Filter and Wrapper Stacking Ensemble (FWSE): a robust approach for reliable biomarker discovery in high-dimensional...

Area of Science:

• Bioinformatics and Computational Biology
• Machine Intelligence
• Genomics and Proteomics

Background:

• Selecting accurate biomarkers is crucial for disease diagnosis, prognosis, and treatment response prediction.
• High-dimensional omics data present challenges in biomarker discovery due to limited samples, method dependency, and reproducibility issues.

Purpose of the Study:

• To propose a novel ensemble feature selection method, Filter and Wrapper Stacking Ensemble (FWSE), for identifying reproducible biomarkers from high-dimensional omics data.
• To address the challenges of feature selection in bioinformatics, including small sample sizes and non-reproducibility.

Main Methods:

• FWSE employs an ensemble approach combining filter and wrapper feature selection techniques.
• Filter methods are applied to data subsets to remove irrelevant features, followed by wrapper methods to rank the most informative features.

Main Results:

• FWSE demonstrated stability and statistical significance in feature selection across four high-dimensional medical datasets (mental illness and cancer).
• Selected features showed biological relevance and outperformed existing methods in reproducibility.
• The method proved effective on diverse high-dimensional datasets.

Conclusions:

• FWSE offers a robust and generic approach for reproducible biomarker discovery in bioinformatics and machine intelligence.
• The method enhances the reliability of biomarker identification from complex medical data.
• FWSE contributes to advancing personalized medicine through improved diagnostic and prognostic tools.