Revealing static and dynamic biomarkers from postprandial metabolomics data through coupled matrix and tensor...

Area of Science:

• Metabolomics
• Systems Biology
• Biostatistics

Background:

• Longitudinal metabolomics data from meal challenges contain both fasting and dynamic signals relevant to metabolic health.
• Analyzing time-resolved metabolomics as a three-way array (subjects, metabolites, time) reveals dynamic markers but struggles to link static and dynamic biomarkers for the same subject groups.
• Existing methods have limited success in extracting both static and dynamic biomarkers for consistent subject stratifications.

Purpose of the Study:

• To jointly analyze fasting and dynamic metabolomics data for a comprehensive understanding of metabolic phenotypes.
• To capture static and dynamic biomarkers within the same subject stratifications for enhanced precision health applications.
• To provide a complete picture of metabolic health by integrating different data states.

Main Methods:

• Utilized coupled matrix and tensor factorizations (CMTF) to jointly analyze fasting (subjects by metabolites) and dynamic (subjects by metabolites by time) metabolomics data.
• Employed data from the COPSAC cohort collected during a meal challenge test.
• Coupled the tensor (dynamic data) and matrix (fasting data) in the subjects mode.

Main Results:

• The data fusion approach successfully extracted shared subject stratifications based on BMI from both fasting and dynamic metabolomics signals.
• Identified static (fasting VLDL) and dynamic (postprandial VLDL particle size) metabolic biomarker patterns associated with BMI-related subject stratifications.
• Observed higher correlations between subject quantifications and the phenotype of interest compared to analyzing fasting and postprandial states independently.

Conclusions:

• The CMTF-based approach effectively integrates static and dynamic metabolomics data to reveal biomarkers for the same subject stratifications.
• This method provides a more complete picture of metabolic health by identifying markers present in both fasting and dynamic states.
• The findings support the utility of data fusion techniques for advancing precision health through comprehensive biomarker discovery.