Closing the Knowledge Gap of Post-Acquisition Sample Normalization in Untargeted Metabolomics.

Area of Science:

• Metabolomics
• Bioinformatics
• Quantitative Biology

Background:

• Sample normalization is critical in metabolomics to minimize variations and enable fair quantitative comparisons.
• Post-acquisition normalization is essential when total metabolite quantity cannot be accurately measured.
• Limited understanding of normalization algorithm differences hinders optimal method selection.

Purpose of the Study:

• To elucidate the mechanisms and performance differences of common post-acquisition sample normalization methods.
• To benchmark six distinct normalization techniques: sum, median, probabilistic quotient normalization (PQN), maximal density fold change (MDFC), quantile, and class-specific quantile.
• To propose an evidence-based strategy for maximizing sample normalization outcomes in metabolomics.

Main Methods:

• Utilized data simulation, experimental simulation, and real experiments to evaluate normalization methods.
• Benchmarked six normalization methods on public metabolomics datasets.
• Analyzed the impact of unbalanced data and data quality factors (noise, saturation, missingness) on normalization performance.

Main Results:

• Demonstrated significant discrepancies in outcomes between different sample normalization methods.
• Revealed that most normalization methods exhibit bias when dealing with unbalanced data (unequal up- and downregulated metabolites).
• Highlighted the influence of data quality issues on normalization effectiveness.

Conclusions:

• Normalization method performance is significantly affected by data characteristics, particularly unbalanced data.
• An evidence-based normalization strategy is proposed to improve quantitative comparisons in metabolomics.
• This work provides a robust bioinformatic solution for advancing metabolomics research through improved normalization.