Chemometrics methods and strategies in metabolomics

Abstract

Chemometrics has been a fundamental discipline for the development of metabolomics, while symbiotically growing with it. From design of experiments, through data processing, to data analysis, chemometrics tools are used to design, process, visualize, explore and analyse metabolomics data. In this chapter, the most commonly used chemometrics methods for data analysis and interpretation of metabolomics experiments will be presented, with focus on multivariate analysis. These are projection-based linear methods, like principal component analysis (PCA) and orthogonal projection to latent structures (OPLS), which facilitate interpretation of the causes behind the observed sample trends, correlation with outcomes or group discrimination analysis. Validation procedures for multivariate methods will be presented and discussed. Univariate analysis is briefly discussed in the context of correlation-based linear regression methods to find associations to outcomes or in analysis of variance-based and logistic regression methods for class discrimination. These methods rely on frequentist statistics, with the determination of p-values and corresponding multiple correction procedures. Several strategies of design-analysis of metabolomics experiments will be discussed, in order to guide the reader through different setups, adopted to better address some experimental issues and to better test the scientific hypotheses.