Machine learning-enhanced identification of fluorophilic interactions for improved SERS detection of PFOA

Abstract

Per- and polyfluoroalkyl substances (PFASs), such as the legacy C8 compound perfluorooctanoic acid (PFOA), pose significant environmental and health risks due to their persistence and widespread use. While surface-enhanced Raman spectroscopy (SERS) has shown promise for PFAS detection, challenges remain in achieving high sensitivity and understanding the underlying molecular interactions. This study combines fluorinated thiol-modified SERS substrates with machine learning techniques to enhance PFOA detection and elucidate fluorophilic interactions. Three different fluorinated thiols were used to modify SERS substrates, and their performance in PFOA detection was compared to bare substrates. Surface modification improved SERS signal enhancement and lowered detection limits compared to unmodified surfaces. Machine learning algorithms, including partial least squares-discriminant analysis (PLS-DA), partial least squares regression (PLSR), and support vector machine (SVM) regression, were employed to classify and quantify PFOA based on Raman spectral features. The PLS-DA model successfully distinguished ligand-specific interaction patterns, demonstrating strong robustness and predictive performance, while SVM regression achieved a limit of detection of 3.61 ppb. We further validated the proposed machine learning-guided SERS approach for PFOA detection directly in water, demonstrating ligand-specific Raman responses arising from fluorophilic ligand–PFOA interactions. A linear and sensitive response was observed at environmentally relevant concentrations, confirming the practical applicability of the approach for aqueous PFAS monitoring. This approach offers a novel SERS-based strategy for detecting PFOA, with results anticipated to inform future developments in applying this strategy to a broader range of PFAS compounds and more complex environmental matrices.