Herbal knowledge and artificial intelligence may unlock smarter disease inference
image: Figure 1. Flowchart of the MediHerb model. (A) Input stage: multi-modal encoders represent the main ingredient of each herb using chemical, structural, physiological, and textual features; (B) Fusion stage: five modalities are integrated via attention and aggregated across herbs in a prescription; (C) Prediction stage: fused features are processed by transformer blocks and a sigmoid layer to produce disease probabilities, optimized with an asymmetric loss for imbalanced multi-label data. SMILES: Simplified Molecular Input Line Entry System; Bi-LSTM: bidirectional long short-term memory; FC: fully connected; ECFP: extended connectivity fingerprint; MACCS: Molecular Access System; ERG: extended reduced graph; MHA: multi-head attention.
Credit: © 2025 Xiaoyi Liu*,Fu Chen, Junnan Pan, Chengwei Ai, Fei Guo*, Jijun Tang*. This is an Open Access article licensed under a Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, sharing, adaptation, distribution and reproduction in any medium or format, for any purpose, even commercially, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Researchers publishing in Computational Biomedicine have introduced a novel multi-modal artificial intelligence framework, "MediHerb," designed to improve disease inference through Traditional Chinese Medicine (TCM) knowledge integration. The study highlights how combining biomedical data, molecular information, and herbal prescription knowledge may advance interpretable AI-driven healthcare models.
As artificial intelligence continues to reshape biomedical research, one major challenge in Traditional Chinese Medicine remains the integration of highly complex prescription data with modern computational approaches. In a recent research article titled "MediHerb: A multi-modal enhanced framework for disease inference via herbal knowledge", researchers Xiaoyi Liu, Fei Guo and Jijun Tang propose a knowledge-enhanced framework capable of uncovering hidden relationships between herbs and diseases.
Bridging Molecular Biology and Traditional Medicine
Traditional Chinese Medicine prescriptions contain rich yet highly interconnected information, including molecular structures, physicochemical properties, herbal compatibility, and clinical symptom descriptions. However, existing computational approaches often struggle to capture the deep semantic relationships embedded within these heterogeneous datasets.
To address this limitation, the researchers developed MediHerb, a multi-modal framework that integrates five complementary information sources:
- molecular sequences;
- chemical fingerprints;
- physicochemical properties;
- graphical prescription representations;
- textual descriptions of TCM prescriptions.
Using an attention-based fusion mechanism, MediHerb aligns biological, herbal, and diagnostic information within a shared latent space, enabling multi-granularity reasoning across different biomedical layers.
Improved Disease Inference Through Multi-Modal Learning
Experimental benchmarking demonstrated that MediHerb substantially outperformed existing baseline methods in herb–disease association prediction tasks. Beyond achieving higher predictive accuracy, the framework also revealed biologically meaningful embedding patterns and attention mechanisms that may help explain the pharmacological basis underlying herb–disease relationships.
The authors emphasize that interpretability is a critical advantage of the system. Rather than functioning solely as a “black-box” prediction tool, MediHerb provides insights into how molecular and herbal features contribute to disease inference, potentially supporting future mechanistic studies in TCM research.
Toward Intelligent and Interpretable TCM Systems
To improve accessibility and usability, the research team also developed a lightweight graphical interface that allows users to interact with both the model and publicly available datasets. This design may facilitate broader adoption among biomedical researchers and clinicians exploring AI-assisted TCM analysis.
The study suggests that knowledge-enhanced multi-modal fusion could serve as an important bridge connecting molecular biology, herbal medicine, and clinical semantics. By integrating diverse biomedical signals into a unified framework, MediHerb offers a more holistic and interpretable approach to understanding complex TCM prescriptions.
AI-Driven Biomedical Research
The work reflects a growing trend in computational biomedicine, where artificial intelligence, graph learning, and multi-modal data integration are increasingly being used to solve long-standing biomedical challenges. Published in Computational Biomedicine, the study demonstrates how AI technologies may accelerate the modernization of Traditional Chinese Medicine and contribute to the future development of intelligent healthcare systems.