TCMHTI: a Transformer-based herb-target interaction prediction model for Qingfu Juanbi Decoction in rheumatoid arthritis
10.1016/j.dcmed.2025.05.007
- VernacularTitle:TCMHTI：基于 Transformer 的青附蠲痹汤治疗类风湿关节炎的 中药-靶点相互作用预测模型
- Author:
Zhenzhong LIANG
;
Changsong DING
- Publication Type:Journal Article
- Keywords:
Transformer;
Qingfu Juanbi Decoction;
Rheumatoid arthritis;
Deep learning;
Network pharmacology
- From:
Digital Chinese Medicine
2025;8(2):206-218
- CountryChina
- Language:English
-
Abstract:
[Objective] :To predict the potential targets of Qingfu Juanbi Decoction (青附蠲痹汤, QFJBD) in treating rheumatoid arthritis (RA) using an improved Transformer model and investigate the network pharmacological mechanisms underlying QFJBD’s therapeutic effects on RA.
Methods:First, a traditional Chinese medicine herb-target interaction (TCMHTI) model was constructed to predict herb-target interactions based on Transformer improvement. The performance of the TCMHTI model was evaluated against baseline models using three metrics: area under the receiver operating characteristic curve (AUC), precision-recall curve (PRC), and accuracy. Subsequently, a protein-protein interaction (PPI) network was built based on the predicted targets, with core targets identified as the top nine nodes ranked by degree values. Gene Ontology (GO) functional and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using the targets predicted by TCMHTI and the targets identified through network pharmacology method for comparison. Then, the results were compared. Finally, the core targets predicted by TCMHTI were validated through molecular docking and literature review.
Results:The TCMHTI model achieved an AUC of 0.883, PRC of 0.849, and accuracy of 0.818, predicting 49 potential targets for QFJBD in RA treatment. Nine core targets were identified: tumor necrosis factor (TNF)-α, interleukin (IL)-1β, IL-6, IL-10, IL-17A, cluster of differentiation 40 (CD40), cytotoxic T-lymphocyte-associated protein 4 (CTLA4), IL-4, and signal transducer and activator of transcription 3 (STAT3). The enrichment analysis demonstrated that the TCMHTI model predicted 49 targets and enriched more pathways directly associated with RA, whereas classical network pharmacology identified 64 targets but enriched pathways showing weaker relevance to RA. Molecular docking demonstrated that the active molecules in QFJBD exhibit favorable binding energy with RA targets, while literature research further revealed that QFJBD can treat RA through 9 core targets.
Conclusion:The TCMHTI model demonstrated greater accuracy than traditional network pharmacology methods, suggesting QFJBD exerts therapeutic effects on RA by regulating targets like TNF-α, IL-1β, and IL-6, as well as multiple signaling pathways. This study provides a novel framework for bridging traditional herbal knowledge with precision medicine, offering actionable insights for developing targeted TCM therapies against diseases.
- Full text:2025071323155469590liangzhenzhong.pdf
