KAUSTNMF

KAUSTNMF applies non-negative matrix factorization (NMF) with a maximum correntropy criterion (MCC) to analyze and cluster gene expression data for cancer classification.

Key Features:

• Non-negative matrix factorization (NMF) with MCC: Implements NMF optimized using the maximum correntropy criterion (MCC) for matrix decomposition of gene expression data.
• Maximum correntropy criterion (MCC): Maximizes correntropy as a robust similarity measure that provides stability against outliers and noise.
• Alternative to traditional losses: Replaces minimization of the l2 norm or the Kullback-Leibler distance with a correntropy maximization approach.
• Optimization algorithm: Solves the MCC-based optimization using an expectation conditional maximization algorithm.
• Improved cancer clustering performance: Demonstrated higher accuracy than existing state-of-the-art methods in cancer clustering tasks.
• Benchmark validation: Validated on six benchmark cancer datasets.
• Pattern extraction from gene expression: Extracts meaningful patterns from complex gene expression data to support cancer classification.

Scientific Applications:

• Cancer clustering and classification: Improves clustering of gene expression profiles for cancer subtype identification.
• Robust analysis of noisy data: Provides noise- and outlier-robust decomposition of gene expression matrices for downstream biological interpretation.
• Method comparison in oncology studies: Enables comparison against state-of-the-art NMF methods on cancer datasets.

Methodology:

Performs NMF using the maximum correntropy criterion (MCC), maximizes correntropy rather than minimizing the l2 norm or the Kullback-Leibler distance, and solves the resulting optimization via an expectation conditional maximization algorithm.