Bladder cancer diagnostic and prognostic models from DNA methylation by multi algorithm machine learning.

This study aimed to identify core DNA methylation sites linked to urothelial bladder cancer (UBC) and to develop prognostic and early diagnostic models using multi-sample datasets from tissue, blood, and urine. By applying six machine learning algorithms, we constructed a diagnostic signature based on four DNA methylation sites and a prognostic signature involving ten sites. The diagnostic model showed high accuracy in detecting bladder cancer in both tissue and urine samples, while the prognostic model effectively predicted survival outcomes. Further analyses revealed associations with survival, enriched pathways, immune infiltration, genetic mutations, and responses to immunotherapy and chemotherapy. Cellular experiments, including Q-PCR, WB, Co-IP, and ChIP, demonstrated that C1QTNFNF6 significantly influences UBC prognosis and that TET2 promotes demethylation of C1QTNFNF6, elevating its expression and accelerating tumor progression. These findings present eleven key methylation sites with strong clinical applicability for diagnosis and prognosis in UBC, and reveal an epigenetic mechanism involving TET2 and C1QTNFNF6 that drives disease development.