Field-Deployable Plasmonic Sensing and Machine Learning Classification of Microplastics Using Peptide–AuNP Conjugates

Microplastic (MP) debris ($1 \mu \mathrm{m}-5 ext{mm}$) is now widely found in oceans, rivers, and even municipal tap water, yet routine monitoring remains limited due to the high cost and time demands of laboratory-based FTIR or Raman spectroscopy. We introduce a peptide-gold nanoparticle (AuNP) assay that converts polymer-specific interactions into a rapid plasmonic colorimetric response, further analyzed through machine learning (ML)-based classification. Cysteineterminated peptides with affinity toward polystyrene (PS) and polypropylene (PP) were immobilized on 20 nm AuNPs. Upon exposure to the target microplastics, peptide-driven aggregation was triggered, producing a red-shift of the localized surface plasmon resonance (LSPR) band from 533 nm within five minutes. Spectral parameters obtained from UV-Vis analysis, together with dynamic light scattering (DLS) measurements, provided features such as $\lambda \max, \Delta \lambda$, full width at half maximum (FWHM), integrated absorbance, and Z-average hydrodynamic size. Random Forest, K-nearest neighbors (KNN), and Hierarchical Clustering were employed to classify MPs across six size ranges ($100 ext{nm}-250 \mu \mathrm{m}$) and four concentration tiers $\left(0.1-5 \mu \mathrm{g} ext{mL}^{-1} ight)$, successfully assigning blind samples to the correct size-concentration cluster. The assay functions effectively in tap water, achieving a limit of detection of $0.1 \mu \mathrm{g} ext{mL}^{-1}$ and a linear detection range up to $10 \mu \mathrm{g} ext{mL}^{-1}$ comparable to electrochemical platforms but without the need for a potentiostat or extended incubations. By integrating nanoscale molecular recognition with spectrum-based ML, this method offers a field-deployable, chemically specific solution for real-time microplastic monitoring.