A Novel Sparse Graph-Regularized Singular Value Decomposition Model and Its Application to Genomic Data Analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene coexpression pattern discovery is crucial for high-dimensional gene expression analysis.
• Sparse Singular Value Decomposition (SVD) is a common method, but it overlooks variable structural information like gene networks.
• Existing graph-regularized methods fail to account for negative correlations between variables.

Purpose of the Study:

• To propose a novel sparse graph-regularized SVD model with an absolute operator (AGSVD).
• To enhance the discovery and interpretability of gene expression patterns by incorporating prior gene network information.
• To address the limitations of existing methods in handling negative correlations and structural information.

Main Methods:

• Developed a novel sparse graph-regularized SVD model (AGSVD) utilizing an absolute operator penalty (|u|^T L|u|).
• Employed an alternating optimization strategy to efficiently solve the resulting nonconvex and nonsmooth optimization problem.
• Validated the method using synthetic datasets and several real gene expression datasets.

Main Results:

• AGSVD demonstrated superior performance compared to existing SVD-based methods on synthetic data.
• The model successfully incorporated prior gene network information.
• Analysis of real gene expression data revealed more biologically interpretable expression patterns.

Conclusions:

• The proposed AGSVD model offers a more effective approach for high-dimensional gene expression pattern discovery.
• Incorporating gene network structures and handling negative correlations leads to improved accuracy and interpretability.
• AGSVD provides a valuable tool for uncovering biologically meaningful insights from gene expression data.