Sparse robust graph-regularized non-negative matrix factorization based on correntropy.

Area of Science:

• Computational biology
• Machine learning
• Bioinformatics

Background:

• Non-negative Matrix Factorization (NMF) is widely used for data dimension reduction.
• Traditional NMF is sensitive to noise and outliers in data.
• Robustness is crucial for accurate analysis of complex biological datasets.

Purpose of the Study:

• To propose a robust NMF model, Sparse Robust Graph-regularized Non-negative Matrix Factorization based on Correntropy (SGNMFC).
• To enhance data analysis by improving robustness against noise and preserving data geometry.
• To validate the effectiveness of SGNMFC in gene expression analysis and sample clustering.

Main Methods:

• Utilizing maximized correntropy instead of minimized Euclidean distance for improved robustness.
• Employing graph regularization to preserve the geometry of high-dimensional data in low-dimensional manifolds.
• Applying sparse constraints to the loss function for reduced matrix complexity.

Main Results:

• The SGNMFC model demonstrates superior robustness compared to traditional NMF.
• Correntropy effectively down-weights outliers and noise, emphasizing meaningful data.
• Graph regularization successfully preserves the underlying data structure.
• Sparse constraints simplify matrix complexity and analysis.

Conclusions:

• SGNMFC offers a more robust and effective approach for dimension reduction and data analysis.
• The method shows significant improvements in identifying differentially expressed genes and clustering samples.
• SGNMFC proves effective across multiple The Cancer Genome Atlas (TCGA) datasets.