Identifying Novel Diagnostic Biomarkers in Bladder Cancer

Abstract

Bladder cancer remains a significant clinical challenge due to limitations in current diagnos tic methods, which are invasive, costly, or lack sensitivity particularly for early-stage detection. This study aimed to identify novel transcriptomic biomarkers that enable non-invasive and accurate detection of bladder cancer using machine learning and bioinformatics approaches. Publicly available gene expression datasets from bladder tumor and normal tissues were analyzed using differential expression analysis and weighted gene co-expression network analysis (WGCNA) to identify genes within tumor-associated modules. From an intersection of 408 candidate genes, LASSO regression selected 21 potential biomarkers. These were further evaluated using cross-validation and model performance metrics, and an 11-gene diagnostic signature was prioritized based on consistent discriminatory power (AUC > 0.80) and biological relevance.The diagnostic performance of this signature was rigorously assessed using multiple machine learning classifiers, including Support Vector Machines, Artificial Neural Networks, and Random Forests, across three independent vali dation cohorts. The signature consistently achieved high area under the ROC curve (AUC) values ranging from 0.926 to 0.963, demonstrating strong cross-platform generalizability. To enhance biological and clinical interpretation, Shapley Additive Explanations (SHAP) were employed to highlight key genes contributing to tumor classification and to reveal their potential roles in tumor biology and immune microenvironment sig naling.Overall, this study presents a biologically coherent, interpretable, and robust 11-gene transcriptomic signature for bladder cancer diagnosis. While the findings are derived from publicly available datasets, they provide a reproducible analytical framework and a promising foundation for future prospective validation and clinical translation toward non-invasive diagnostic testing.