Abstract
Bladder cancer (BC) has a high recurrence rate and marked molecular heterogeneity, yet effective biomarkers and druggable targets remain scarce. Honghua Longdan (Gentiana rhodantha Franch., GR), a traditional Chinese medicine, exhibits antitumor activity, but its therapeutic targets and mechanisms in BC remain poorly defined. Here, differentially expressed genes (DEGs) in BC from TCGA-BLCA and GEO were integrated with weighted gene coexpression network analysis (WGCNA) to identify BC-related gene modules. Active GR-derived compounds were obtained from HERB v 2.0 and SymMap. Three machine-learning algorithms were employed to identify hub genes and build a neural-network diagnostic model. Functional enrichment, survival, and gene set enrichment analyses together with immunohistochemistry (IHC) were used to characterize key genes. Molecular docking and all-atom molecular dynamics simulations were performed to validate interactions between hub proteins and GR-derived compounds. Integrated analyses identified eight core drug-disease genes enriched in cell-cycle-related pathways. Six hub genes were consistently selected by the three algorithms, and the neural-network model showed excellent diagnostic performance. High CCNB1 expression was significantly associated with poorer overall survival (OS), and its expression is markedly upregulated in BC tissues. In silico simulations suggested that the GR-derived iridoid glycoside swertiamarin forms a stable complex with CCNB1. This study integrates network pharmacology, machine learning, and computational biophysics to identify CCNB1 as a biomarker with diagnostic and prognostic value in BC. The predicted interaction between swertiamarin and CCNB1 suggests that CCNB1 is a GR-mediated therapeutic target and supports the rational discovery of traditional Chinese medicine-derived therapeutics.