High-dimensional data analysis: selection of variables, data compression and graphics--application to gene expression

Area of Science:

• Bioinformatics
• Statistical Genetics
• Computational Biology

Background:

• High-dimensional data, such as gene expression data, presents significant challenges in statistical analysis.
• Effective variable selection is crucial for increasing statistical power and enabling biological interpretation in complex datasets.
• Existing methods may struggle with the scale and complexity of modern biological data, necessitating robust approaches.

Purpose of the Study:

• To present mathematically exact and effective procedures for variable selection in very high-dimensional settings.
• To enhance statistical power and facilitate biological interpretation through the construction of variable sets.
• To ensure statistical stability and avoid overfitting while maintaining control over familywise type I error.

Main Methods:

• Variables are selected by considering each single variable as the center of potential variable sets.
• Significance testing is performed using the Westfall-Young principle (resampling-based) or parametric spherical tests.
• A specific data compression technique is employed for generating graphical representations like heat maps and curves.

Main Results:

• The proposed procedures are effective for high-dimensional data, including gene expression analysis.
• High statistical power is achieved, and familywise type I error is controlled despite large data dimensions.
• Graphical representations using heat maps and curves are generated via a data compression technique.

Conclusions:

• The developed variable selection methods offer a robust solution for analyzing high-dimensional biological data.
• These procedures successfully balance statistical power, interpretability, and error control.
• The methodology is demonstrated effectively using gene expression data from B-cell lymphoma patients.