Exploratory Analysis of Metabolomic Data

Abstract This chapter focuses on the exploratory analysis of metabolomic data. The goal of data exploration is to uncover any possible differences in sampling procedures, to examine the between- or within-laboratory variations in the measurement processes, to understand the issues that are associated with the appropriateness of the selected experimental design, to investigate the relationships among samples and between samples and the metabolites that are measured, to obtain information about the occurrence of groups of similar samples, to detect outlying biological samples and finally to formulate new hypotheses about the presence of subgroups in the data and, eventually, to verify them experimentally later. In order to fulfil these tasks, various classic projection and clustering methods such as principal component analysis, hierarchical and nonhierarchical clustering methods as well as self-organizing maps have gained popularity in metabolomics, but here, special attention is paid to some new developments and extensions of the chemometric methods that already exist, which are designed to look for a clustering tendency or extreme biological samples in data, to find natural groups of arbitrary shapes, to incorporate the measurement uncertainty estimation within the construction of models and to deal with missing and left-censored data.