Related Experiment Video
Updated: Feb 20, 2026
Magnetic Resonance Imaging Assessment of Carcinogen-induced Murine Bladder Tumors
Published on: March 29, 2019
Integrating machine learning and Mendelian randomization for identifying genetic biomarkers in bladder cancer:
Chaojie Xu1, Ying Dong2,3, Jiou Li4
1Department of Urology, Peking University First Hospital, Peking University, Beijing, China.
Background:
Bladder cancer (BC) continues to be a major public health challenge due to its high recurrence and mortality rates, compounded by difficulties in early detection. Identifying novel genetic biomarkers is crucial for improving diagnosis and therapy. This study integrates machine learning with Mendelian randomization (MR) to identify and validate potential biomarkers for BC.
Methods:
We analyzed five publicly available Gene Expression Omnibus (GEO) datasets, including both BC and normal tissue samples, to identify differentially expressed genes (DEGs). Data preprocessing included batch effect correction, followed by differential expression and weighted gene co-expression network analysis (WGCNA). We evaluated the diagnostic performance of candidate genes using 113 machine learning models, selecting the best models based on receiver operating characteristic curve analysis. MR analysis was used to assess causal relationships between gene expression and BC risk. In vitro validation was performed using BC cell lines (SW780, UMUC3, 5637, and T24) and normal uroepithelial cells (SV-HUC-1), with gene expression measured by quantitative real-time PCR and functional assays evaluating cell proliferation and migration.
Results:
A total of 48 DEGs were identified, with WGCNA revealing the green module as most associated with BC. Machine learning analysis highlighted FAM107A as a key biomarker, achieving high diagnostic accuracy across datasets. The Stepwise Generalized Linear Model (Stepglm)[backward] + Gradient Boosting Machine (GBM) model performed optimally, achieving the highest area under the curve (AUC). MR analysis confirmed that higher expression of FAM107A is protective against BC risk, with experimental validation showing that FAM107A overexpression inhibits BC cell proliferation and migration.
Conclusion:
This study demonstrates the power of combining machine learning and MR to uncover genetic biomarkers for BC. FAM107A emerges as a promising diagnostic and therapeutic target, offering new insights for personalized BC treatment strategies.
