The Multifaceted Challenge of Normalizing Mass Spectrometry-Based Metabolomics Data.

Normalization is a critical step in metabolomics studies to ensure the quality of metabolomics data, reduce quantitative variability, and enable confident and robust statistical analyses. From an analytical perspective, metabolomics normalization encompasses multiple distinct processes. Broadly, normalization can refer to (1) sample normalization, which mitigates variation due to differences in total metabolite amounts; (2) signal correction, which reduces batch effects, instrumental fluctuations, and retention time drifts during data collection; and (3) statistical transformation and scaling, which prepare data for statistical analyses. Each of these normalization processes addresses unique analytical and bioinformatic needs, but the term "normalization" is often used broadly, leading to confusion in method development, selection, and implementation. Moreover, many well-established normalization algorithms in genomics and proteomics are not always transferable to metabolomics due to differences in analytical workflows and data characteristics. To address these issues, we believe it is crucial to gain a clear understanding of the purpose of each normalization type, its appropriate implementation, and the evaluation criteria. This perspective outlines the key normalization tasks in metabolomics, reviews existing tools, and provides recommendations for their appropriate applications. We also highlight two critical considerations: (1) selecting an appropriate missing value imputation method and (2) establishing strategies to evaluate and compare normalization outcomes. The goal of this work is to provide recommendations for the rigorous development and implementation of normalization techniques in metabolomics, thereby enhancing analytical accuracy and precision, improving data interpretability, and ultimately advancing the biological insights gained from metabolomics studies.