APPLICATION OF ELASTIC NET FOR SPECTRAL INTERFERENCES CORRECTION IN ICP-MS DETERMINATION OF REE IN PLANT SAMPLES

Rare Earth Elements (REE) are important nowadays for multiple industrial applications. The growing in the consumption of REE increases their emission into the environment, as well as their absorption by living organisms. Inductively coupled plasma mass spectrometry (ICP-MS) is the preferred technique for the determination of REE in different matrices due to its multielemental capability, and consequent speed of analysis, and high sensitivity. However, problems with spectral interference from oxides and hydroxides of some REE and Ba on other REE reduce the accuracy of the measurements. Several studies have been proposing approaches to overcome these difficulties, such as desolvation systems [1], mathematical corrections [2], chromatographic separation [3] and high resolution ICP-MS [4]. Other option is the employment of chemometric methods, being principal component regression (PCR) and partial least squares (PLS) the two most popular, which have proven to overcome problems of spectral overlaps [5]. The elastic net (EN) is a hybrid regularized least squares regression method used in learning and variable selection. EN linearly combines l1 penalty term (least absolute shrinkage selection operator -LASSO) and l2 penalty term (ridge regression - RR), which ensure democracy among groups of correlated variables [6]. The aim of this study was to investigate the efficiency of EN regression model in overcoming spectral interference in REE determination in plant samples, by ICP-MS. For that, the ability of elastic net to predict the concentrations of REE was compared to ordinary least squares (OLS), by applying them to a certified reference material (CRM). Table 1 presents the results obtained for the CRM BCR 670 (Aquatic plant). REE concentrations obtained with EN were more in agreement with the certified values, showing that this regression model overcame satisfactorily problems due to spectroscopic interference. By applying this model, concentrations of unknown samples can be predicted more accurately than with OLS.

Table1: Average REE concentrations (g g-1) obtained for the CRM BRC670 by ICP-MS using OLS and EN regression models.
La Ce Pr Nd Sm Eu Gd Tb Dy Ho Er Tm
Certified 0.487 0.99 0.121 0.473 0.094 0.0232 0.098 0.014 0.079 0.0158 0.044 0.0057
OLS 0.430 0.86 0.100 0.410 0.080 0.0320 0.123 0.016 0.070 0.0200 0.040 0.0100
EN 0.484 1.01 0.123 0.480 0.094 0.0229 0.100 0.013 0.080 0.0163 0.043 0.0065
Acknowledgements: CNPq, FAPERJ and CAPES for financial suppo