Interface Self-Assembly of AuNPs Gap Monolayers Combined with RamEfficientNet for Label-Free Detection of Multiple Antibiotics in Milk Without Preprocessing

Due to signal suppression caused by large interfering macromolecules such as proteins, reliable and label-free detection of antibiotics in milk remains a significant challenge. Here, a surface-enhanced Raman scattering (SERS) platform based on self-assembled molecular layers combined with deep learning was developed to selectively enrich and enhance representative antibiotics. A gold nanoparticle monolayer SERS substrate with an average gap of 2.5   nm was successfully fabricated, achieving a detection limit as low as 4.88×10 -10   M for 4-MBA. Notably, this technique effectively suppressed the Raman signals from macromolecules in milk while allowing antibiotic molecules to access the nanogap hotspot regions. The detection limits for penicillin and ciprofloxacin in milk were determined to be 8.4×10⁻ 9 M and 5.34×10 -9   M, respectively. Additionally, to address the issue of spectral overlap caused by the simultaneous presence of multiple antibiotics, which traditional chemometric methods cannot resolve for accurate quantification, a novel deep learning model, RamEfficientNet, was developed. Unlike conventional single-peak approaches, this model analyzes the entire spectral range through CBAM attention mechanisms, enabling extraction and decomposition of individual antibiotic spectra from complex mixtures for precise quantification. The model successfully achieved rapid classification of Raman spectra for various concentrations and types of antibiotics, with a classification accuracy exceeding 95%. Furthermore, by combining this model with an MLP, an innovative artificial intelligence visualization of the spectral regression process was realized, with an R² value exceeding 0.99. In conclusion, the combination of gold nanoparticle gaps and artificial intelligence provides a solution for the simultaneous quantitative analysis of antibiotics.