Species-level identification of Nocardia spp. from clinical samples via intelligent analysis of Raman spectroscopic...

Nocardia spp. are clinically opportunistic pathogens that are frequently underdiagnosed. They often lead to severe clinical consequences. These infections are often invasive, involving the lungs, nervous system, skin, and soft tissues. Different Nocardia spp. show significant differences in virulence and antimicrobial susceptibility. However, clinical manifestations are highly diverse, and species-level identification remains technically difficult. The precise diagnosis of Nocardia spp. is challenging. Therefore, rapid identification of Nocardia spp. is essential for guiding clinical treatment.

In this study, an intelligent analytical model integrating machine learning (ML) with surface-enhanced Raman spectroscopy (SERS) was developed for the rapid identification of seven clinically common Nocardia spp. We isolated and cultured 46 Nocardia strains, including seven Nocardia spp. from clinical samples. For each strain, a total of 64 SERS spectra were generated to enhance data reproducibility. We performed principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) to evaluate spectral differences among Nocardia spp. Subsequently, we developed and optimized nine machine learning models. We quantitatively assessed model performance using Accuracy, Precision, Recall, F1-score, fivefold cross-validation, and further evaluated classification performance using confusion matrices and receiver operating characteristic (ROC) curves. We applied the SHapley Additive exPlanations (SHAP) method to interpret and analyze the optimal model.

Among all models, the support vector machine (SVM) achieved the highest identification accuracy of 99.47%, demonstrating superior classification performance.

SERS-SVM is an accurate and effective analytical technique for the rapid identification of seven clinically common Nocardia spp.