Stable feature selection based on the ensemble L ₁ -norm support vector machine for biomarker discovery.

Area of Science:

• Biomedical research
• Genomics
• Bioinformatics

Background:

• Biomarker discovery is crucial in biomedicine.
• High-throughput technologies yield vast genomic data (RNA-seq, microarray).
• Conventional feature selection methods struggle with high-dimensional, low-sample data, leading to reproducibility issues.

Purpose of the Study:

• To propose a stable feature selection method for high-dimensional datasets.
• To enhance biomarker discovery and improve classification performance.
• To address limitations of existing feature selection techniques.

Main Methods:

• An ensemble L1-norm support vector machine was employed to reduce irrelevant features.
• A stability score was defined by aggregating ensemble results.
• Backward feature elimination was utilized on a purified feature set for optimal feature selection.

Main Results:

• The proposed method demonstrated superior performance in classification and stability compared to established algorithms.
• It showed moderate performance on datasets with many features and few samples.
• The methodology was validated using RNA-seq data for renal clear cell carcinoma staging.

Conclusions:

• The developed approach offers a robust solution for stable feature selection in biomarker discovery.
• It is expected to be broadly applicable to various biomarker discovery research areas.
• The method enhances the reliability and performance of biomarker identification from complex genomic data.