Machine Learning-Driven Multi-Emission Fluorescence Array for Simultaneous Size Discrimination and Quantification of Gold Nanoparticles

Gold nanoparticles (AuNPs) exhibit size-dependent environmental behaviors and bioaccumulation risks, necessitating precise characterization of their hydrodynamic dimensions and concentrations for toxicity assessment. Existing analytical platforms are limited in their ability to perform efficient multidimensional analysis due to time-consuming procedures and the reliance on large, costly equipment. Herein, we developed a multiemission fluorescence sensing array that assisted with machine learning analysis for the simultaneous size discrimination and quantification of AuNPs. The sensor integrates europium-doped ZnS quantum dots encapsulated in bovine serum albumin (BSA-Eu:ZnS QDs) with an extreme random forest algorithm. Upon incubation with AuNPs of different sizes, the triple emission channels (450/592/616 nm) showed distinct quenching patterns due to the inner filter effect. The ratiometric response (F₆₁₆/F₄₅₀) allowed for concentration quantification with a limit of detection (LOD) of 0.022 nM. Meanwhile, the triple-channel fingerprint enabled size discrimination using linear discriminant analysis and ExtraTrees algorithms. Notably, the discriminant model achieved 100% classification accuracy for 5-100 nm AuNPs based on feature machine learning analysis of spectral fingerprints. Validation using spiked environmental waters showed recovery rates of 90%-104%, demonstrating robustness against complex matrices. This sensing strategy offers significant potential for rapid nanomaterial characterization, enhanced analysis efficiency, and multidimensional data acquisition for environmental monitoring.