Statistics in metabolomics for biomarker research in medicine

Metabolomics aims at studying chemical processes involving small molecules in living organisms. It is applied on plants, bacteria and animals, but in particular to the human metabolism with a variety of health applications. Typical metabolomics experiments generates large amount of data that require appropriate processing. Consequently, an adequate statistical analysis is crucial for the interpretation. Metabolomic community is well aware about the importance and impact of statistical treatment and publicly available repositories of metabolomics data (raw or processed) and the associated metadata are being developed nowadays. Due to the features of analytical systems used in metabolomics most of the data in metabolomics are of relative nature. It means that relative contributions of metabolites are of primary interest, not the absolute values obtained from raw signals with peaks. Nevertheless, this basic feature is not taken into account by standard statistical methodology, thus taking the raw data and any kind of normalization will lead to different results of the statistical analysis. A way out is to use the logratio approach to the statistical analysis of metabolomic (compositional) data that provides a reasonable statistical treatment of metabolomic observations. It results in easily interpretable coordinate systems, new variables, where any standard statistical analysis (up to slight modifications) can be performed. The aim of the project will be to form a concise methodology for statistical processing of metabolomics data using the logratio approach, involving pre-processing issues (like proper zero treatment), the corresponding processing and interpretation of all statistical methods, popular in metabolomics (principal component analysis, partial least squares, and others). We will focus also on recent approaches that improve relevance and interpretability of statistical processing (robustness, sparsity constraints). The developed methodologies will be tested on high-quality data matrices generated in compliance with strategies of Metabolomics Standards Initiative. The results will be disseminated through publications in high-quality peer-reviewed journals and contributions to conferences, not necessarily related strictly to metabolomics (chemometrics, statistical methodology) as the anticipated project outputs are expected to have a much broader impact than just for the primary research field of metabolomics. For this reason, one of the outputs will be also a new library in the open-source statistical software R, freely available for all interested researchers.

New statistical methods for the analysis of data from metabolomics have been developed. These methods make use of relative information in terms of ratios between the variable values of an observation. Thus they no longer depend on the absolute information of the measured values, which are among different samples usually not comparable due to the specifics how samples are taken. Moreover, the developed methods are robust against data outliers, i.e. unusual observations, even for single variables of an observation. All methods have been implemented in the free software environment R, and thus they are available for other problems where relative information is of main interest.