Evaluation of statistical techniques to normalize mass spectrometry-based urinary metabolomics data

Area of Science:

• Metabolomics
• Biomarker Discovery
• Urine Analysis

Background:

• Human urine is a non-invasive medium for metabolomics biomarker discovery.
• Renal dilution in urine can complicate biomarker analysis.
• Existing normalization techniques have limitations.

Purpose of the Study:

• To evaluate various statistical normalization techniques for mass spectrometry-based urinary metabolomic data.
• To compare the performance of normalization methods in differentiating cancer types.
• To assess the impact of normalization on statistical significance in biomarker analysis.

Main Methods:

• Analyzed urinary metabolomic data from prostate cancer, bladder cancer, and control groups.
• Investigated normalization techniques including Creatinine Ratio, Log Value, Linear Baseline, Cyclic Loess, Quantile, Probabilistic Quotient, Auto Scaling, Pareto Scaling, and Variance Stabilizing Normalization.
• Utilized summary statistics, principal component analysis, and hypothesis testing for comparison.

Main Results:

• The choice of normalization technique significantly influences the determination of statistical significance.
• No single normalization method demonstrated universally superior performance across all analyses.
• Principal component analysis revealed varying degrees of clustering based on cancer type depending on the normalization method used.

Conclusions:

• Careful selection of an appropriate normalization technique is essential for reliable urinary metabolomics biomarker discovery.
• The performance of normalization methods varies, impacting the ability to differentiate between cancer types.
• Normalization method choice directly affects the statistical validity of biomarker findings.