Novel harmonic regularization approach for variable selection in Cox's proportional hazards model.

Area of Science:

• Statistics
• Bioinformatics
• Genomics

Background:

• Variable selection is crucial in regression analysis, with many methods utilizing nonconvex penalty functions.
• Microarray gene expression data analysis requires robust variable selection to identify key risk factors.

Purpose of the Study:

• To introduce and evaluate a novel harmonic regularization method for variable selection in Cox's proportional hazards model.
• To approximate nonconvex Lq regularizations (1/2 < q < 1) for enhanced risk factor identification.

Main Methods:

• Developed a harmonic regularization method for variable selection.
• Employed a direct path seeking approach for efficient computation.
• Validated the method using artificial and real microarray gene expression datasets (e.g., DCBCL, lung cancer, AML).

Main Results:

• The harmonic regularization method demonstrated efficient solvability via the direct path seeking approach.
• Solutions closely approximated those of convex loss functions and nonconvex regularization.
• The method showed higher accuracy in variable selection compared to existing Lasso series methods on real-world datasets.

Conclusions:

• Harmonic regularization offers a more accurate approach to variable selection in Cox's proportional hazards models for gene expression data.
• The proposed method provides an efficient and effective alternative to existing techniques for identifying key risk factors.