Integrating\nMetal–Phenolic Networks-Mediated\nSeparation and Machine Learning-Aided Surface-Enhanced Raman Spectroscopy\nfor Accurate Nanoplastics Quantification and Classification

Increasing\naccumulation of nanoplastics across ecosystems poses\na significant threat to both terrestrial and aquatic life. Surface-enhanced\nRaman scattering (SERS) is an emerging technique used for nanoplastics\ndetection. However, the identification and classification of nanoplastics\nusing SERS faces challenges regarding sensitivity and accuracy as\nnanoplastics are sparsely dispersed in the environment. Metal–phenolic\nnetworks (MPNs) have the potential to rapidly concentrate and separate\nvarious types and sizes of nanoplastics. SERS combined with machine\nlearning may improve prediction accuracy. Herein, we report the integration\nof MPNs-mediated separation with machine learning-aided SERS methods\nfor the accurate classification and high-precision quantification\nof nanoplastics, which is tailored to include the complete region\nof characteristic peaks across diverse nanoplastics in contrast to\nthe traditional manual analysis of SERS spectra on a singular characteristic\npeak. Our customized machine learning system (e.g., outlier detection,\nclassification, quantification) allows for the identification of detectable\nnanoplastics (accuracy 81.84%), accurate classification (accuracy\n> 97%), and sensitive quantification of various types of nanoplastics\n(polystyrene (PS), poly(methyl methacrylate) (PMMA), polyethylene\n(PE), and poly(lactic acid) (PLA)) down to ultralow concentrations\n(0.1 ppm) as well as accurate classification (accuracy > 92%) of\nnanoplastic\nmixtures at a subppm level. The effectiveness of this approach is\nsubstantiated by its ability to discern between different nanoplastic\nmixtures and detect nanoplastic samples in natural water systems.