Multi-omics and machine learning based on clinical samples: revealing the toxicity characteristics in adverse reactions to Evodia rutaecarpa
8 affiliations
- 1Chinese Medicine Germplasm Resources Innovation and Effective Uses Key Laboratory of Sichuan Province, School of Pharmacy, Chengdu University of Traditional Chinese Medicine
- 2Key Laboratory of Standardization of Chinese Medicine (Chengdu University of Traditional Chinese Medicine), Ministry of Education
- 3Department of Pharmacy, Chinese PLA General Hospital
- 4Lung Disease Diagnosis and Treatment Center, National Medical Center, The First Affiliated Hospital of Henan University of Chinese Medicine
- 5Collaborative Innovation Center for Chinese Medicine and Respiratory Diseases Co-constructed by Henan Province & Education Ministry of P.R. China / Henan Key Laboratory of Chinese Medicine for Respiratory Diseases, Henan University of Chinese Medicine
- 6The First Clinical Medical College, Henan University of Chinese Medicine
- 7College of Medicine and Life Sciences, Chengdu University of Traditional Chinese Medicine
- 8Hospital of Chengdu University of Traditional Chinese Medicine, School of Clinical Medicine, Chengdu University of Traditional Chinese Medicine
Abstract
Background: Evodia rutaecarpa (EVR) is a widely used low-toxicity traditional Chinese medicine in clinical practice. Its adverse reactions are often subtle with prominent inter-individual variability, and the associated risks are frequently underestimated. This study conducted multi-omics analysis of the metabolic and microbiological characteristics of its related adverse reactions, and constructed a prediction model to provide a reference for individualized safe medication.
Methods: 86 patients with gastric diseases receiving EVR-containing compound formulas were enrolled and divided into a control group (n = 50) and an adverse reaction group (n = 36) based on the presence or absence of adverse reactions. Serum, urine, and saliva samples were collected for untargeted metabolomics and 16S rRNA sequencing, respectively. Seven machine learning algorithms (logistic regression, random forest, LightGBM, etc. ) were employed to construct predictive models in combination with LOOCV, screen key features, and derive a multi-omics joint prediction formula.
Results: Clinically, adverse events related to EVR predominantly presented as upper-gastrointestinal symptoms. Omics profiling of patient samples identifies tentative correlative trends between EVR-associated adverse events and systemic changes including disturbed amino acid metabolism, shifted neuroactive molecule profiles, altered bile secretion signatures and oral microbiota dysbiosis. Machine learning screening yields multiple model-associated classification features from multi-omics datasets. Serum PE (P-18:1/22:0), hydroxyisocaproic acid and ureidoisobutyric acid show statistical trends linked to lower model prediction outputs for adverse events. Urinary 1, 3-Dihydro-(2H)-indol-2-one and MG (0:0/20:4/0:0) correlate with reduced predictive values, whereas arachidoyl ethanolamide associates with elevated model prediction scores. Higher relative abundances of Granulicatella , Alysiella , Leptotrichia hongkongensis and Streptococcus in saliva correspond to higher predictive probabilities from the classification model. An AUC-weighted multi-omics joint formula is established to calculate individual predictive likelihood of EVR-related adverse reactions. Scores ≥ 0.597 indicate higher model-derived predictive tendency, while scores below 0.597 indicate lower predictive possibility.
Conclusion: The ten omics-derived discriminative features and the multi-omics joint prediction formula developed in this work enable quantitative estimation of individual-level risk for EVR-related adverse reactions. They offer preliminary references for safety assessment, risk stratification, and personalized application of low-toxicity traditional Chinese medicines.
Related
- Clinical characteristics of cadonilimab-associated infusion reactions: a severity-stratified analysis of 52 cases and preliminary management framework
- Revealing the toxicological effects and mechanisms of tris(1-chloro-2-propyl) phosphate exposure on osteoarthritis based on network toxicology, machine learning, and molecular docking
- Clinical Recognition, Safe Triage, and Pharmacovigilance Reporting Readiness for Suspected Serious Adverse Drug Reactions in Community Pharmacies: A Covert Simulated-Patient Study in Central Jordan
- Clinical, Pathological, and Molecular Characteristics in Colorectal Cancer
- Age-related clinical characteristics of respiratory syncytial virus infection in hospitalized children: a retrospective study
BibTeX
@article{tan2026multiomics,
title = {{Multi-omics and machine learning based on clinical samples: revealing the toxicity characteristics in adverse reactions to Evodia rutaecarpa}},
author = {Xiyue Tan and Jin Xie and Xiaochuan Guo and Lv Wu and Senpan Shi and Lingping Fu and Wenwen Zhang and Lingshun Zeng and Xiao Ma and Jinhao Zeng and Yanling Zhao},
journal = {Frontiers in Pharmacology},
year = {2026},
month = oct,
doi = {10.3389/fphar.2026.1901889},
url = {https://www.frontiersin.org/articles/10.3389/fphar.2026.1901889/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Multi-omics and machine learning based on clinical samples: revealing the toxicity characteristics in adverse reactions to Evodia rutaecarpa
AU - Xiyue Tan
AU - Jin Xie
AU - Xiaochuan Guo
AU - Lv Wu
AU - Senpan Shi
AU - Lingping Fu
AU - Wenwen Zhang
AU - Lingshun Zeng
AU - Xiao Ma
AU - Jinhao Zeng
AU - Yanling Zhao
PY - 2026
DA - 2026/10/19
JO - Frontiers in Pharmacology
DO - 10.3389/fphar.2026.1901889
UR - https://www.frontiersin.org/articles/10.3389/fphar.2026.1901889/full
AB - Background: Evodia rutaecarpa (EVR) is a widely used low-toxicity traditional Chinese medicine in clinical practice. Its adverse reactions are often subtle with prominent inter-individual variability, and the associated risks are frequently underestimated. This study conducted multi-omics analysis of the metabolic and microbiological characteristics of its related adverse reactions, and constructed a prediction model to provide a reference for individualized safe medication. Methods: 86 patients with gastric diseases receiving EVR-containing compound formulas were enrolled and divided into a control group (n = 50) and an adverse reaction group (n = 36) based on the presence or absence of adverse reactions. Serum, urine, and saliva samples were collected for untargeted metabolomics and 16S rRNA sequencing, respectively. Seven machine learning algorithms (logistic regression, random forest, LightGBM, etc. ) were employed to construct predictive models in combination with LOOCV, screen key features, and derive a multi-omics joint prediction formula. Results: Clinically, adverse events related to EVR predominantly presented as upper-gastrointestinal symptoms. Omics profiling of patient samples identifies tentative correlative trends between EVR-associated adverse events and systemic changes including disturbed amino acid metabolism, shifted neuroactive molecule profiles, altered bile secretion signatures and oral microbiota dysbiosis. Machine learning screening yields multiple model-associated classification features from multi-omics datasets. Serum PE (P-18:1/22:0), hydroxyisocaproic acid and ureidoisobutyric acid show statistical trends linked to lower model prediction outputs for adverse events. Urinary 1, 3-Dihydro-(2H)-indol-2-one and MG (0:0/20:4/0:0) correlate with reduced predictive values, whereas arachidoyl ethanolamide associates with elevated model prediction scores. Higher relative abundances of Granulicatella , Alysiella , Leptotrichia hongkongensis and Streptococcus in saliva correspond to higher predictive probabilities from the classification model. An AUC-weighted multi-omics joint formula is established to calculate individual predictive likelihood of EVR-related adverse reactions. Scores ≥ 0.597 indicate higher model-derived predictive tendency, while scores below 0.597 indicate lower predictive possibility. Conclusion: The ten omics-derived discriminative features and the multi-omics joint prediction formula developed in this work enable quantitative estimation of individual-level risk for EVR-related adverse reactions. They offer preliminary references for safety assessment, risk stratification, and personalized application of low-toxicity traditional Chinese medicines.
SN - 1663-9812
ER -