Mind your Ps and Qs – Caveats in metabolomics data analysis

Abstract

Metabolomics studies use high-throughput analytical platforms to measure metabolites in biological samples. These mass spectrometry and/or NMR spectroscopy platforms generate complex data sets, and the analysis of such data poses many challenges, in particular the high dimensionality with relatively fewer number of samples means that sophisticated statistical models are required to analyse these data and these models come with caveats. In this review, we discuss some of these common caveats associated with most popular statistical tests and models. We present common mistakes found in metabolomics data analysis, along with recommendations on how to avoid them. The aim of this review is to raise awareness of the potential risks of misusing or abusing statistical models, and to promote good practices for reliable and reproducible metabolomics research. A new form of permutation test with emphasis on assessing the statistical significance level of the effect captured by supervised model is also proposed.