Performance comparison of three scaling algorithms in NMR-based metabolomics analysis

Area of Science:

• Metabolomics
• Nuclear Magnetic Resonance (NMR) Spectroscopy
• Bioinformatics

Background:

• Preprocessing is crucial for metabolomics data analysis.
• Common scaling algorithms include Unit Variance (UV), Mean Centering (CTR), and Pareto (Par) scaling.
• The impact of these scaling methods on clustering and metabolite identification in NMR metabolomics is not fully understood.

Purpose of the Study:

• To compare the performance of UV, CTR, and Par scaling methods in NMR-based metabolomics.
• To evaluate their effectiveness in clustering identification and discriminative metabolite extraction.
• To propose an optimal pipeline for selecting scaling algorithms in NMR metabolomics.

Main Methods:

• Utilized NMR-based metabolomics data from diverse biological samples (urine, spleen, serum, bacterial cells).
• Applied Unit Variance (UV) scaling, Mean Centering (CTR) scaling, and Pareto (Par) scaling.
• Assessed clustering identification performance and metabolite discriminative power based on coefficient values.

Main Results:

• Scaling method performance varied significantly for clustering identification.
• UV scaling demonstrated robust performance for clustering analysis in NMR metabolomics, even with technical errors.
• UV, CTR, and Par scaling were equally effective in extracting discriminative metabolites.

Conclusions:

• Unit Variance (UV) scaling is recommended for robust clustering in NMR metabolomics.
• All tested scaling methods efficiently extract discriminative metabolites.
• An optimal working pipeline is proposed to guide the selection of scaling algorithms for NMR-based metabolomics research.