Scaling and normalization effects in NMR spectroscopic metabonomic data sets

Area of Science:

• Metabonomics
• Nuclear Magnetic Resonance (NMR) Spectroscopy
• Data Science

Background:

• Metabonomics research often faces confusion regarding the necessity and methods of spectroscopic data preprocessing.
• Data in metabonomics are typically structured as tables of sample measurements, such as spectral peak intensities or metabolite concentrations.

Purpose of the Study:

• To clarify the roles of normalization (row operations) and scaling (column operations) in metabonomics data preprocessing.
• To highlight the specific challenges of analyzing urine samples using 1H NMR spectroscopy.
• To discuss the implications of data preprocessing on statistical analyses, like correlation coefficients.

Main Methods:

• Defined and discussed normalization and scaling operations for metabonomic data.
• Demonstrated the application and necessity of these preprocessing steps using 1H NMR spectroscopic data from urine.
• Analyzed the impact of "binned" data and normalization methods on correlation coefficient calculations.

Main Results:

• Normalization and scaling are essential preprocessing steps for 1H NMR spectroscopic data from urine.
• "Binned" data present unique challenges in metabonomics analysis.
• Careful consideration is needed when calculating correlation coefficients for normalized data, especially with constant sum normalization.

Conclusions:

• Data preprocessing in metabonomics is highly context-dependent, with no universal optimal method.
• These preprocessing considerations apply to various biofluids, analytical techniques (e.g., HPLC-MS), and other omics fields (e.g., transcriptomics, proteomics).
• Integrated analysis of "fused" datasets requires analogous preprocessing considerations.