Self-organizing maps for the detection and classification of natural nanoparticles, nanoparticle systems and engineered nanoparticles characterized using single particle ICP-time-of-flight-MS

Abstract

The development of single-particle inductively coupled plasma time-of-flight mass spectrometry (spICP-ToF-MS) heralds a breakthrough in our ability to measure the multi-elemental composition of natural nanoparticles and colloids (NPs), and to characterize the dynamics, responses, and impacts of systems of natural NPs (NNPs). However, further developments and associated comparisons across studies and research groups are hindered by the lack of a consistent, reliable and comparable approach for detecting and differentiating NPs and NNPs. Self-organizing maps (SOM, aka Kohonen networks) are single-layer artificial neural networks that are widely used for pattern recognition and classification in the natural sciences and beyond. The SOM is a nonparametric statistical method which adapts to data structures and is robust to noise, outliers, and sparse data, making it especially suitable for peak detection and particle classification using raw spICP-ToF-MS time-series. This article provides a brief review of SOM and their outputs before demonstrating their ability to detect particles in spICP-ToF-MS time-series, and to characterize and compare NNPs. Additional considerations and research directions for the application of SOM to spICP-ToF-MS and particle data are then discussed. The raw data and algorithms used in this study are provided in the SI to facilitate the testing of SOM across research groups, and for comparing their performance with other methods.