GC-Derived Volatile Profiling of Commercial Thyme and Oregano Labeled Products and Evaluation of Portable NIR Screening
Foods, cilt.15, sa.18, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 15 Sayı: 18
- Basım Tarihi: 2026
- Doi Numarası: 10.3390/foods15183245
- Dergi Adı: Foods
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Food Science & Technology Abstracts, Directory of Open Access Journals, Natural Science Collection (ProQuest)
- Anahtar Kelimeler: chemometrics, class imbalance, compositional data, GC-FID/MS, herb authentication, oregano, portable NIR, thyme
- Ankara Üniversitesi Adresli: Evet
Özet
Commercial aromatic herbs sold under the names thyme and oregano may encompass heterogeneous plant materials, making chemical-profile screening important even when botanical identity is unknown. This study characterized 110 commercial products marketed in Türkiye and evaluated whether portable near-infrared (NIR) spectroscopy could predict chemical-profile classes derived independently from GC-FID/MS data. Commercial labels were recorded but were not treated as verified species identities. Essential oils were measured in technical duplicate for 23 volatile compounds; replicate means were subjected to compositional data treatment, principal component analysis and Ward clustering. Two clusters were selected by the silhouette criterion (0.485), and cluster stability under technical-replicate resampling was high (median adjusted Rand index, 0.907). Cluster 1 (n = 17) was relatively enriched in sabinene, caryophyllene, carvacrol and delta-3-carene, whereas Cluster 2 (n = 93) showed higher myrcene, p-cymene, alpha-pinene and beta-pinene. Two NIR spectra were collected per sample. To prevent replicate leakage, spectra were averaged before repeated nested cross-validation; a grouped-replicate analysis was also performed. A majority-class baseline achieved 84.55% accuracy but only 0.500 balanced accuracy. The best primary model, shrinkage linear discriminant analysis, yielded a mean accuracy of 0.689, a balanced accuracy of 0.565, a macro-F1 of 0.530 and a Matthews correlation coefficient of 0.103. The grouped-replicate sensitivity analysis gave a balanced accuracy of 0.592. Thus, GC-FID/MS revealed a stable chemical-profile structure, but portable NIR did not reliably recover the minority chemical class. NIR may support exploratory triage after broader calibration, but the present data do not support stand-alone botanical authentication or definitive chemical-class assignment.