A Generic Sure Independence Screening Procedure.

Area of Science:

• Statistical Learning
• Information Theory
• Genomics
• Bioinformatics
• Computational Biology

Background:

• Feature extraction from ultra-high dimensional data is crucial for fields like precision medicine.
• Existing screening methods often rely on specific models and assumptions, limiting their applicability.
• A flexible, model-free approach is needed for diverse data types and complex models.

Purpose of the Study:

• To propose a generic, nonparametric sure independence screening procedure (BCor-SIS).
• To leverage Ball correlation, a universal dependence measure, for model-free feature screening.
• To demonstrate the method's effectiveness in challenging biological discovery settings.

Main Methods:

• Developed BCor-SIS, a sure independence screening procedure based on Ball correlation.
• Established strong screening consistency for BCor-SIS, even with exponential dimensionality relative to sample size.
• Investigated iterative BCor-SIS, interaction pursuit, and survival outcome analysis.

Main Results:

• BCor-SIS demonstrates strong screening consistency without sub-exponential moment assumptions.
• The procedure is effective in challenging settings like iterative screening and interaction detection.
• Simulations and real-world data analyses confirm the method's versatility and practicality.

Conclusions:

• BCor-SIS offers a powerful, flexible, and assumption-light approach for feature screening in ultra-high dimensional data.
• The method shows significant promise for applications in biological discovery and precision medicine.
• BCor-SIS advances the field by providing a generic nonparametric screening solution.