Multi-Omics Profiling and Machine Learning Uncover High-Stemness Bladder Cancer Risk Genes

Researchers integrating single-cell RNA sequencing, spatial transcriptomics, bulk RNA sequencing, and machine-learning models have characterized highly stem-like malignant epithelial cells in bladder cancer. Bladder cancer displays marked molecular heterogeneity and carries substantial risks of recurrence, disease progression, and treatment resistance, which highlights the need for new biomarkers defined at the level of specific cellular states. In the investigation, malignant epithelial cells were identified from single-cell transcriptomic datasets using CopyKAT and subsequently stratified into LowStem, IMStem, and HighStem groups based on within-patient CytoTRACE scores.

Cell-to-cell communication analyses and pathway evaluations revealed extensive inferred interactions connecting HighStem malignant epithelial cells with surrounding immune, fibroblast, and endothelial cell populations. When comparing pathway activation profiles, HighStem malignant epithelial cells exhibited higher scores for DNA repair, oxidative phosphorylation, MYC, and E2F pathways. Conversely, these high-stemness cells demonstrated lower pathway scores for transforming growth factor-β, Notch, and WNT/β-catenin signaling cascades.

To identify core genetic drivers across these cellular phenotypes, the investigators combined high-dimensional weighted gene co-expression network analysis, differential-expression analysis, survival analysis, spatial transcriptomics validation, and machine-learning modelling. Through this integrative pipeline, the team identified eight core risk genes: TXNL4A, APEX1, TIMM8B, MDH2, TRMT112, C1QBP, CYC1, and BANF1. Subsequent model testing demonstrated that a Gaussian naïve Bayes model achieved favorable overall classification performance across both evaluated patient cohorts.

Interpretability assessments using Shapley additive explanations indicated that all eight identified risk genes contributed directly to the classification and identification of HighStem cells. Furthermore, analyses conducted across immunotherapy cohorts suggested that TXNL4A, BANF1, and MDH2 expression levels may be associated with patient response to immunotherapy regimens. In complementary in silico virtual knockout experiments, perturbation of the BANF1-associated molecular network implicated the gene in protein synthesis, mitochondrial function, and cell adhesion pathways.

The study authors noted that these findings provide a potential molecular basis for understanding stemness heterogeneity in bladder cancer. By resolving the gene expression networks and microenvironmental communication patterns of high-stemness malignant epithelial cells, the computational and transcriptomic framework may help inform the future development of individualized treatment strategies.

The study was published in Discover Oncology as an open access analysis, with the version available on 23 October 2026.

Subject of Research: Molecular features and risk genes of high-stemness malignant epithelial cells in bladder cancer

Article Title: Integrative single-cell and spatial transcriptomics combined with machine learning reveals molecular features and key risk genes of high-stemness malignant epithelial cells in bladder cancer

Article References: Liu, X., Xiao, Z., Wang, Y., & Huang, L. (2026). Integrative single-cell and spatial transcriptomics combined with machine learning reveals molecular features and key risk genes of high-stemness malignant epithelial cells in bladder cancer. Discover Oncology. https://doi.org/10.1007/s12672-026-06196-2

Image Credits: AI Generated

DOI: 10.1007/s12672-026-06196-2

Keywords: bladder cancer, single-cell RNA sequencing, spatial transcriptomics, machine learning, cellular stemness, malignant epithelial cells, CytoTRACE, CopyKAT, TXNL4A, APEX1, BANF1, MDH2, immunotherapy response, Gaussian naive Bayes, SHAP analysis, hdWGCNA