Structured Dataset of Traditional Efficacies for Chinese Herbal Medicines

Description: This dataset provides a structured and computable resource of traditional therapeutic effects (efficacies) for Chinese herbal medicines. Its main purpose is to transform classical TCM efficacy descriptions from natural-language expressions into standardized herb-efficacy relations that can be directly used for knowledge graph construction, formula efficacy prediction, herb-symptom association mining, and AI-based reasoning over traditional therapeutic knowledge.

The dataset links Chinese herbs to explicit and rule-expanded efficacy nodes. Explicit efficacy nodes are directly structured from authoritative TCM textual sources, while expanded efficacy nodes are inferred through rule-based semantic reasoning over a TCM knowledge graph. In this way, the dataset not only records what therapeutic effects are explicitly stated for each herb, but also provides computable extensions of those effects to related symptoms, patterns, sub-patterns, and pathogenesis-related efficacy concepts.

Version 2.1 Release Notice: This is Version 2.1 of the Structured Dataset of Traditional Efficacies for Chinese Herbal Medicines. Version 2.1 corrects the implementation and output of Rule C while preserving the explicit herb-efficacy relations and the Rule A, Rule B, and Rule D results from Version 2.0. The earlier Rule C implementation included an overly broad range of symptom-related node types in reverse reasoning. The corrected Rule C restricts reverse reasoning to Symptom (pattern) and Symptom (disease) nodes with explicitly linked Pattern name or Disease name upstream nodes. Accordingly, the Rule C expanded relation file and Rule C evidence file have been replaced in Version 2.1, while the other data files remain unchanged.

Upstream Knowledge Graph Dependency Note: The rule-based expanded efficacy relations in this dataset are based on Version 2.0 of the related TCM knowledge graph. Version 2.1 uses the same upstream knowledge graph as that used for Version 2.0 of this dataset; only the implementation and output of Rule C have been corrected in this release. The related knowledge graph is available at: https://doi.org/10.5281/zenodo.20061549

Dataset Content: The dataset consists of five herb-efficacy relation files and four rule-specific evidence files.

Herb-Efficacy Relation Files:

1. explicit_herb_efficacy_relations_v2.csv: Directly curated herb-to-efficacy relations for explicit efficacy nodes.

2. expanded_herb_efficacy_relations_ruleA_v2.csv: Herb-to-efficacy relations expanded through Rule A.

3. expanded_herb_efficacy_relations_ruleB_v2.csv: Herb-to-efficacy relations expanded through Rule B.

4. expanded_herb_efficacy_relations_ruleC_v2.1.csv: Herb-to-efficacy relations expanded through the corrected Rule C.

5. expanded_herb_efficacy_relations_ruleD_v2.csv: Herb-to-efficacy relations expanded through Rule D.

Evidence Files:

1. herb_efficacy_expansion_evidence_ruleA_v2.csv: Rule A evidence table documenting the reasoning paths for Rule A expansions.

2. herb_efficacy_expansion_evidence_ruleB_v2.csv: Rule B evidence table documenting the reasoning paths for Rule B expansions.

3. herb_efficacy_expansion_evidence_ruleC_v2.1.csv: Rule C evidence table documenting the reasoning paths for the corrected Rule C expansions.

4. herb_efficacy_expansion_evidence_ruleD_v2.csv: Rule D evidence table documenting the reasoning paths for Rule D expansions.

Key Statistics: The five herb-efficacy relation files contain 28,820 relation records in total, covering 483 unique herbs and 2,519 unique target efficacy nodes. After deduplication by source_id and target_id, the dataset contains 28,792 unique herb-efficacy pairs. The four rule-specific evidence files contain 21,268 evidence records in total.

Data Structure: The relation files are provided in CSV format. Most CSV files in this release are UTF-8 encoded; the retained explicit_herb_efficacy_relations_v2.csv file from Version 2.0 is encoded in GB18030. The shared fields include source_id, source_ENGLISHNAME, source_LABEL, target_id, target_ENGLISHNAME, target_LABEL, TYPE, and Efficacy_Node_Type. The TYPE field represents the directed has_effect relation between herbs and efficacy nodes. The Efficacy_Node_Type field indicates whether the entry belongs to explicit efficacy nodes or rule-based expanded nodes.

The evidence files provide rule-specific provenance information for expanded efficacy relations. Because Rules A-D involve different semantic reasoning paths, the evidence files contain different intermediate-node fields. The number of evidence records may differ from the number of final expanded herb-efficacy relations because a single herb-efficacy relation may be supported by multiple reasoning paths or intermediate evidence nodes.

Potential Applications: This resource is intended for researchers in TCM informatics, biomedical knowledge graphs, and AI-based prescription analysis. It supports formula efficacy prediction, herb-efficacy relation mining, symptom-herb association analysis, knowledge graph enrichment, semantic reasoning, and computational analysis of traditional therapeutic knowledge.

Related Publication:

Yuanbai L, Fangzhou L, Yihao L, Yu D, Meng L, Qin Q, Yang Y, Hongming M. A Knowledge Graph-Driven Hypergeometric Efficacy Prediction Model for Classical Traditional Chinese Herbal Formulas. Methods Inf Med. 2026 Apr 7. doi: 10.1055/a-2841-4549. Epub ahead of print. PMID: 41895302.

Contact:
For questions, please contact:
LI Yuanbai: liyuanbai126@126.com