Selectively Quantify Toxic Pollutants in Water by Machine Learning Empowered Electrochemical Biosensors

Abstract Electroactive biofilm (EAB) sensors have become pivotal in water quality detection and early ecological risk warnings due to their remarkable sensitivity. However, it is challenging to identify multiple toxicants in complex water bodies concurrently. This research developed an innovative biosensor detection strategy combined with machine learning. To simultaneously quantify and qualitatively predict the presence of each toxin in a multitoxic system, we developed a prediction model (MEA-ANN) based on machine learning analysis of EABs by analyzing the electrochemical toxicity response parameters of various toxicants (Cd2+, Cr6+, triclosan, and trichloroacetic acid). Furthermore, the mean impact value was utilized to filter the characteristic response parameters of toxicants, enhancing the prediction accuracy and efficiency of the model. The optimized model (OMEA-ANN) demonstrated strong performance in predicting target toxicants within interference systems containing analogs. The practicability and feasibility of this model were validated using seven real water samples and spiked natural water samples, achieving R2 > 0.9. The novel, eco-friendly, and intelligent water ecological risk early warning strategy presented in this paper addresses the limitations of traditional EAB sensors. It expands the applicability of EAB sensors for detecting multiple toxicants in water, significantly advancing their role in water quality monitoring. This approach provides valuable insights for the intelligent management of sewage.