Rapid discrimination between wild and cultivated Ophiocordyceps sinensis through comparative analysis of label-free SERS technique and mass spectrometry

Abstract

Ophiocordyceps sinensis is a genus of ascomycete fungi that has been widely used as a valuable tonic or medicine. However, due to over-exploitation and the destruction of natural ecosystems, the shortage of wild O. sinensis resources has led to an increase in artificially cultivated O. sinensis. To rapidly and accurately identify the molecular differences between cultivated and wild O. sinensis, this study employs surface-enhanced Raman spectroscopy (SERS) combined with machine learning algorithms to distinguish the two O. sinensis categories. Specifically, we collected SERS spectra for wild and cultivated O. sinensis and validated the metabolic profiles of SERS spectra using Ultra-Performance Liquid Chromatography coupled with Orbitrap High-Resolution Mass Spectrometry (UPLC-Orbitrap-HRMS). Subsequently, we constructed machine learning classifiers to mine potential information from the spectral data, and the spectral feature importance map is determined through an optimized algorithm. The results indicate that the representative characteristic peaks in the SERS spectra are consistent with the metabolites identified through metabolomics analysis, confirming the feasibility of the SERS method. The optimized support vector machine (SVM) model achieved the most accurate and efficient capacity in discriminating between wild and cultivated O. sinensis (accuracy = 98.95%, 5-fold cross-validation = 98.38%, time = 0.89s). The spectral feature importance map revealed subtle compositional differences between wild and cultivated O. sinensis. Taken together, these results are expected to enable the application of SERS in the quality control of O. sinensis raw materials, providing a foundation for the efficient and rapid identification of their quality and origin.

Keywords: Cultivation; Machine learning; Metabolomics; Ophiocordyceps sinensis; Surface-enhanced Raman spectroscopy.

Graphical abstract

Fig. 1

Workflow of classification and prediction of wild and cultivated O. sinensis samples based on SERS technology and ML algorithms.

Fig. 2

SERS spectral characteristic peaks and metabolite analysis. (A) Average SERS spectra of cultivated (N = 1500) and wild (N = 1200) O. sinensis, gray regions represent standard errors of SERS spectra within groups. (B) Deconvoluted SERS spectra of cultivated and wild O. sinensis, the highlighted color represents differential characteristic peaks, the dark gray indicates common characteristic peaks. (C) Common characteristic peaks and peak areas. (D) Loading plots of PC1 and PC2. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 3

SERS and metabolomics profiles of wild and cultivated O. sinensis. (A) PCA cluster analysis, and (B) OPLS-DA cluster analysis of SERS spectra. (C–K) Box plot of major SERS peak intensities for cultivated (green) and wild (orange) O. sinensis. (L) Correlation analysis of significant SERS peaks. (M) PCA cluster analysis, and (N) OPLS-DA cluster analysis of metabolites discovered by metabolomics. (O) Heatmap analysis for metabolites in cultivated and wild groups. (P–R) Box plot of metabolites distinguishing among cultivated and wild groups. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 4

SVM performance evaluation and feature importance matching of metabolites. (A) SVM hyperparameter optimization. (B) Training curve. (C) ROC curve and AUC values. (D) Binary confusion matrix of SVM. (E) The spectral feature importance map found by SVM, where wavenumbers with high feature importance scores are important for cultivated and wild O. sinensis recognition. (F) Feature importance ranking and biological significance.