Spectroscopic discrimination of bacterial species of variable pathogenicity through explainable machine learning

This is the authors' abstract. We don't add key points for this paper.

- two strains of the same species differing only in resistance phenotype. To ensure transparent decision-making and eliminate the black-box nature of machine learning models, SHapley Additive exPlanations (SHAP) analysis was applied to both RF and CNN models, which facilitated the convergence of both frameworks on the same discriminatory Raman regions, revealing conserved biochemical determinants of species identity. Complementary MCR-ALS decomposition further resolved the spectra into interpretable biochemical components and provided the rubric for biochemical differentiation. Together, this study aims to demonstrate an end-to-end explainable workflow that couples SERS with interpretable AI, offering a rapid, transparent approach for pathogenic disease diagnosis.