Sparse Graph Regularization Non-Negative Matrix Factorization Based on Huber Loss Model for Cancer Data Analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Non-negative matrix factorization (NMF) is widely used for dimensionality reduction in cancer gene expression data.
• The standard NMF's square loss function is sensitive to non-Gaussian noise and outliers common in gene expression data.
• Existing methods often struggle with the inherent noise and variability in biological datasets.

Purpose of the Study:

• To develop a more robust NMF algorithm for cancer data analysis.
• To improve the clustering performance and accuracy of gene selection in cancer genomics.
• To address the limitations of traditional NMF in handling noisy gene expression data.

Main Methods:

• Proposed a novel sparse graph regularization NMF model incorporating Huber loss (Huber-SGNMF).
• Huber loss function, situated between L1 and L2 norms, enhances robustness against non-Gaussian noise and outliers.
• Incorporated sparse penalty and graph regularization to improve matrix sparsity and capture data manifold structure.

Main Results:

• Huber-SGNMF demonstrated superior robustness compared to other NMF models.
• Experiments on The Cancer Genome Atlas (TCGA) data showed improved sample clustering accuracy.
• The method excelled in selecting differentially expressed genes compared to state-of-the-art techniques.

Conclusions:

• Huber-SGNMF offers a significant advancement in analyzing noisy cancer gene expression data.
• The proposed method enhances the reliability of NMF for cancer subtyping and biomarker discovery.
• This approach provides a more accurate and robust tool for cancer data mining.