Machine Learning-Enabled Multidimensional Ratiometric Sensor Array for Accurate Discrimination and Quantification of Tetracyclines.

Extensive use of tetracycline antibiotics (TCs) threatens ecosystems and human health, necessitating sensitive and selective analytical monitoring approaches. Herein, we report an amino-functionalized mesoporous-silica-confined nanosensor (PLNP@SiO2-NH2) that enables self-referenced ratiometric detection and pattern recognition-based discrimination and quantification of TCs. Aminated mesoporous silica effectively enriches and rigidly immobilizes TCs, triggering the turn-on of the intrinsic green fluorescence signal (IG) of TCs. Meanwhile, the strong UV absorption of TCs at 254 nm induces an inner-filter effect, attenuating the effective excitation of the nanoprobe and leading to the decay of the 700 nm near-infrared emission reference signal (I700). These directional response signals exhibit remarkable intrinsic anti-interference capability and signal amplification. Leveraging the multichannel IG/I700 ratiometric fingerprints under single-wavelength excitation, combined with linear discriminant analysis (LDA) and hierarchical cluster analysis (HCA), the sensor effectively discriminates tetracycline (TC), oxytetracycline (OTC), chlortetracycline (CTC), and doxycycline (DOX) across wide concentration gradients in both single-component and binary mixture systems, with discrimination sensitivity surpassing that of existing sensor arrays. The detection limits for the four TCs are 0.136 μM (OTC), 0.021 μM (DOX), 0.039 μM (TC), and 0.247 μM (CTC). The sensor demonstrates excellent antimatrix interference capability (matrix effect values from 0.92 to 1.04) and quantitative accuracy in milk samples (recoveries ranging from 95.40% ± 1.68% to 109.40% ± 1.12%). The sensor array also exhibits outstanding TCs discrimination capability in milk samples, confirming its robustness and reliability for high-throughput screening in complex food matrices.