“Sweet Nanosheet”: An Antibody Mimic for Machine Learning-Assisted Ultra-Sensitive Immunochromatographic Assay for Pathogens

The bacterial surface is rich in diverse molecular features, and fully exploiting these natural recognition mechanisms provides innovative avenues for multimechanism detection of foodborne pathogens. Here, we developed a label-free, dual-modal LFIA platform based on the glycan-cluster effect for the efficient capture of Salmonella. The platform employs dextran-functionalized tungsten diselenide nanosheets (Dex-WSe₂) as core probes, where dextran coatings provide antibody-like high-affinity capture through multivalent glycan-bacteria interactions, while WSe₂ nanosheets act as dual signal transducers. Benefiting from exciton-plasmon coupling and charge-transfer effects, WSe₂ nanosheets not only possess inherent surface-enhanced Raman scattering (SERS) activity but also display distinct visible coloration due to their unique optical properties. Leveraging these dual features, the platform enables highly sensitive bimodal detection, achieving a visual detection limit of 10³ CFU/mL and an ultralow SERS detection limit of 52 CFU/mL. Furthermore, machine learning was introduced for multidimensional signal analysis: k-nearest neighbors (KNN) for qualitative concentration classification and random forest (RF) regression for quantitative prediction. The integrated model achieved 100% classification accuracy and an R² of 0.9977, demonstrating outstanding robustness. By combining glycan-based molecular recognition with machine learning strategies, the Dex-WSe₂ probe offers an efficient, stable, and intelligent platform for rapid on-site pathogen detection and food safety monitoring.