Bridging traditional medicine with modern technology: “Omics”-guided herbal pharmacovigilance for integrative liver health

Traditional herbal medicines (THMs) are widely used globally, but herb‑induced liver injury (HILI) remains underreported and poorly understood. Current pharmacovigilance systems fail to address product heterogeneity, nomenclature confusion, and regulatory gaps. This review proposes an integrated framework combining "omics" (genomics, transcriptomics, proteomics, metabolomics) with artificial intelligence (AI) to transform herbal safety into a personalized, evidence‑based system. By identifying genetic susceptibilities, metabolic biomarkers, and phytochemical fingerprints, this approach enables risk stratification, early signal detection, and improved causality assessment, bridging traditional practice with modern precision medicine.

Introduction
About 80% of the global population uses THMs, yet safety data are limited. The liver is vulnerable to reactive metabolites from herbal extracts, driving a rising HILI burden. Current pharmacovigilance (PV) systems, designed for single‑drug molecules, cannot handle complex botanical mixtures. This review advocates an "omics"‑AI paradigm to shift from reactive monitoring to proactive molecular surveillance.

PV Challenges in THM

• Complexity: Polyherbal products contain hundreds of bioactive metabolites, hindering causative agent identification.

• Heterogeneity: Chemical profiles vary with genetics, climate, and processing, causing inconsistent pharmacokinetics.

• Nomenclature confusion: Lack of standardized botanical names leads to attribution errors.

• Underreporting: >76% of users self‑medicate; ADR reporting rates for herbs are only 0.03–29.84%.

• Regulatory gaps: Many herbs are classified as supplements, bypassing GMP.

• Contamination: 30.5% of herbal products contain over‑limit heavy metals (Pb, Cd, As, Hg, Cu).

"Omics" Applications
Multi‑omics detects biochemical damage within 5–10 hours, before ALT/AST rise.

• Genomics: DNA barcoding prevents toxic substitution (e.g., Gynura japonica misidentification caused >2,000 liver injury cases). CYP450 polymorphisms and HLA variants stratify host risk.

• Transcriptomics: Detects stress pathways (e.g., Nrf2 downregulation) before symptoms; interindividual response varies 864‑fold, explaining idiosyncratic toxicity.

• Proteomics: Identifies injury biomarkers (GDF‑15 for mitochondrial stress, inter‑alpha‑trypsin inhibitor).

• Metabolomics: Functional readout of metabolic disruption (e.g., purine and bile acid alterations). Multi‑omics integration has revealed 575 altered proteins and 14 key indicators in Fructus Psoraleae hepatotoxicity.

AI‑Powered PV

• Predictive toxicology: ML models achieve 76–84% accuracy; multi‑omics classifiers reach AUC 0.81–0.87.

• Automated signal detection: NLP and ML extract ADEs from reports and EHRs; gradient boosting detects signals 6 months earlier than manual analysis.

• Quality control: Chemometric modeling detects adulteration and geographical variation.

• HDI prediction: Computational models predict herb‑drug interactions.

Operational Pipeline (5 stages)

1. High‑dimensional data acquisition (genomics, transcriptomics, phytochemistry).

2. AI‑based data harmonization and normalization.

3. Algorithmic synthesis (graph neural networks, transformers, Bayesian neural networks).

4. Signal benchmarking against RUCAM.

5. Clinical validation.

Clinical Scenarios

• CYP450 genotyping before high‑risk herbal prescription.

• Transformer models detect rising GDF‑15 weeks before jaundice.

• AI predicts herb‑drug interference in multi‑drug regimens.

Ethical & Technical Challenges

• Need for tiered consent, differential privacy, and harmonized regulations (WHO, FDA, EMA).

• Limitations: bulk tissue analysis lacks single‑cell resolution; no universal validation protocols; small sample sizes limit generalizability; deep learning "black box" hinders trust; regulatory lag for AI‑generated signals.

Future Directions

• Single‑cell and spatial omics for high‑resolution injury mapping.

• Blockchain for data security and product traceability.

• Open data platforms (e.g., Omics Discovery Index).

• Global regulatory harmonization.

Conclusions
Integrating multi‑omics with AI transforms herbal safety from reactive to proactive, personalized medicine. By identifying genetic risks, metabolic biomarkers, and phytochemical fingerprints, this framework addresses underreporting, causality ambiguity, and quality failures. AI‑driven fingerprinting ensures product integrity and predicts hepatotoxicity in silico. This synergy harmonizes ancient wisdom with modern precision, providing a scientific foundation for safer integrative liver health worldwide.

Full text:

https://www.xiahepublishing.com/2835-6357/FIM-2025-00059

The study was recently published in the Future Integrative Medicine.

Future Integrative Medicine (FIM) is the official scientific journal of the Capital Medical University. It is a prominent new journal that promotes future innovation in medicine.It publishes both basic and clinical research, including but not limited to randomized controlled trials, intervention studies, cohort studies, observational studies, qualitative and mixed method studies, animal studies, and systematic reviews.

Follow us on X: @xiahepublishing

Follow us on LinkedIn:  Xia & He Publishing Inc.