Researchers have unveiled a pioneering machine learning-driven biosensor system that promises rapid, accurate, and calibration-free detection of microcystin-lysine-arginine (MC-LR), a potent toxin produced by cyanobacterial harmful algal blooms. MC-LR poses serious threats to human health, including liver damage and elevated cancer risks, leading the World Health Organization to set a strict safety threshold of 1 microgram per liter in drinking water. This breakthrough leverages integrated biosensing and multi-parameter water quality data to overcome a major limitation in current detection methods: the need for repeated sensor recalibration due to shifting water conditions.
The team, composed of scientists from Hanbat National University in South Korea and the University of Central Florida, developed a hybrid framework that unites portable screen-printed carbon electrode (SPCE) biosensors with an advanced machine learning algorithm, Extreme Gradient Boosting (XGBoost). While SPCE biosensors provide a low-cost and rapid means to detect MC-LR by measuring electrochemical impedance changes, their performance typically suffers from interference by varying environmental factors such as pH, turbidity, electrical conductivity, and other water quality parameters.
To address this challenge, the researchers collected an extensive dataset comprising 201 measurements from 27 diverse aquatic sites across Florida. These spanned freshwater, estuarine, and transitional environments, capturing a wide spectrum of physicochemical water characteristics. For each sample, parameters including pH, turbidity, electrical conductivity, total dissolved solids, ultraviolet absorbance at 254 nm (UV254), and the biosensor’s electrochemical impedance were recorded. Feeding this comprehensive dataset into the XGBoost model enabled the prediction of actual MC-LR concentrations without the need for individual sensor recalibrations tailored to each unique water matrix.
Performance metrics demonstrated the model’s robustness, with a Nash-Sutcliffe efficiency of 0.89 and a root mean square error of just 13.21, confirming its high accuracy across heterogeneous environmental samples. The application of Shapley Additive Explanations (SHAP), an interpretable artificial intelligence technique, revealed the dominant predictive features influencing toxin concentration estimation. The biosensor’s electrical impedance emerged as the most critical factor, followed closely by electrical conductivity, pH, UV254 absorbance, and turbidity, underscoring the necessity of integrating multi-parameter water quality data for reliable predictions.
This innovative approach fundamentally transforms the existing workflow for MC-LR detection. Unlike conventional protocols requiring time-consuming and labor-intensive sensor recalibration for different water samples, this unified machine learning model permits on-site toxin monitoring with reduced sensor consumption, lowering both expenses and environmental impact. The method offers a practical route to improving analytical efficiency and expanding the accessibility of real-time environmental surveillance.
Given the escalating incidence of harmful algal blooms fueled by climate change, the development represents a timely and critical advance. Rapid, accurate, and cost-effective toxin detection technologies are essential for safeguarding drinking water and protecting public health. According to Professor Jungsu Park of Hanbat National University, “This robust data-driven framework enhances the speed and precision of MC-LR detection in complex waters, paving the way for scalable and field-deployable monitoring solutions.”
The integration of biosensors with machine learning and comprehensive water quality monitoring stands as a notable example of how artificial intelligence can revolutionize environmental health technologies. As harmful algal blooms continue to threaten freshwater resources worldwide, such smart sensor systems will be vital for early warning, mitigation efforts, and ensuring water safety.
Article Title: Calibration-free on-site detection of microcystin-LR using integrated biosensing, multi-parameter water quality monitoring, and machine learning
News Publication Date: 15 June 2026
References: DOI: 10.1016/j.watres.2026.125832
Image Credits: Jungsu Park, Woo Hyoung Lee
Keywords: Artificial intelligence, Machine learning, Biosensors, Microcystin-LR, Water quality, Environmental health, Electrochemical impedance, Harmful algal blooms
