Matrix Linear Models for Connecting Metabolite Composition to Individual Characteristics

Area of Science:

• Biochemistry
• Bioinformatics
• Systems Biology

Background:

• High-throughput metabolomics generates complex datasets linking molecular profiles to biological states.
• Current analysis often involves a two-step process: metabolite-individual association followed by enrichment analysis.
• This stepwise approach can obscure intricate relationships between metabolite and sample characteristics.

Purpose of the Study:

• To develop a unified statistical framework for analyzing high-throughput metabolomics data.
• To integrate metabolite and sample characteristics within a single analytical model.
• To improve the assessment of how metabolite levels associate with individual characteristics, considering metabolite properties.

Main Methods:

• A bilinear model based on the matrix linear model (MLM) framework was adapted for metabolomics.
• The method estimates relationships considering both categorical (e.g., pathways) and numerical (e.g., double bonds) metabolite characteristics.
• The approach was implemented in the open-source Julia package, MatrixLM.

Main Results:

• The MLM approach successfully integrated external information into metabolomics data analysis.
• Demonstrated ability to disentangle overlapping metabolite characteristics, such as in triglyceride analysis (e.g., double bonds vs. carbon atoms).
• The framework proved flexible and interoperable across three diverse metabolomic studies.

Conclusions:

• The matrix linear model provides a powerful, efficient, and interpretable method for complex metabolomics data analysis.
• This approach enhances the integration of biological context and external information.
• The MatrixLM package offers a practical tool for researchers in the field.