Hyper-Graph Regularized Constrained NMF for Selecting Differentially Expressed Genes and Tumor Classification.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Non-negative Matrix Factorization (NMF) is a dimensionality reduction technique for non-negative data.
• Standard NMF overlooks data manifold structure and prior class label information.
• These limitations hinder its effectiveness in complex biological data analysis, such as cancer genomics.

Purpose of the Study:

• To propose a novel matrix decomposition method, Hyper-graph regularized Constrained Non-negative Matrix Factorization (HCNMF).
• To enhance gene selection and tumor sample classification by integrating hyper-graph regularization and label information.
• To improve the identification of pathogenic genes and the discriminative ability of decomposition matrices in cancer datasets.

Main Methods:

• Developed HCNMF, a matrix decomposition method incorporating hyper-graph regularization.
• Integrated higher-order data sample relationships using hyper-graph theory.
• Incorporated prior class label information into the objective function for supervised learning.

Main Results:

• HCNMF effectively captures local spatial information and higher-order relationships in high-dimensional data.
• The integration of label information significantly improved the classification performance.
• Experiments on The Cancer Genome Atlas (TCGA) datasets demonstrated the superiority of HCNMF over existing algorithms.

Conclusions:

• HCNMF offers a robust approach for selecting differentially expressed genes and classifying tumor samples.
• The method effectively leverages hyper-graph learning and supervised information for enhanced biological data analysis.
• HCNMF shows significant potential for applications in cancer genomics and precision medicine.