Integrated metabolomic and pharmacogenomic analysis for prediction of valproic acid efficacy and hepatotoxicity in pediatric epilepsy.

PURPOSE: Valproic acid (VPA) is one of the most commonly prescribed broad-spectrum antiseizure medications for pediatric epilepsy. However, substantial interindividual variability exists in therapeutic efficacy and hepatotoxicity during VPA treatment, and reliable biomarkers for individualized prediction remain limited. This study aimed to identify metabolomic biomarkers associated with VPA therapeutic response and hepatotoxicity in pediatric patients with epilepsy and to construct integrated metabolomic-genomic prediction models for VPA efficacy and hepatotoxicity. METHODS: A total of 194 pediatric epilepsy patients receiving VPA monotherapy were enrolled in this study. Untargeted metabolomics analysis was performed using LC-MS/MS to identify endogenous metabolites associated with VPA therapeutic efficacy and adverse reactions. Differential metabolites were analyzed using principal component analysis, volcano plot analysis, hierarchical clustering, and metabolite enrichment analysis. Candidate SNPs identified in our previous pharmacogenomic study were further integrated with metabolomic features for multi-omics modeling analysis. Logistic regression analysis was used to construct prediction models, and model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, confusion matrix analysis, and bootstrap internal validation. RESULTS: Untargeted metabolomics analysis identified multiple differential metabolites associated with VPA therapeutic response and hepatotoxicity. Differential metabolites related to VPA efficacy were mainly enriched in pyrimidine metabolism, vitamin B6 metabolism, pantothenate and CoA biosynthesis, and beta-alanine metabolism pathways. Differential metabolites associated with hepatotoxicity were primarily enriched in arginine biosynthesis, pyrimidine metabolism, purine metabolism, and steroid hormone biosynthesis pathways. Integrated metabolomic-genomic prediction models demonstrated good predictive performance. For VPA therapeutic response, the combined model achieved an AUC of 0.830 in the training set and 0.817 in the testing set. For VPA-related hepatotoxicity, the model achieved an AUC of 0.816 in the training set and 0.791 in the testing set. Calibration curve and confusion matrix analyses further demonstrated acceptable robustness and predictive capability of the models. CONCLUSION: This study identified multiple endogenous metabolites and metabolic pathways associated with VPA therapeutic efficacy and hepatotoxicity in pediatric epilepsy patients. Integration of metabolomic and pharmacogenomic features may improve individualized risk stratification of VPA treatment outcomes; however, prospective multicenter external validation and clinically standardized metabolite assays are required before these models can be implemented in routine pediatric epilepsy care.