Efficient ℓ₀ -norm feature selection based on augmented and penalized minimization.

Area of Science:

• Biostatistics
• Computational Biology
• Genomics
• Medical Imaging

Background:

• High-throughput genomics and imaging generate vast numbers of prognostic biomarkers.
• Penalized regression is crucial for identifying biomarkers associated with disease outcomes.
• Exact L0-norm minimization for variable selection is computationally intractable (NP-hard).

Purpose of the Study:

• To develop a computationally tractable and efficient method for L0-norm penalized variable selection.
• To introduce the augmented penalized minimization-L0 (APM-L0) procedure for biomarker discovery.
• To improve upon existing methods in terms of selection accuracy and computational speed.

Main Methods:

• Proposed a novel 2-stage procedure, APM-L0, for L0-norm penalized variable selection.
• Iterative approach combining convex regularized regression and hard-thresholding estimation.
• Utilized a 1-step coordinate descent algorithm for computational efficiency in the first stage.

Main Results:

• APM-L0 closely targets the L0-norm while maintaining computational tractability.
• Demonstrated superior performance in selection accuracy and computational speed via simulations and real data.
• The method is available as an R-package (APML0).

Conclusions:

• APM-L0 offers an effective and efficient solution for biomarker selection using L0-norm regularization.
• The proposed method advances the field of penalized regression for high-dimensional data analysis.
• APM-L0 provides a valuable tool for researchers in genomics, imaging, and disease outcome studies.