Integrating cyclodextrin fluorescence sensing array and machine learning for rapid screening and detection of...

Perfluoroalkyl substances (PFASs) are persistent environmental toxins posing significant health risks, with drinking water being a major exposure route. Current gold-standard detection via liquid chromatography-mass spectrometry (LC-MS) is hindered by high cost, complexity, and lack of portability, limiting rapid on-site screening. While fluorescence sensors offer alternatives, existing designs lack multiplex detection capability or require intricate fabrication. To address this gap, we developed a supramolecular fluorescence sensor array exploiting host-guest chemistry between β-cyclodextrin polymer (β-CDP) and dye probes, integrated with machine learning for rapid multi-PFAS screening in water.

A 4 x 6 sensor array was constructed using β-cyclodextrin polymer (β-CDP) complexes with four dyes (NPN, AFR, CC, PR). Differential competitive binding of six PFASs generated unique fluorescence response patterns. These PFASs included perfluorooctanoic acid (PFOA), perfluorooctane sulfonate (PFOS), perfluorodecanoic acid (PFDA), perfluorononanoic acid (PFNA), perfluoroheptanoic acid (PFHpA), and perfluorohexanoic acid (PFHxA). Linear discriminant analysis (LDA) enabled simultaneous discrimination and quantification of all six PFASs within 10 min, achieving detection limits of 38 ng/L (PFOA) and 31 ng/L (PFOS). The array accurately classified PFAS mixtures (binary to quaternary) at μg/L levels. A modular deep learning platform quantified PFASs in real water samples with only 0.66 % relative error versus LC-MS. Validation using surface water spiked with 5 μg/L PFOA confirmed high accuracy (predicted: 5.2345 μg/L; LC-MS: 5.2692 μg/L).

This work establishes the host-guest fluorescence array coupled with machine learning for multiplex PFASs detection, overcoming limitations of single-analyte sensors and lab-bound instruments. The method provides a portable, cost-effective platform for on-site screening with LC-MS-level accuracy, addressing urgent needs for environmental monitoring. Its modular design allows seamless integration into field-deployable devices, offering transformative potential for rapid water quality assessment and regulatory compliance.