Bi-clustering of metabolic data using matrix factorization tools.

Area of Science:

• Metabolomics
• Bioinformatics
• Data Analysis

Background:

• Metabolic phenotyping using Nuclear Magnetic Resonance (NMR) and Mass Spectrometry (MS) generates large, complex datasets.
• Clustering is vital for uncovering relationships between samples and metabolites in this data.
• Conventional clustering methods often require pre-defined cluster numbers and analyze all data points.

Purpose of the Study:

• To propose a novel bi-clustering strategy for metabolic phenotyping data analysis.
• To discover subsets of metabolites with similar behaviors across subsets of samples.
• To offer an alternative to traditional clustering approaches that require a priori cluster specification.

Main Methods:

• Utilized non-negative matrix factorization (NMF) for a bi-clustering strategy.
• Incorporated bi-cross-validation and statistical segmentation for automatic cluster determination.
• Compared the proposed method with existing bi-clustering techniques from genomics and transcriptomics.

Main Results:

• The proposed NMF-driven bi-clustering strategy demonstrated superior performance.
• Successfully identified interrelated metabolites and sample subsets.
• Validated on both simulated NMR and real MS bacterial metabolic data.

Conclusions:

• The developed bi-clustering approach offers an effective and automated method for metabolic phenotyping data analysis.
• It provides a powerful alternative to conventional clustering, especially when the number of clusters is unknown.
• The strategy shows promise for advancing the interpretation of complex metabolomic datasets.