Related Experiment Video
Updated: Sep 18, 2025
07:41
Performing Data Mining And Integrative Analysis Of Biomarker in Breast Cancer Using Multiple Publicly Accessible Databases
Published on: May 17, 2019
9.1K
XGB-BIF: An XGBoost-Driven Biomarker Identification Framework for Detecting Cancer Using Human Genomic Data
Veena Ghuriani1, Jyotsna Talreja Wassan1, Priyal Tripathi1
1Maitreyi College, University of Delhi, New Delhi 110021, India.
International Journal of Molecular Sciences
|June 26, 2025
Summary
This study introduces XGB-BIF, a machine learning framework that identifies key genomic biomarkers for gastric, lung, and breast cancers. The approach achieves over 90% accuracy in distinguishing diseased from non-diseased states, aiding targeted therapy development.
Area of Science:
- Genomics
- Bioinformatics
- Machine Learning
Background:
- The human genome plays a critical role in health and disease, with cancer detection and targeted therapies requiring advanced analytical methods.
- Identifying specific genomic biomarkers is crucial for effective cancer diagnosis and treatment strategies.
Purpose of the Study:
- To develop and validate a machine learning framework (XGB-BIF) for identifying genomic biomarkers associated with gastric, lung, and breast cancers.
- To enhance disease detection and stratify cancer risk using genomic data analysis.
Main Methods:
- Analysis of human genomic data from cancer patients using the XGBoost-Driven Biomarker Identification Framework (XGB-BIF).
- Feature selection via XGBoost (eXtreme Gradient Boosting) for capturing complex feature interactions.
- Classification using Support Vector Machines (SVM), Logistic Regression (LR), and Random Forest (RF) models.
- Interpretability through SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME).
- Validation of biomarker significance using pathway enrichment and survival analysis (Kaplan-Meier curves, Cox regression).
Main Results:
- The XGB-BIF framework achieved high predictive performance with >90% accuracy in classifying cancer states.
- Strong agreement between predicted and actual risk categories confirmed by Cohen's Kappa statistic (0.80-0.99).
- External validation on the METABRIC dataset yielded an AUC-ROC of 93%, accuracy of 79%, and Kappa of 74%.
- Identification of specific high-impact biomarker genes for gastric (e.g., CBX2, CLDN1), breast (e.g., CAVIN2, ADAMTS5), and lung cancers (e.g., CLDN18, MYBL2).
Conclusions:
- XGB-BIF effectively identifies genomic biomarkers for distinct cancer types, demonstrating translational value.
- The identified biomarkers can significantly aid clinicians in developing targeted therapies for cancer patients.
- The framework offers a robust approach for biomarker discovery using genomic data, improving cancer detection and risk stratification.
