Robust Metabolomics Data Normalization across Scales and Experimental Designs

Area of Science:

• Analytical Chemistry
• Bioinformatics
• Systems Biology

Background:

• Metabolomics studies using liquid chromatography-mass spectrometry (LC-MS) are susceptible to signal drift and batch effects.
• These technical variations introduce noise, hindering the discovery of true biological insights.
• Current quality control (QC) sample-based normalization methods are often compromised by outliers, limiting their effectiveness.

Purpose of the Study:

• To develop and evaluate robust normalization methods for LC-MS metabolomics data.
• To enhance the accuracy and reliability of biological discoveries by minimizing technical variance.
• To provide a versatile R package, Metanorm, for implementing these advanced normalization strategies.

Main Methods:

• Introduction of three robust normalization methods: rLOESS, rGAM, and tGAM, designed to resist outliers.
• Utilization of additive models in rGAM and tGAM for flexible nonlinear modeling and differential sample weighting.
• Development of the Metanorm R package for integrated normalization, visualization, and parallel processing.

Main Results:

• Robust methods demonstrated improved replicate concordance and reduced drift and batch effects compared to existing strategies in both in silico and experimental datasets.
• Simulations showed enhanced recovery of underlying signals with robust methods.
• Distinct differential abundance results were observed, underscoring the impact of normalization on statistical inference.

Conclusions:

• The proposed robust normalization methods, particularly tGAM, offer superior performance in metabolomics data analysis.
• Metanorm package provides a versatile and efficient tool for robust normalization across various metabolomics study scales and setups.
• These advancements are crucial for improving the reliability and reproducibility of metabolomics research.