Hessian regularization based non-negative matrix factorization for gene expression data clustering.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Gene expression data analysis relies on identifying genes with similar patterns.
• Non-negative Matrix Factorization (NMF) is a common data representation technique.
• Laplacian regularization (LR) improves NMF but has weak extrapolation power, especially with limited data.

Purpose of the Study:

• To propose and evaluate a novel Hessian regularization-based NMF (HR-NMF) algorithm.
• To enhance data representation for gene expression clustering.
• To address the extrapolation limitations of traditional NMF methods.

Main Methods:

• Development of the HR-NMF algorithm incorporating Hessian regularization.
• Application of HR-NMF for data representation in gene expression analysis.
• Clustering experiments on five standard gene expression datasets.

Main Results:

• The proposed HR-NMF algorithm demonstrated superior performance compared to LR-based NMF and standard NMF.
• HR-NMF showed improved clustering accuracy on gene expression datasets.
• The method's effectiveness was particularly noted for small sample data scenarios.

Conclusions:

• HR-NMF offers a promising approach for gene expression data representation and clustering.
• The enhanced extrapolation capabilities of HR-NMF are beneficial for analyzing complex biological data.
• This work suggests significant potential for HR-NMF in bioinformatics applications.