Machine Learning-Assisted Poly( l -Cysteine)-Functionalized Sensor Array for Heavy Metal Detection in Complex Aqueous Environments

Abstract Electrochemical analysis, valued for its rapidity and sensitivity, is an important tool for the rapid detection of environmental pollutants. However, its accuracy and reliability in real-world settings are often compromised by complex environmental interference. To address this challenge, this study developed a multifunctional electrochemical sensor array via screen-printing technique. The array integrates sensing units for heavy metals, pH, temperature, and conductivity, enabling the synchronous acquisition of multiple environmental signals. The heavy metal detection unit was fabricated by electropolymerizing l-cysteine onto a carbon nanotube (CNT)-modified electrode, forming a stable poly(l-cysteine)/CNT (P(L-Cys)/CNT) composite interface for the highly sensitive detection of Zn2+, Cd2+, and Pb2+. Under optimized conditions, the achieved detection limits were 15.54, 0.74, and 0.29 μg L−1 for Zn2+, Cd2+, and Pb2+, respectively. The specific interaction mechanism between the modified layer and the target metal ions was elucidated through a combination of density functional theory (DFT) calculations and experimental analysis. To effectively mitigate interference from complex environmental matrices, a multi-task hybrid model based on a Convolutional Neural Network and Long Short-Term Memory network (CNN-LSTM) was designed. This model utilizes the complete differential pulse voltammetry (DPV) curves and key environmental parameters collected by the sensor array as comprehensive input features, allowing the algorithm to automatically learn and compensate for the influence of environmental factors on the detection signals. This work provides an innovative solution that integrates sensor design with intelligent algorithms for the rapid and data-driven analysis of multiple heavy metal ions in complex aqueous environments.