Metabolite profile analysis: from raw data to regression and classification

Abstract

Successful metabolic profile analysis will aid in the fundamental understanding of physiology. Here, we present a possible analysis workflow. Initially, the procedure to transform raw data into a data matrix containing relative metabolite levels for each sample is described. Given that, because of experimental issues in the technical equipment, the levels of some metabolites cannot be universally determined or that different experiments need to be compared, missing value estimation and normalization are presented as helpful preprocessing steps. Regression methods are presented in this review as tools to relate metabolite levels with other physiological properties like biomass and gene expression. As the number of measured metabolites often exceeds the number of samples, dimensionality reduction methods are required. Two of these methods are discussed in detail in this review. Throughout this article, practical examples illustrating the application of the aforementioned methods are given. We focus on the uncovering the relationship between metabolism and growth-related properties.

Metabolic Networks: How to Identify Key Components in the Regulation of Metabolism and Growth

“…Deeper insights were provided by applying canonical correlation analysis and partial least squares regression. These multivariate methods identify which combination and weighting of a large set of predictors (in this case, about 150 metabolites) provide the best prediction of a target trait (in this case, biomass; Steinfath et al, 2008). These authors identified a set of metabolites that provided a highly significant prediction of biomass.…”

“…In complex multifactorial systems, additional information can be extracted using more sophisticated statistical methods, including principal components analysis, canonical correlation analysis, partial least squares regression, and mutual information (Janes and Yaffe, 2006;Steinfath et al, 2008). For example, in a recombinant inbred line (RIL) Arabidopsis (Arabidopsis thaliana) population, many low-M r metabolites correlated negatively with biomass, but the correlations were not significant (Meyer et al, 2007).…”

“…There have been many reviews of different aspects of metabolomics in the last years. They provide an overview of the available technology for analysis of metabolite levels (Kopka et al, 2004;Hall, 2006;Lisec et al, 2006;Last et al, 2007), the use of metabolic flux analysis to analyze fluxes in complex metabolic systems (Fernie et al, 2005;Noh et al, 2006;Antoniewicz et al, 2007;Libourel and Shachar-Hill, 2008;Kruger and Ratcliffe, 2009), reporting standards (Goodacre et al, 2007;Taylor et al, 2008), and statistical and modeling methods (Liao et al, 2003;Fernie et al, 2004;Janes and Yaffe, 2006;Goodacre et al, 2007;Guy et al, 2008;Steinfath et al, 2008). This update focuses on a set of nontechnical questions that are also of central importance for any systems analysis of plant metabolism.…”

Rice Metabolomics

“…Additionally, they have been found to be applicable to the safety assessment of GM crops. Therefore, many positive studies on plant metabolomics have been performed, and therefore, many recent review articles on plant metabolomics can be consulted for general statements [9,28,31,35,41,80,90], methodology [13,18,30,46,69,81,89,99,101], and applications [42,67,76,79,84,97].…”

Improvement of the quantitative differential metabolome pipeline for gas chromatography-mass spectrometry data by automated reliable peak selection

“…Input file format is based on the output files of FragmentAlign but it is possible to adjust to text format output file of other peak alignment programs. The algorithm and parameters of the peak selection process were designed based on conventional manual procedures and were implemented using the script language Perl (De Souza et al 2006;Steinfath et al 2008). This program consists of the four following steps: 1) selection of the available peaks, 2) evaluation of the reliability of peaks in sample groups and calculation of the mean of peak intensity, 3) comparison of the mean peak intensity between sample groups for all aligned peaks, and 4) selection of aligned peaks that are reliable and show intensity differences among the sample groups (Figure 2).…”