Cancer classification and functional pathway discovery using TCGA transcriptomic profiles: A matched case-control...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional transcriptomic data from The Cancer Genome Atlas (TCGA) is crucial for precision oncology.
• Matched Case-Control Design (MCCD) enhances statistical power but faces challenges like overfitting and instability in high-dimensional settings.
• Feature selection is vital for mitigating the 'curse of dimensionality' by identifying key variables and reducing redundancy.

Purpose of the Study:

• To develop and validate a unified analytical workflow for cancer classification using MCCD with high-dimensional transcriptomic data.
• To compare paired versus unpaired feature selection approaches in simulated MCCD scenarios.
• To enhance classification accuracy, feature stability, and biological interpretability for precision medicine.

Main Methods:

• Developed a modular, pluggable pipeline integrating mean-centering, gene filtering, and a Corrected Feature Matrix (CFM) transformation preserving matched structure.
• Applied Incremental Feature Selection (IFS) for gene subset refinement and gene set enrichment analysis for interpretability.
• Validated the workflow using simulated and real TCGA datasets with machine learning classifiers.

Main Results:

• The integrated workflow demonstrated superior performance over uncorrected approaches in cancer classification accuracy.
• The proposed method showed improved feature stability and enhanced biological interpretability.
• The workflow effectively leverages MCCD for high-dimensional transcriptomic data analysis.

Conclusions:

• The developed workflow offers a practical and scalable tool for enhancing cancer classification accuracy in precision medicine.
• This approach facilitates biomarker discovery and aids in building interpretable diagnostic models.
• The method effectively addresses the challenges of high-dimensional data in MCCD settings.