Identifying and assessing matrix effect severity in inductively coupled plasma optical emission spectrometry using non-analyte signals and unsupervised learning

An unsupervised data-driven methodology is used to quantify matrix effects caused by carbon and easily ionizable elements (ElEs) in inductively coupled plasma optical emission spectrometry (ICP OES). Background signals from nine plasma naturally-occurring species of Ar, H and O are used with principal component analysis (PCA) and affinity propagation (AP) clustering to evaluate the effects of complex matrices on ionic emission lines of Cd, Co, Cr and Pb. Matrix effect severity is then quantified based on Euclidean distance in principal component space from an average calibration curve point. The method has been applied to spiked solutions of Mediterranean Sea and Dead Sea water samples, and a significant correlation (- 0.997) was found between Euclidean distance and analyte recoveries. For sea water analysis, accurate results are found using external standard calibration (EC) when Euclidean distance <1 for a given sample, and/or when that sample point groups with the calibration curve after affinity propagation clustering. Thus, by applying the PCA-AP strategy, one needs to perform no addition/recovery experiment to evaluate EC applicability. In addition, it can be carried out on the fly, as the background species used to monitor plasma changes are simultaneously recorded with the analytical signals, and a specific algorithm can be added to the instrument control software to flag instances in which EC may be ineffective. This is a proof-of-concept study, and additional work is required to evaluate the method's applicability to a larger number of analytes and sample matrices. However, the PCA-AP method described here for ICP OES can be used to quantify matrix effects, allowing for informed decisions regarding calibration. It requires no additional sample preparation and can be easily implemented in routine analyses of such complex-matrix samples as sea water. (C) 2019 Elsevier B.V. All rights reserved.