Factor-aware cross-modal mapping of GC-MS and FTIR resolves geographic and stalk-position contributions to volatile-matrix covariation in tobacco leaves from China.
- 1Shanghai Tobacco Group Co., Ltd Shanghai 200082 China [email protected] [email protected] [email protected].
- 2Department of Chemistry, Fudan University Shanghai 200433 China [email protected] [email protected] [email protected].
Abstract
Cross-platform correlations in heterogeneous biological materials can reflect shared experimental structure rather than intrinsic chemical coupling. Here, 90 flue-cured tobacco leaves from ten production areas and three stalk positions were analysed in a balanced 10 × 3 × 3 design using headspace solid-phase microextraction gas chromatography-mass spectrometry (HS-SPME-GC-MS) and Fourier transform infrared spectroscopy (FTIR). For a prespecified 8 × 8 cross-modal panel, pooled rank covariance was partitioned into production-area, stalk-position, area × position and within-cell residual components, with robustness assessed by design-preserving permutation, nested classification and leave-one-production-area-out validation. FTIR replicate consistency was high (median mean within-sample correlation ≈0.996). Both analytical platforms contained reproducible stalk-position information, and feature-level fusion achieved 84.4% internal accuracy, but this decreased to 42.2% when complete production areas were held out. Geographic origin dominated global analytical heterogeneity and 13 of 14 retained cross-modal associations (|ρ| ≥ 0.30; false-discovery-rate q < 0.05). The nicotine-related-aromatic/lignin association (ρ = 0.479) was almost entirely attributable to between-area covariance and approached zero after area adjustment, whereas selected nicotine-carbohydrate and phytol-glycosidic relationships contained larger area × position contributions. The framework therefore provides source-aware interpretation of pooled multimodal correlations; its principal limitations are dependence on structured sampling, tentative compound annotation from archived total-ion-current data and limited transfer to unseen geographic origins.
Related
- Geometry-Controlled Dynamic Tensiometry Resolves Intrinsic Surfactant Adsorption Kinetics
- Solid-State Host–Guest Chemistry of Cyanoborate-Bridged Cubane-Type Tetranuclear Ruthenium Complexes: Solvate Crystal Formation and Reversible Sorption of Volatile Organic Compounds
- New Insights into the Pharmacological Effects and Potential Health Benefits of Volatile Phenylpropenes
- Recombination-zone position, not charge balance, sets the efficiency ceiling of deep-blue TADF host OLEDs
- Direct separation of intra- and inter-molecular contributions in pulse dipolar EPR experiments
BibTeX
@article{li2026factoraware,
title = {{Factor-aware cross-modal mapping of GC-MS and FTIR resolves geographic and stalk-position contributions to volatile-matrix covariation in tobacco leaves from China.}},
author = {Deguo Li and Weican Chen and Chi Ma and Yang Gao and Hui Chen and Ren Zhang},
journal = {RSC advances},
year = {2026},
month = oct,
doi = {10.1039/d6ra08992e},
url = {https://europepmc.org/article/MED/42910033},
note = {CC BY-NC},
}
RIS
TY - JOUR
TI - Factor-aware cross-modal mapping of GC-MS and FTIR resolves geographic and stalk-position contributions to volatile-matrix covariation in tobacco leaves from China.
AU - Deguo Li
AU - Weican Chen
AU - Chi Ma
AU - Yang Gao
AU - Hui Chen
AU - Ren Zhang
PY - 2026
DA - 2026/10/21
JO - RSC advances
DO - 10.1039/d6ra08992e
UR - https://europepmc.org/article/MED/42910033
AB - Cross-platform correlations in heterogeneous biological materials can reflect shared experimental structure rather than intrinsic chemical coupling. Here, 90 flue-cured tobacco leaves from ten production areas and three stalk positions were analysed in a balanced 10 × 3 × 3 design using headspace solid-phase microextraction gas chromatography-mass spectrometry (HS-SPME-GC-MS) and Fourier transform infrared spectroscopy (FTIR). For a prespecified 8 × 8 cross-modal panel, pooled rank covariance was partitioned into production-area, stalk-position, area × position and within-cell residual components, with robustness assessed by design-preserving permutation, nested classification and leave-one-production-area-out validation. FTIR replicate consistency was high (median mean within-sample correlation ≈0.996). Both analytical platforms contained reproducible stalk-position information, and feature-level fusion achieved 84.4% internal accuracy, but this decreased to 42.2% when complete production areas were held out. Geographic origin dominated global analytical heterogeneity and 13 of 14 retained cross-modal associations (|ρ| ≥ 0.30; false-discovery-rate q < 0.05). The nicotine-related-aromatic/lignin association (ρ = 0.479) was almost entirely attributable to between-area covariance and approached zero after area adjustment, whereas selected nicotine-carbohydrate and phytol-glycosidic relationships contained larger area × position contributions. The framework therefore provides source-aware interpretation of pooled multimodal correlations; its principal limitations are dependence on structured sampling, tentative compound annotation from archived total-ion-current data and limited transfer to unseen geographic origins.
SN - 2046-2069
ER -