Development and validation of an interpretable prognostic model for bladder cancer based on lactylation associated genes using SHAP analysis

OBJECTIVE: The high recurrence rate and poor clinical outcomes of bladder cancer (BLCA) underscore the urgent need for novel biomarkers to guide precision medicine. Focusing on lactylation-a novel epigenetic modification-this study aimed to construct and validate a prognostic model for BLCA based on lactylation-related genes. METHODS: Transcriptomic and clinical data were integrated from the Gene Expression Omnibus (GEO, GSE13507) and The Cancer Genome Atlas (TCGA-BLCA) cohorts. Core prognostic genes were identified, and a risk scoring model was constructed utilizing a stepwise framework comprising univariate Cox regression, Least Absolute Shrinkage and Selection Operator (LASSO) regression, and multivariate Cox regression. The SHapley Additive exPlanations (SHAP) algorithm was employed for personalized, interpretable survival predictions. Subsequently, the associations between the risk model and the tumor immune microenvironment (TIME), tumor mutational burden (TMB), and drug sensitivity were comprehensively evaluated. RESULTS: A 4-gene risk model comprising ATAD3A, MKI67, VWF, and CCL2 was ultimately established. The model demonstrated robust predictive efficacy in both the training cohort (1-, 3-, and 5-year AUCs of 0.802, 0.818, and 0.856, respectively) and the validation cohort (AUCs of 0.636, 0.601, and 0.627, respectively). Multivariate analysis confirmed that the risk score served as an independent prognostic factor. The SHAP analysis facilitated precise interpretability, bridging global feature importance with individual mortality risk. Immune infiltration profiling revealed a significant enrichment of M2 macrophages coupled with high CCL2 expression in the high-risk group. Furthermore, combined survival analysis indicated that patients with a "low TMB and high-risk" profile exhibited the poorest clinical outcomes. CONCLUSION: To our knowledge, this study is the first to establish a 4-gene lactylation-related prognostic model for BLCA integrating a LASSO-SHAP framework. This model not only provides precise, visualizable predictions for individualized survival risk but also elucidates the lactylation-driven immunosuppressive microenvironment in BLCA, offering crucial data support for the development of personalized immunotherapeutic and targeted strategies.