Machine learning-driven construction and validation of an intra-tumoral heterogeneity-associated prognostic model for bladder urothelial carcinoma.

Abstract

Intra-tumoral heterogeneity (ITH) plays a crucial role in tumor progression and prognosis. This study aimed to construct a prognostic model for bladder urothelial carcinoma (BLCA) based on ITH-related genes. Transcriptomic and clinical data from multiple public cohorts were collected and processed. Hub genes associated with ITH were identified using Weighted Gene Co-expression Network Analysis. A 14-gene prognostic signature was developed using a combined LASSO and Random Survival Forest algorithm. The model demonstrated strong predictive performance, with high concordance index values and favorable time-dependent ROC curves in the training set and five independent validation cohorts. Furthermore, single-cell RNA sequencing analysis confirmed that several model genes were significantly overexpressed in BLCA samples compared to controls and showed distinct expression patterns across different cell types. These findings highlight the prognostic relevance of ITH-related genes and support the application of the proposed model in improving outcome prediction and guiding personalized therapy for BLCA patients.