MHSNMF: multi-view hessian regularization based symmetric nonnegative matrix factorization for microbiome data...

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• High-throughput techniques generate vast amounts of heterogeneous omics data (genomics, proteomics, metabolomics).
• Integrating multi-view data is crucial for understanding complex microbe-nutrient-host interactions.
• Existing methods face challenges in leveraging high-order information and complex relationships within microbiome data.

Purpose of the Study:

• To propose a multi-view Hessian regularization based symmetric nonnegative matrix factorization (MHSNMF) algorithm.
• To effectively cluster heterogeneous microbiome data by integrating multiple data sources.
• To develop a novel prediction method for classifying new microbiome samples.

Main Methods:

• Developed a multi-view Hessian regularization approach combined with symmetric nonnegative matrix factorization (SNMF).
• Leveraged high-order information from multiple data representations of the same samples.
• Utilized a consensus matrix for sample clustering and a novel prediction strategy.

Main Results:

• The MHSNMF algorithm demonstrated superior performance on real-world datasets (Three-source and Human Microbiome Plan).
• Achieved high accuracy (95.28%) and normalized mutual information (91.79%) in microbiome data analysis.
• Outperformed baseline and state-of-the-art methods in clustering and classification tasks.

Conclusions:

• MHSNMF effectively integrates phylogenetic, transporter, and metabolic profiles for microbiome sample analysis.
• The proposed prediction method accurately classifies new microbiome samples.
• MHSNMF shows significant potential for advancing microbiome data analysis and understanding host-microbe interactions.