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

Human urine recently became a popular medium for metabolomics biomarker discovery because its collection is non-invasive. Sometimes renal dilution of urine can be problematic in this type of urinary biomarker analysis. Currently, various normalization techniques such as creatinine ratio, osmolality, specific gravity, dry mass, urine volume, and area under the curve are used to account for the renal dilution. However, these normalization techniques have their own drawbacks. In this project, mass spectrometry-based urinary metabolomic data obtained from prostate cancer (n = 56), bladder cancer (n = 57) and control (n = 69) groups were analyzed using statistical normalization techniques. The normalization techniques investigated in this study are Creatinine Ratio, Log Value, Linear Baseline, Cyclic Loess, Quantile, Probabilistic Quotient, Auto Scaling, Pareto Scaling, and Variance Stabilizing Normalization. The appropriate summary statistics for comparison of normalization techniques were created using variances, coefficients of variation, and boxplots. For each normalization technique, a principal component analysis was performed to identify clusters based on cancer type. In addition, hypothesis tests were conducted to determine if the normalized biomarkers could be used to differentiate between the cancer types. The results indicate that the determination of statistical significance can be dependent upon which normalization method is utilized. Therefore, careful consideration should go into choosing an appropriate normalization technique as no method had universally superior performance. (C) 2019 Elsevier B.V. All rights reserved.