Machine learning-guided structure-reactivity modeling of lignin depolymerization by high resolution mass spectrometry
Bioresource Technology · 2026-11 · 136245
DOI: 10.1016/j.biortech.2026.136245 ↗Kurz gefasst
Fachartikel in der Zeitschrift Bioresource Technology, 2026-11. Verfasst von Zhenqin Wang und 9 weiteren Autor:innen. Beteiligte Einrichtung: Department of Energy, Environmental and Chemical Engineering, McKelvey School of Engineering, Washington University in St. Louis, MO 63130, the Unite…. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Gordon and Betty Moore Foundation.
Abstract (Originalsprache)
Lignin molecular heterogeneity makes it difficult to understand and predict reactivity during depolymerization. Fourier transform ion cyclotron resonance mass spectrometry (FTICR-MS) characterized molecular changes during depolymerization of organosolv poplar lignin (OPL) in methanol (MeOH), with a copper porous metal oxide catalyst (CuPMO) in MeOH, and with CuPMO in a MeOH/dimethyl carbonate (DMC) mixture. FTICR-MS peaks were assigned to elemental formulas (i.e., CxHyOz), enabling calculation of molecular descriptors including mass, hydrogen-to-carbon ratio (H/C), oxygen-to-carbon ratio (O/C), and Kendrick mass defect (KMD). Van Krevelen classification grouped assigned formulas into broad chemical regions. Each assigned formula was compared with untreated OPL and classified as Removed, Not Removed, Produced, or Never Present, then modeled using extreme gradient boosting (XGBoost). SHapley Additive exPlanation (SHAP) analysis identified the molecular descriptors that most strongly predicted formula transformation. In the MeOH models, predicted formula outcomes were associated with a broader range of molecular descriptors, suggesting a less selective formula-level response. The CuPMO and CuPMO/DMC models showed distinct prediction patterns across oxygenated aromatic and lignin-like formulas. Under CuPMO/DMC, higher probabilities of Removed or Produced outcomes were concentrated in a narrower formula space, suggesting greater selectivity in the predicted formula-level response. The framework converts thousands of FTICR-MS elemental formulas into interpretable reaction-outcome predictions, providing a new approach for identifying non-obvious relationships between molecular composition and lignin reactivity.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Energy, Environmental and Chemical Engineering, McKelvey School of Engineering, Washington University in St. Louis, MO 63130, the United States of America.
- Department of Molecular Biosciences and Bioengineering, University of Hawai'i at Mānoa, Honolulu, HI 96822, the United States of America.
- Department of Neurosurgery and Institute for Informatics, Data Science, and Biostatistics, Washington University in St. Louis, MO 63130, the United States of America.
- Environmental Molecular Sciences Laboratory, Pacific Northwest National Laboratory, Richland, WA 99354, the United States of America.
- Department of Energy, Environmental and Chemical Engineering, McKelvey School of Engineering, Washington University in St. Louis, MO 63130, the United States of America. Electronic address: mfoston@wustl.edu.
Förderung
- Gordon and Betty Moore Foundation
Zitieren
APA
Wang, Z., Gao, Y., Roell, G., Wang, J., Meyer, J., Zhang, J., DiDonato, N., Nikola, T., Chu, R., & Foston, M. (2026). Machine learning-guided structure-reactivity modeling of lignin depolymerization by high resolution mass spectrometry. Bioresource Technology, 136245. https://doi.org/10.1016/j.biortech.2026.136245
Vancouver
Wang Z, Gao Y, Roell G, Wang J, Meyer J, Zhang J, et al. Machine learning-guided structure-reactivity modeling of lignin depolymerization by high resolution mass spectrometry. Bioresource Technology. 2026:136245. doi:10.1016/j.biortech.2026.136245
BibTeX
@article{Wang2026,
author = {Wang, Zhenqin and Gao, Yu and Roell, Garrett and Wang, Jing and Meyer, James and Zhang, Jialiang and DiDonato, Nicole and Nikola, Tolic and Chu, Rosalie and Foston, Marcus},
title = {Machine learning-guided structure-reactivity modeling of lignin depolymerization by high resolution mass spectrometry},
journal = {Bioresource Technology},
year = {2026},
pages = {136245},
doi = {10.1016/j.biortech.2026.136245}
}
EUVIMED