Recently, Chen Yuqian — Director of the AI4S Platform Center at the School of Information Engineering and Joint Researcher at the School of Chemical Biology and Biotechnology, Peking University Shenzhen Graduate School — was invited to publish a review focusing on the empowerment of artificial intelligence (AI) in traditional Chinese medicine (TCM) research. The review summarizes over two decades of efforts in establishing, maintaining, and updating TCMBank, the world’s largest TCM database, and explores the integrated applications of TCM and AI across multiple research fields, including herbal screening, new drug discovery, diagnostic and therapeutic principles, pharmacological mechanisms, and network pharmacology. It also details how AI reveals the active ingredients and mechanisms of action in complex TCM formulations through data mining, pattern recognition, and predictive analysis. The integration of AI and TCM not only helps people understand traditional TCM knowledge from new perspectives but also opens up new research methods and therapeutic strategies. This achievement was published inChemical Sciencein September 2024. The first authors are Song Zhilin, a doctoral student from the School of Chemical Biology and Biotechnology at Peking University, and Chen Guanxing, a doctoral student from the School of Intelligent Engineering at Sun Yat-sen University, with Chen Yuqian as the corresponding author. Other related works include publications inChemical Science(2023) andSignal Transduction and Targeted Therapy(STTT, IF=40.8) (2023).
TCM is a vital source of natural products. From 1981 to 2019, over 60% of FDA-approved small-molecule drugs were directly or indirectly derived from natural products. Additionally, TCM offers diverse treatment options tailored to patients’ specific needs and preferences. In recent years, advances in AI technology have unlocked enormous potential for TCM research. By accelerating drug discovery, optimizing formulations, and advancing TCM modernization, AI ensures that TCM evolves in step with the latest scientific progress while preserving its traditional roots. Currently, the integration of AI in clinical research at TCM hospitals has not only improved diagnostic accuracy but also promoted the development of personalized medicine, making TCM research more efficient and precise.
AI-Based Chemical Data Analysis of TCM Facilitates Component Identification, Drug Discovery, Personalized Therapy, and Elucidation of Pharmacological Effects, Driving the Modernization and Sustainable Development of TCM
Currently, network pharmacology research in TCM primarily focuses on validating TCM theories, lacking clear definitions of network design and optimization principles. Rational prescription design can be divided into top-down and bottom-up approaches. The top-down method designs new prescriptions based on existing ones, while the bottom-up method creates new prescriptions entirely based on disease networks without considering existing formulas. Chen Yuqian’s team combined multiple feasible approaches and first defined several common future top-down optimization models. Both top-down and bottom-up design methods use biological networks to establish correlations based on computational systems biology principles but differ in their consideration of existing prescriptions and TCM theories. Before TCM theories are quantitatively expressed, new formulations designed without reference to existing prescriptions are more likely to conflict with TCM principles. In recent years, TCM has gradually shifted toward the classification and organization of natural products. Against the backdrop of excessive human exploitation, many TCM species face extinction. Thus, Chen Yuqian’s team developed a method to calculate correlations between natural products and individual TCM species, using highly reliable models to quickly screen prescription components that can be replaced, optimized, or removed. The authors argue that the design of new prescription formulas will become a core research area, significantly impacting the sustainable development of TCM.
Overview of Network Design and Optimization Methods for TCM Prescriptions
In the field of TCM, understanding interactions between herbal medicines and conventional drugs is crucial for ensuring patient safety and treatment efficacy. The authors proposed a framework for comprehensive clinical research on TCM-western medicine interactions. This framework establishes an integrated clinical research system to thoroughly explore these interactions, as shown in Figure 3. The system is divided into three main modules: clinical research, integration system, and basic research, each focusing on different aspects of chemical analysis and data integration to predict and analyze drug incompatibilities.
The clinical research module primarily collects and analyzes clinical data to evaluate the pharmacokinetic and pharmacodynamic properties of herbal-drug combinations. Leveraging advanced chemical extraction and analytical techniques, this module aims to identify potential adverse interactions through rigorous clinical testing and chemical analysis. It also uses data mining and machine learning algorithms to process complex clinical data and identify patterns predicting drug interaction outcomes.
The core of the integrated approach is the integration system, which includes the HAZA@home platform. This system combines data from clinical and basic research to provide real-time health monitoring and predictive analysis for patients. It also uses an extensive network of databases and simulation tools to deliver personalized health assessments and drug interaction alerts, with AI adjusting recommendations based on individual patient data.
The basic research module supports the integration system through in-depth chemical analysis and computational modeling, predicting drug targets and simulating drug interactions. This includes using TCM and clinical drug databases, ADMET prediction models, and creating small-molecule databases. These tools are critical for understanding the biochemical mechanisms of drug interactions and developing strategies to prevent potential adverse reactions.
This proposal embodies the integration of clinical and basic scientific methods in TCM interaction research. By combining chemical analysis, real-time data monitoring, and AI-driven predictive modeling, the system not only enhances understanding of complex drug interactions but also pioneers new approaches to patient safety in the integration of TCM and drug therapy, setting benchmarks for future research and applications. Related research results were published inChem. Sci.2024, 15 (41), 16844–16886 (https://doi.org/10.1039/D4SC04107K).
A Proposal for a Comprehensive Automated Clinical Research System
In the modernization of TCM, identifying active ingredients in herbs and elucidating the mechanisms between active ingredients and targets are two key aspects. Constructing a comprehensive and highly reliable TCM database is therefore particularly important. Since its establishment in 2011, the team’s TCM Database@Taiwan has been widely used and cited, and was incorporated into the ZINC database. To further advance TCM research, the authors developed TCMBank — an upgraded version of TCM Database@Taiwan — using natural language processing to build knowledge graphs and molecular signaling pathways. TCMBank contains 9,192 herbs, 61,966 chemical components, 15,179 targets, and 32,529 diseases. Related research results were published inChem. Sci.2023, 14 (39), 10684–10701 (https://doi.org/10.1039/D3SC02139D).
Schematic Diagram of TCMBank’s Data Processing Framework and Objectives
In today’s medical field, combined TCM-western medicine therapy is increasingly common, but it also brings risks of drug interactions. To better predict and manage interactions between TCM and western medicine and reduce medical risks, this research developed multiple innovative models and databases using advanced AI technology and big data analysis, providing strong support for studying TCM-western medicine interactions.
The research team found that over 10% of patients need to take 5 drugs simultaneously, and 20% of elderly patients require at least 10 drugs. This phenomenon poses new challenges for drug interaction research. Based on pharmaceutical chemistry knowledge, the team proposed that the functional groups/chemical substructures of drugs determine their pharmacokinetic, pharmacodynamic properties, and incompatibility with TCM. Thus, they developed an innovative algorithm to predict interactions between multiple drugs by constructing a drug interaction network, where compounds serve as nodes and their causal relationships as edges. Nodes corresponding to all components in TCM form a subnetwork. By analyzing connections between subnetworks, the team can predict whether adverse reactions exist between TCM formulations or between TCM and western medicine.
In the future, AI-assisted models will integrate large language models, natural language processing, and text-mined knowledge graphs to develop a TCM-western medicine adverse reaction database, further improving prediction accuracy and practicality. The research team proposed two models — 3DGT-DDI and SA-DDI — on two real-world public drug-drug interaction (DDI) datasets, achieving state-of-the-art prediction performance. Based on these models, the team extended their predictions to TCM-western medicine adverse reactions. Using TCMBank’s big data, the team conducted unsupervised learning to predict such adverse reactions. It is assumed that if no adverse reactions occur between all components in TCM and a western drug, there is no mutually exclusive reaction. If one or more components in TCM react adversely with a western drug, there is a risk of adverse reactions. The team used AI-assisted DDI prediction models to obtain risk predictions for potential TCM-western medicine adverse reactions, with related results published inSignal Transduction and Targeted Therapy(https://doi.org/10.1038/s41392-023-01339-1).
Comprehensive Analysis of TCMBank, the World’s Largest Intelligent TCM Database
Chen Yuqian is the corresponding author of the three aforementioned papers. The research was supported by General Program and Key Program of the National Natural Science Foundation of China.
References
[1] Song, Z.; Chen, G.; Chen, C. Y.-C. AI Empowering Traditional Chinese Medicine? Chem. Sci. 2024, 15 (41), 16844–16886. https://doi.org/10.1039/D4SC04107K.
[2] Lv, Q.; Chen, G.; He, H.; Yang, Z.; Zhao, L.; Chen, H.-Y.; Chen, C. Y.-C. TCMBank: Bridges between the Largest Herbal Medicines, Chemical Ingredients, Target Proteins, and Associated Diseases with Intelligence Text Mining. Chem. Sci. 2023, 14 (39), 10684–10701. https://doi.org/10.1039/D3SC02139D.
[3] Lv, Q.; Chen, G.; He, H.; Yang, Z.; Zhao, L.; Zhang, K.; Chen, C. Y.-C. TCMBank-the Largest TCM Database Provides Deep Learning-Based Chinese-Western Medicine Exclusion Prediction. Signal Transduct. Target. Ther. 2023, 8 (1), 127. https://doi.org/10.1038/s41392-023-01339-1.
