The L(1/2) regularization approach for survival analysis in the accelerated failure time model

Area of Science:

• Bioinformatics
• Genomics
• Biostatistics

Background:

• Analyzing high-dimensional, low-sample size microarray data for cancer survival analysis presents significant challenges.
• Selecting relevant biomarkers from gene expression datasets where genes vastly outnumber samples is a major hurdle.

Purpose of the Study:

• To develop a robust prediction approach for patient survival time using a L(1/2) regularization estimator within the accelerated failure time (AFT) model.
• To address the challenge of relevant gene selection in high-dimensional biological data.

Main Methods:

• Implementation of the L(1/2) regularized AFT model using a coordinate descent algorithm.
• Utilization of a renewed half thresholding operator for optimizing gene selection.
• Application to five real DNA microarray datasets.

Main Results:

• The L(1/2) regularized AFT model yields more accurate and sparse predictors for survival analysis compared to L1 regularization methods.
• Simulation experiments validate the model's effectiveness in high-dimensional gene expression data analysis.

Conclusions:

• The proposed L(1/2) regularization approach offers an efficient and effective method for predicting cancer patient survival time.
• This technique enhances the identification of prognostic gene signatures and clinical factors from complex biological datasets.