The Performance of Partial Least Squares Methods in Virtual Nanosensor Array—Multiple Metal Ions Sensing Based on Multispectral Fluorescence of Quantum Dots

The design of chemical sensors and probes is usually based on selective receptors for individual analytes, however, many analytical tasks are dedicated to multi-analyte sensing or recognizing properties of the sample related to more than one analyte. While it is possible to simultaneously use multiple sensors/receptors in such cases, multi-responsive probes could be an attractive alternative. In this work, we use thiomalic acid-capped CdTe quantum dots as a multiple-response receptor for the detection and quantification of six heavy metal cations: Ag(I), Cd(II), Co(II), Cu(II), Ni(II), and Pb(II) at micromolar concentration levels. Multiplexing is realized via multispectral fluorescence (so-called virtual sensor array). For such a sensing strategy, the effective decoding of the excitation-emission spectrum is essential. Herein, we show how various parameters of chemometric analysis by the Partial Least Squares method, such as preprocessing type and data structure, influence the performance of discrimination and quantification of the heavy metals. The established models are characterized by respective performance metrics (accuracy, sensitivity, precision, specificity/RMSE, a, b, R²) determined for both train and test sets in replicates, to obtain reliable and repeatable results.