Effect of scaling and normalisation in ranking of metabolites in metabolomics studies

Metabolomics has been successfully applied in many fields including clinical research, drug discovery, toxicology, and phytochemistry. In all of these methods pathophysiological stimuli or genetic modification is performed on living system and a quantitative measure of the dynamic metabolic response of living systems is made to explain the disease process and mechanism could be investigated in a synthesis induction way[1]. However extracting relevant biological information from large data sets is a major challenge in this field [2]. From data acquisition to statistical analysis, metabolomics data need to undergo several processing steps, which all of them is critical in correct interpretation of data [3]. Prior to multivariate analysis, preprocessing of the data must be carried out to remove unwanted variation such as instrumental or experimental artifacts. In this study effect of scaling and normalization has been investigated by both simulated and 1HNMR nanotoxicometabolomics real data. Different scaling methods such as centering, autoscaling, pareto scaling, range scaling, vast scaling and different normalization methods such as total area normalization, probabilistic quotient normalization, and quantile normalization have been used in this study [1-4]. Finally, it was shown that preprocessing of metabolomics data is an important step prior to statistical analysis and selecting a proper data pretreatment method is a critical step in the analysis of metabolomics data and strongly affects the identification of the most important metabolites in the considered biological system.