Integrating transcriptomics, network pharmacology, and GraphBAN neural networks to identify biomarkers and regulatory mechanisms of classical traditional Chinese medicine formulations in ulcerative colitis.

Ulcerative colitis (UC) poses substantial therapeutic challenges with limited treatment options. Although traditional Chinese medicine (TCM) shows clinical efficacy, the underlying mechanisms of many formulations remain unclear. This study aimed to identify biomarkers and regulatory mechanisms of classical TCM formulations in UC treatment. We applied an integrated strategy combining transcriptomics, network pharmacology, GraphBAN neural networks, and molecular simulations. Differentially expressed genes (DEGs) in UC were identified from the GSE75214 dataset (97 controls and 11 patients with UC). The Traditional Chinese Medicine Systems Pharmacology and BATMAN-TCM databases were used to analyze 10 classical TCM formulations. Active constituents were screened using thresholds including oral bioavailability ≥ 30%, drug-likeness ≥ 0.18, and ADMET properties. GraphBAN predicted constituent-target interactions, which were subsequently validated through molecular docking and 20-ns molecular dynamics simulations. The analysis identified 3050 UC DEGs (1712 upregulated and 1338 downregulated) enriched in cytokine regulation, PI3K-Akt, and MAPK pathways. Among 797 constituents and 372 targets across 36 herbs, 7 key active constituents from 6 herbs were identified, with PPARG and PTGS2 recognized as core biomarkers and 4 herbs acting via 4 active constituents. Molecular docking demonstrated potential binding capability (Vina score <  - 5), further supported by molecular dynamics simulations showing minimal root mean square deviation fluctuations and stable hydrogen bonding. This study establishes a multi-omics framework identifying PPARG and PTGS2 as potential UC biomarkers and clarifying TCM-mediated inflammatory pathway modulation, providing a foundation for precision diagnosis and drug development.