PCA Based on Graph Laplacian Regularization and P-Norm for Gene Selection and Clustering

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Identifying characteristic genes from gene expression data is crucial in molecular biology.
• Traditional principal component analysis (PCA) methods are sensitive to outliers and noise in biological data.
• Robust methods are needed for accurate gene identification and sample clustering.

Purpose of the Study:

• To develop a novel principal component analysis (PCA) method robust to outliers and noise.
• To enhance the identification of characteristic genes and sample clustering in gene expression data.
• To address the limitations of traditional PCA in handling noisy biological datasets.

Main Methods:

• Developed a novel PCA method, PgLPCA, incorporating a P-norm error function and graph-Laplacian regularization.
• Utilized a non-convex proximal P-norm for the error function to mitigate outlier and noise sensitivity.
• Employed an augmented Lagrange multiplier method for efficient optimization of the minimization problem.

Main Results:

• PgLPCA demonstrates superior accuracy in selecting characteristic genes compared to existing methods.
• The method effectively clusters samples from large biological datasets.
• The proposed approach shows improved performance in the presence of outliers and noise.

Conclusions:

• PgLPCA offers a robust and accurate solution for characteristic gene identification and sample clustering.
• The novel P-norm error function and Laplacian regularization enhance data representation and analysis.
• This method advances the analysis of complex and noisy gene expression data in molecular biology.