Quantification of nitrite in beverages and pickled foods with a high-sensitivity amperometric biosensor enhanced by a multilayer perceptron neural network

Abstract

Accurate nitrite detection in beverages and pickled foods is crucial for food safety but remains challenging due to matrix complexity, particularly interference from salinity. To address this, a highly sensitive electrochemical sensor was constructed by modifying a screen-printed electrode with an electrodeposited platinum–palladium nanoparticles/gold layer (Pt–Pd NPs/Au/SPE). The electrocatalytic effect of the bimetallic nanoparticles conferred a 1.60-fold sensitivity enhancement, enabling the sensor to achieve a wide linear range (1–7500 µM), a low detection limit of 0.11 µM, and high sensitivity (226.03 µA mM⁻¹ cm⁻²). Crucially, quantification errors caused by salinity were corrected through a novel strategy that couples the developed Pt–Pd NPs/Au/SPE sensor with a commercial salinity meter. The NaCl concentration measured with the salinity meter served as the key input to a multilayer perceptron (MLP) neural network, which specifically compensated for the matrix effect. This intelligent compensation reduced the mean absolute error of nitrite quantification from 45.99% to 4.14%. The method was successfully applied to commercial beverages and pickled food, such as cola, milk, and pickled ginger, onion, garlic, and mustard, achieving recoveries of 92.77–106.56%. This work provides a reliable tool for food analysis and demonstrates a practical AI-assisted approach to overcome matrix interference in electrochemical sensing.