Enhancing Selectivity in Electrochemical PFAS Sensing: Leveraging Machine Learning to Overcome Selectivity Challenges

High Resolution Image Download MS PowerPoint Slide Per- and polyfluoroalkyl substances (PFAS) are persistent pollutants that require highly selective and sensitive detection in complex water samples. Although sensors labeled as molecularly imprinted polymers (MIPs) are widely reported, sensors fabricated by electropolymerizing ortho-phenylenediamine on glassy carbon electrodes using literature MIP protocols exhibited inconsistent, template-dependent selectivity among four PFAS targets─PFOS, PFOA, PFBS, and PFBA─when evaluated by differential pulse voltammetry with a ferrocenemethanol redox probe. However, the PFOS-templated sensor did not show a measurable response to octanesulfonic acid, the hydrogenated analogue of PFOS, indicating sensor selectivity to CF 2 groups but not to specific CF 2 chain length. In particular, MIPs prepared with smaller templates (PFBS, PFBA) demonstrated stronger selectivity toward their respective targets, whereas those with larger templates (PFOS, PFOA) showed reduced sensitivity and increased cross-reactivity. Despite these limitations, the sensors exhibited stable and reproducible redox responses to increasing PFAS concentrations, achieving detection limits of 3.48 ng L –1 (PFOS), 2.00 ng L –1 (PFOA), 1.33 ng L –1 (PFBS), and 1.08 ng L –1 (PFBA). Leveraging machine learning, a classification model was trained on differential pulse voltammetry data, reaching 95% accuracy in distinguishing the four PFAS. This work demonstrates a methodological advancement in the integration of ML models that reached system-level selectivity even greater than chemical selectivity, offering a promising platform for future PFAS monitoring.