Kernel approaches for differential expression analysis of mass spectrometry-based metabolomics data

Area of Science:

• Biostatistics
• Bioinformatics
• Metabolomics

Background:

• Metabolomics data presents unique challenges, including frequent missing values in mass spectrometry (MS)-based experiments.
• Traditional statistical methods struggle to integrate the dual information of metabolite presence/absence and quantitative abundance levels.

Purpose of the Study:

• To develop a novel statistical approach for differential expression analysis in metabolomics.
• To address the challenge of analyzing metabolomics data with both continuous and discrete patterns, including missing values.

Main Methods:

• Proposed a kernel-based score test specifically designed for metabolomics differential expression analysis.
• Introduced two novel kernels: a distance-based kernel and a stratified kernel, to capture both continuous and discrete data patterns.
• Extended the methodology for single-metabolite analysis to metabolite set analysis.

Main Results:

• The proposed kernel method demonstrated superior performance compared to existing methods.
• Evaluations using both simulated and real-world liver cancer metabolomics data confirmed the method's effectiveness.

Conclusions:

• The novel kernel-based score test provides a robust solution for differential expression analysis in metabolomics.
• The method successfully integrates diverse data types, outperforming current alternatives.
• An R implementation of the kernel method is publicly available for researchers.