Combining UHPLC profiling and random walk network-based in vitro analysis to predict herb-induced liver injury

Herbal medicines are widely used, yet their hepatotoxic potential remains underexplored in predictive toxicology. UHPLC-based compound profiling was combined with a Random Walk with Restart (RWR) network approach using herb compound target associations filtered by P-value and Z-score thresholds. Predictions were evaluated in HepG2 cells using microscopy-based phenotypic assessment, mitochondrial membrane potential measurement, ALT and AST activities in culture supernatants, transcriptomic profiling by RNA sequencing with enrichment analysis, and qRT-PCR as supportive validation. RWR prioritized apoptosis, oxidative stress, and inflammatory pathways for Camellia sinensis, Piper longum, Atractylodes lancea, Angelica gigas, Xanthium sibiricum, and Cynanchum wilfordii, whereas Astragalus membranaceus showed limited enrichment. Consistent with these predictions, the six prioritized extracts induced injury-associated morphological changes, loss of mitochondrial membrane potential, and increased ALT and AST release, while A. membranaceus showed minimal changes. RNA sequencing showed broad transcriptomic perturbations and clustering of the predicted hepatotoxic extracts with coordinated changes across hepatotoxicity-relevant gene categories. Overall, this framework supports scalable preclinical screening of herbal products by linking computational pathway prioritization with experimental validation, and broader herb–compound–target coverage with expanded toxicological datasets may further improve predictive performance for safety assessment.