A New Pipeline for the Normalization and Pooling of Metabolomics Data

Area of Science:

• Molecular Epidemiology
• Bioinformatics
• Biostatistics

Background:

• Pooling metabolomics data across studies increases statistical power but faces challenges due to preanalytical and analytical variations.
• Differences in sample types, collection protocols, storage, and assay conditions can introduce significant variability between datasets.
• Existing methods struggle to effectively harmonize diverse metabolomics datasets for robust analysis.

Purpose of the Study:

• To develop and validate a novel computational pipeline for normalizing and pooling metabolomics data from multiple studies.
• To address and mitigate inter-study variability and biases in metabolomics datasets.
• To enhance the comparability and statistical power of multi-study metabolomics analyses.

Main Methods:

• A sequential pipeline involving data cleaning, outlier removal, and missing data imputation.
• Principal component partial R-square (PC-PR2) analysis to identify sources of variability.
• Linear mixed models to remove unwanted variability (study, batch) while preserving biological signals and accounting for residual variances.

Main Results:

• The pipeline successfully normalized and pooled targeted metabolomics data from eight European Prospective Investigation into Cancer and Nutrition (EPIC) studies.
• Demonstrated improved comparability of metabolomics measurements across diverse datasets.
• The method effectively reduced biases introduced by inter-study variability.

Conclusions:

• The developed pipeline provides a robust method for normalizing and pooling metabolomics data, enhancing its utility for large-scale analyses.
• This approach can be adapted for other molecular data types, including proteomics and biomarkers, facilitating multi-consortia data integration.
• The pipeline offers a valuable tool for molecular epidemiologists seeking to maximize insights from diverse biological datasets.