Machine learning-assisted ratiometric fluorescence sensor array for recognition of multiple quinolones antibiotics

Abstract

Developing analytical methods for simultaneous detection of multiple antibiotic residues is crucial for environmental protection and human health. In this study, a dual lanthanide fluorescence probe (GDP-Eu-Tb) based on nucleotides has been designed. The addition of quinolone antibiotics (QNs) quench the Eu³⁺ fluorescence signal through the inner filter effect (IFE) and exhibit characteristic peaks, enabling ratio fluorescence detection of levofloxacin (LVLX), gatifloxacin (GTLX), and moxifloxacin (MXLX). A ratiometric fluorescence sensor array is constructed using a single sensor element (GDP-Eu-Tb), combined with principal component analysis (PCA) and decision tree (DT) algorithms to model the relationship between fluorescence intensity ratios (I₄₅₀/I₆₁₆, I₄₆₀/I₆₁₆, I_463/I₆₁₆, I₄₆₈/I₆₁₆) and QNs. The performance of the DT model is evaluated using accuracy, precision, recall, and F1 score, with stability and generalizability confirmed by stratified ten-fold cross-validation. This approach demonstrates high sensitivity, selectivity and applicability and provides an effective solution for antibiotic residue detection.