Edge-group sparse PCA for network-guided high dimensional data analysis.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-dimensional gene expression data analysis is crucial in modern biology.
• Principal Component Analysis (PCA) is a common technique for dimensionality reduction.
• Existing sparse PCA methods may not fully leverage biological network information.

Purpose of the Study:

• To propose a novel Edge-group Sparse PCA (ESPCA) model.
• To incorporate prior gene network group structures into PCA for improved feature interpretation.
• To enhance the analysis of multiple gene expression matrices.

Main Methods:

• Developed an alternating iterative algorithm to solve the ESPCA model.
• Addressed the k-edge sparse projection problem using a greedy strategy.
• Applied ESPCA to analyze multiple gene expression datasets simultaneously.

Main Results:

• ESPCA identified more biologically relevant genes than PCA and sparse PCA.
• The method improved biological interpretations of gene modules.
• ESPCA revealed distinct sample characteristics in real biological datasets.

Conclusions:

• ESPCA offers a powerful approach for high-dimensional gene expression data analysis.
• Incorporating prior biological network knowledge enhances PCA's interpretability.
• The ESPCA R package is available for public use.