Machine learning-driven SERS fingerprinting for high-accuracy serum screening of hepatic inflammation in mice

Hepatitis screening is clinically challenging. Conventional serological and biochemical assays rely on cumbersome procedures, require long detection times, and lack sufficient sensitivity to identify pathological changes. Surface-enhanced Raman scattering (SERS) enables rapid, label-free, and high-sensitivity molecular fingerprinting of biological fluids. It holds great promise for noninvasive serum biochemical analysis. Here, we report a machine learning-enhanced SERS platform based on a sugar-cube-like Ag nanocrystal (SC-AgN) substrate for accurate discrimination between healthy and DMN-induced hepatitis mouse serum. We fabricated the SC-AgN substrate with favourable uniformity, reproducibility and SERS enhancement to capture robust and well-resolved spectra from mouse serum samples. Clear spectral differences were identified in hepatitis serum, which correspond to protein denaturation, lipid peroxidation, and disturbed nucleic acid metabolism. Principal component analysis (PCA) was applied to reduce dimensionality, remove spectral noise, and eliminate redundant spectral features. The optimized spectral variables were further imported into a sparse partial least-squares discriminant analysis (SPLS-DA) model for precise sample classification. The finalized PCA-SPLS-DA model achieved an accuracy of 96.7%, a sensitivity of 100.0%, a specificity of 94.6%, and an AUC of 0.973 (95% CI, 0.945-0.992) for distinguishing DMN-induced hepatitis from healthy serum samples. This diagnostic performance outperforms that of conventional spectroscopic approaches. This work establishes a robust SC-AgN-based SERS diagnostic strategy integrated with the PCA-SPLS-DA model, which is expected to provide a rapid, low-cost and high-accuracy solution for hepatitis screening and auxiliary clinical diagnosis.