A dendrimer-based platform integrating surface-enhanced Raman scattering and class-incremental learning for rapidly detecting four pathogenic bacteria

Illustration of a novel dendrimer-based platform for detecting multiple foodborne pathogens. Poly(amidoamine) (PAMAM) dendrimers were combined with gold nanoparticles (Au NPs) to prepare PAMAM-based gold nanoassemblies (PGNAs). These PGNAs were then applied to a PAMAM-treated silicon wafer (Si) to fabricate the PGNAs/Si substrate for analyzing both spiked and real samples. Surface-enhanced Raman scattering (SERS) spectra obtained from this setup were processed using the LightGBM machine learning algorithm for bacterial classification. Key features contributing to accurate classification were identified using the Shapley additive exPlanations (SHAP) method. • A new bacterial detection platform integrating SERS with CIL model. • Precise nanostructure control using PAMAM to improve SERS performance. • Efficient CIL models were developed for accurate categorization of pathogens. • The SHAP method was utilized on CIL model for improving model interpretability. Rapid monitoring of pathogens is crucial for preventing foodborne diseases, thus making it an urgent need to develop efficient, fast, and simple methods for on-site detection of multiple pathogens. With advances in SERS-based label-free biosensors and machine learning, progress is evident, yet challenges such as complex substrates and limited model interpretability persist. In this work, we reported a novel dendrimer-based platform that integrated surface-enhanced Raman scattering (SERS) with class-incremental learning (CIL) for quick and simultaneous detection of four pathogenic bacteria, namely Escherichia coli , Salmonella Paratyphi B, Pseudomonas aeruginosa , and Staphylococcus aureus . First, the p oly( am ido am ine) (PAMAM)-based gold nanoassemblies on silicon wafer (PGNAs/Si) was designed as a highly active SERS substrate with an easy fabrication process. Compared with naked gold nanoparticles, both sensitivity and repeatability were improved by introducing a controllable dendritic nanostructure with ultra-small internal nanogaps. The detection limits of four pathogens in water, food, and dietary supplement matrices were all on the order of 10 CFU/mL. Subsequently, to analyze the complex SERS spectra and distinguish samples with different pathogens, CIL models were established using the light gradient boosting machine (LightGBM) algorithm. Notably, the discriminative models demonstrated excellent classification performance with an accuracy of over 93.44 %. Further, the SHapley Additive exPlanations (SHAP) method was utilized to identify the key SERS features in accurately discriminating different pathogens. Together, these results demonstrate that the dendrimer-based platform, by integrating SERS with CIL, can rapidly detect four pathogenic bacteria, underscoring its potential for food safety testing or quality monitoring in medical supplies manufacturing.