Machine Learning-Assisted Classification and Identification of Water-Soluble Polymers Using Protein-Based Fluorescent Sensors

Water-soluble polymers (WSPs) are widely used in diverse industrial, biomedical, and consumer applications; however, their release into the environment has raised growing concerns regarding their potential ecological impacts. The development of effective methods for detecting and identifying WSPs in aqueous systems is therefore of increasing importance. In this study, we developed a series of protein-based fluorescent sensors by conjugating a microenvironment-responsive fluorophore (N-(1-anilinonaphthyl-4)maleimide, ANM) into proteins through a linker-mediated conjugation reaction. The resulting sensors exhibited unique fluorescence responses in the presence of various synthetic and biorelated WSPs. These responses varied depending on the protein species and the ANM conjugation rate, reflecting distinct modes of interaction between the sensors and WSPs. The application of multivariate statistical analyses to the obtained fluorescence spectra enabled the precise and reproducible identification of multiple WSPs. Remarkably, nearly all sensors achieved 100% classification accuracy in two cross-validations, confirming their high potential in WSP identification. Moreover, the combination of multiple sensors further enhanced the robustness and reliability of classification, demonstrating complementary sensing behaviors among different proteins. This study provides the first proof of concept for employing naturally derived proteins as molecular sensors for WSP classification and identification.