Integrated GC-MS, LC-MS/MS and FTIR-deep-learning analysis for chemical differentiation and rapid discrimination of four porcini mushroom species.

Porcini mushrooms have high culinary and commercial value, but species-dependent differences in their flavor-related chemical composition and rapid authentication remain insufficiently characterized. In this study, four wild porcini mushroom species, Boletus bainiugan, Butyriboletus roseoflavus, Lanmaoa asiatica, and Rugiboletus extremiorientalis, were investigated using an integrated analytical strategy combining GC-MS, LC-MS/MS, Fourier transform infrared spectroscopy (FTIR), chemometrics, deep learning, and SHAP-based model interpretation. In this workflow, GC-MS and LC-MS/MS were used as the metabolomic characterization layer to define species-associated chemical differences within the investigated sample set in volatile organic compounds (VOCs), organic acids, and amino acid-related metabolites, whereas FTIR was used as the rapid spectral acquisition layer for non-destructive classification. The metabolomic results identified 770 volatile organic compounds, 59 organic acids, and 67 amino acid metabolites, among which 200 VOCs, 27 organic acids, and 29 amino acid metabolites were screened as differential markers. These data revealed that the four porcini mushroom species differed markedly in aroma-active compounds and taste-related metabolites, including pyrazines, lipid oxidation-derived aldehydes/ketones, sulfur-containing volatiles, organic acids, and amino acid derivatives. FTIR spectra were then processed using different preprocessing methods and modeled using back propagation neural network (BPNN) and convolutional neural network (CNN) classifiers. Among the tested models, the standard normal variate (SNV)-CNN model achieved the best classification performance, with F1 scores of 0.96-1.00 for the four species. SHAP analysis further linked the key discriminative wavenumbers to characteristic molecular vibrations, mainly O-H, N-H, C-H, C-O, and C-C bands, which were consistent with the metabolomic differences revealed by GC-MS and LC-MS/MS. Therefore, this study establishes a two-level analytical platform in which chromatographic-mass spectrometric profiling provides chemical annotation, compositional interpretation, and hypothesis generation, while FTIR combined with deep learning enables rapid species discrimination. The proposed strategy offers both chemical insight and practical potential for quality evaluation of porcini mushrooms.