Machine learning calibration of biosensors for microcystin toxin monitoring in freshwater

Portable screen-printed carbon electrode (SPCE) biosensors provide a rapid and low-cost method to detect microcystin-lysine-arginine (MC-LR), a highly potent toxin produced by cyanobacteria during harmful algal blooms in freshwater. Because MC-LR can damage the liver even at low concentrations and is associated with an increased risk of liver and colon cancer, the World Health Organization has set a guideline value for MC-LR in drinking water at 1 microgram per liter.

SPCE sensors work by measuring changes in electrochemical signals that reflect the concentration of toxins. However, the accuracy of these sensors is highly influenced by the water being tested. Factors such as pH, turbidity, electrical conductivity, and other water quality parameters can interfere with sensor readings, often requiring recalibration after each water sample.

Researchers from Hanbat National University in South Korea and the University of Central Florida in the United States have developed a machine learning framework that accounts for differences in water quality, enabling accurate MC-LR measurements without repeated sample-specific calibrations. The research was led by Professor Jungsu Park of Hambat National University and Professor Woo Hyoung Lee of the University of Central Florida. This paper was made available online on March 26, 2026 and was published in Volume 298 of the journal. water research June 15, 2026.

“This study provides a robust data-driven framework to characterize biosensor-water matrix interactions and provides a practical approach to improve the speed and accuracy of in situ MC-LR detection in complex environmental waters.” Professor Park says.

To build and train the model, the team collected 201 measurements from 27 sites across Florida representing a wide range of water quality conditions, including freshwater, estuarine, and transition environments. For each water sample, pH, turbidity, electrical conductivity, total dissolved solids, and ultraviolet absorbance at 254 nanometers (UV) were measured.₂₅₄), and the electrochemical impedance (Z’) of the biosensor, which changes in response to MC-LR. These measurements served as input variables and the model was trained to predict the actual concentration of MC-LR.

Among the different machine learning models evaluated, Extreme Gradient Boosting (XGBoost) showed the best performance, achieving a Nash-Sutcliffe efficiency of 0.89 and a root mean square error of 13.21. This level of performance demonstrated that a single integrated model can accurately predict MC-LR concentrations across a variety of water samples without the need for separate calibration models for each condition.

To determine which input variables had the greatest impact on the model’s predictions, the researchers used an explainable artificial intelligence technique called Shapley Additive Explains (SHAP). They found that the electrical impedance of the biosensor was the strongest predictor of toxin levels, followed by electrical conductivity, pH, UV absorbance, and turbidity, indicating that incorporating water quality parameters improved the predictive accuracy of the biosensor.

“This framework eliminates the need for repeated sample-specific calibrations, reducing time, effort, and sensor consumption. Compared to traditional workflows, sensor usage is reduced.” and Reduce costs and environmental impact while improving analytical efficiency.,” Professor Park says.

As harmful algal blooms become more frequent as the climate changes, this data-driven approach could make monitoring for toxins faster, more accurate, and easier to implement in drinking and recreational water testing.

reference
Original paper title: Calibration-free on-site detection of microcystin-LR using integrated biosensing, multiparameter water quality monitoring, and machine learning
journal: water research
DOI: https://doi.org/10.1016/j.watres.2026.125832

About Hanbat National University (HBNU)
Hanbat National University (HBNU) is a public university located in Daejeon, South Korea, with its origins dating back to 1927 as the Hongseong Public High School Engineering Training School. In 2023, it was renamed Hanbat National University and has built a strong reputation for engineering, technology, and industry-academia collaboration. The university is known for its extensive partnerships with industry and its role in driving innovation through applied research and technology transfer. HBNU has been selected for several major national higher education initiatives, including university-industry collaboration and university innovation programs. HBNU has a vision of becoming a “world-standard industrial innovation university” and aims to strengthen its international competitiveness.
https://www.hanbat.ac.kr/eng/sub01_04.do

About Professor Park Jeong-soo of Hanbat University
Professor Jungsu Park is a researcher in environmental engineering at National Hanbat University in Daejeon, South Korea. He is the author of over 40 scientific publications. His research focuses on the application of artificial intelligence to water environmental management, including water quality and algal bloom prediction, water treatment process management, and PFAS analysis. His work includes the development of machine learning approaches such as ensemble learning and deep learning models, including generative adversarial networks, to support water environmental management.