XGB-BIF: An XGBoost-Driven Biomarker Identification Framework for Detecting Cancer Using Human Genomic Data
Authors: Veena Ghuriani, Jyotsna Talreja Wassan*, Priyal Tripathi, Anshika Chauhan
A machine learning framework leveraging XGBoost feature ranking to discover robust genomic biomarkers across human cancer cohorts (breast cancer, lung cancer, and gastric cancer), attaining >90% classification accuracy.
Overview
High-throughput transcriptomic sequencing generates tens of thousands of gene expression features per sample, presenting a classic “curse of dimensionality” challenge (p » n). In this peer-reviewed publication in the International Journal of Molecular Sciences (IJMS), we introduce XGB-BIF (XGBoost-Driven Biomarker Identification Framework), an interpretable machine learning pipeline designed to isolate minimal, highly discriminative gene subsets for cancer detection.
Key Highlights & Contributions
- Boosting-Driven Feature Ranking: Evaluated feature importance metrics (gain, coverage, and frequency) across iterative gradient boosting trees to prioritize stable gene sets.
- Dimensionality Reduction & High Accuracy: Compressed large-scale RNA-sequencing matrices down to candidate biomarker panels while exceeding 90% diagnostic accuracy across validation cohorts.
- Biological Validation: Mapped prioritized biomarkers to biological signaling pathways using gene ontology (GO) and KEGG pathway enrichment, confirming biological relevance to tumorigenesis and oncogenic regulation.
Citation
@article{ghuriani2025xgbbif,
title={XGB-BIF: An XGBoost-Driven Biomarker Identification Framework for Detecting Cancer Using Human Genomic Data},
author={Ghuriani, Veena and Wassan, Jyotsna Talreja and Tripathi, Priyal and Chauhan, Anshika},
journal={International Journal of Molecular Sciences},
volume={26},
number={12},
pages={5590},
year={2025},
publisher={MDPI},
doi={10.3390/ijms26125590}
}