Variable Selection in High-Dimensional Logistic Regression Models Using a Whitening Approach.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Omics data analysis faces challenges due to high feature dimensions and limited sample sizes.
• Identifying informative biomarkers is crucial for accurate classification in biomedical research.
• High correlations between biomarkers often hinder effective feature selection.

Purpose of the Study:

• To develop an innovative feature selection method, WLogit, for binary classification of omics data.
• To address the challenge of high correlations among biomarkers.
• To improve classification accuracy by effectively identifying active biomarkers.

Main Methods:

• The WLogit method employs whitening of the design matrix to decorrelate biomarkers.
• A penalized criterion tailored for logistic regression is utilized for feature selection.
• The method is implemented in the WLogit R package.

Main Results:

• WLogit successfully identifies nearly all active biomarkers, even when highly correlated.
• Numerical experiments demonstrate superior performance compared to existing methods.
• Evaluation on public datasets shows WLogit achieves higher prediction accuracy.

Conclusions:

• WLogit offers a robust solution for biomarker identification in high-dimensional, correlated omics data.
• The method enhances classification accuracy in bioinformatics.
• The WLogit R package provides a practical tool for researchers.