Machine learning framework for multi-regulatory safety assessment of traditional Chinese medicine compounds

Introduction: Traditional Chinese Medicine faces unprecedented regulatory challenges due to multi-component complexity incompatible with conventional single-molecule assessment frameworks, necessitating innovative computational approaches for cross-jurisdictional safety and efficacy evaluation. Methods: We developed an intelligent AI-driven multi-regulatory framework integrating 127 bioactive compounds from 24 herb species across systematic multi-database mining (TCMSP, TCMID, PubChem, ChEMBL). The agency-specific scoring algorithms modeled FDA, EMA, and CFDA regulatory priorities through the mathematical weighting systems. The machine learning models employed the Random Forest classification for toxicity prediction and Gradient Boosting regression for quality assessment; utilizes 20 molecular descriptors and ADMET parameters. Network pharmacology analysis revealed herb species interactions through centrality measures and community detection algorithms. Results: The cross-regulatory assessment demonstrated the substantial inter-agency variations (CFDA: 95.43±6.39; FDA: 89.96±10.70; EMA: 80.57±13.95). The AI models achieved a robust performance with the quality prediction R²=0.746 and the toxicity classification accuracy of 0.497±0.052. The feature importance analysis identified oral bioavailability as the dominant predictor (0.38), whereas the network analysis showed hub species with the superior regulatory compliance. Risk-benefit profiling positioned anti-inflammatory and immunomodulatory compounds optimally for regulatory approval. Conclusions: This intelligent, multi-regulatory framework represents a paradigm shift toward precision regulatory science, enhancing traditional medicine safety assessment.