Mycobacterium tuberculosis H37Ra Particles

“…The overall single particle BAMS mass spectra of M tuberculosis H37Ra, M. smegmatis, B. atrophaeus, and B. cereus looked similar to each other (Fig. 3) A novel pattern recognition algorithm (8), which is similar to an ART-2a clustering algorithm previously described (33,40), was used in the analysis of individual BAMS mass spectra and classified single particle types in real time. Used on its own, this data analysis scheme could not efficiently differentiate between these vegetative bacterial cells.…”

Internal Mixing of Pollutants for Submicron Particles Observed during Springtime in Japan

“…The principles were reported previously (eg. Zhao et al, 2005;Phares et al, 2001). Briefly, (1) Mass spectra are considered as vectors containing ion intensities for all m/z; (2) The vectors are normalized; (3) The vectors are categorized into a moderate number of clusters which consist of similar vectors.…”

Chemically-assigned classification of aerosol mass spectra

“…When classes were made more spectrally homogenous even for minor ion peaks by using a vigilance factor Ͼ0.7 in ART-2a, or equivalently in HCA by specifying a higher number of clusters, this resulted in many classes of only subtle differences and complicated interpretation. When a vigilance factor Ͻ0.7 (recommended for standard chemicals [8]) or a decreased number of cluster groups was chosen, spectra classified together were not always chemically similar. These observations indicated that a fixed vigilance factor or number of groups each had drawbacks for interpretation purposes, but the initial 0.7 and 25 values, respectively, were adequate for initially identifying the major chemical groups in the ambient dataset.…”

“…In work completed by others thus far, a variety of methods including principal components analysis (PCA) [2][3][4][5], hierarchical cluster analysis (HCA) [4 -6], and adaptive resonance theory (ART2a-an artificial neural network) [7,8], were applied. PCA was used to identify highly correlated chemical compounds, but it is inherently a group analysis, making individual particle information like particle size, surface area, or mass difficult to attribute among the principle (chemical) components.…”