Classifying groups of individuals based on their metabolic profile is one of the main topics in metabolomics research. Due to the low number of individuals compared to the large number of variables, this is not an easy task. PLSDA is one of the data analysis methods used for the classification. Unfortunately this method eagerly overfits the data and rigorous validation is necessary. The validation however is far from straightforward. Is this paper we will discuss a strategy based on cross model validation and permutation testing to validate the classification models. It is also shown that too optimistic results are obtained when the validation is not done properly. Furthermore, we advocate against the use of PLSDA score plots for inference of class differences.

Current research increasingly recognizes the human gut microbiome as a metabolically versatile biological 'digester' that plays an essential role in regulating the host metabolome. Gut microbiota recover energy and biologically active molecules from food that would otherwise be washed out of the intestinal tract without benefit. In this study, a protocol for NMR-based metabolite profiling has been developed to access the activity of the microbiome. The physicochemical properties of fecal metabolites have been found to strongly affect the reproducibility and coverage of the profiles obtained. Metabolite profiles generated by water and methanol extraction of lyophilized feces are reproducible and comprise a variety of different compounds including, among others, short-chain fatty acids (e.g. acetate, propionate, butyrate, isobutyrate, isovalerate, malate), organic acids (e.g. succinate, pyruvate, fumarate, lactate), amino acids, uracil, trimethylamine, ethanol, glycerol, glucose, phenolic acids, cholate, and lipid components. The NMR profiling approach was validated on fecal samples from a double-blinded, placebo-controlled, randomized cross-over study, in which healthy human subjects consumed a placebo and either a grape juice extract or a mix of grape juice and wine extract over a period of 4 weeks, each. The considerable inter- and intra-individual variability observed originates in the first instance from variable metabolite concentrations rather than from variable metabolite compositions, suggesting that different colonic flora share general biochemical characteristics metabolizing different substrates to specific metabolic patterns. Whereas the grape juice extract did not induce changes in the metabolite profiles as compared with the placebo, the mixture of grape juice and wine extract induced a reduction in isobutyrate, which may indicate that polyphenols are able to modulate the microbial ecology of the gut.

A new method is introduced for the analysis of 'omics' data derived from crossover designed drug or nutritional intervention studies. The method aims at finding systematic variations in metabolic profiles after a drug or nutritional challenge and takes advantage of the crossover design in the data. The method, which can be considered as a multivariate extension of a paired t test, generates different multivariate submodels for the between- and the within-subject variation in the data. A major advantage of this variation splitting is that each submodel can be analyzed separately without being confounded with the other variation sources. The power of the multilevel approach is demonstrated in a human nutritional intervention study which used NMR-based metabolomics to assess the metabolic impact of grape/wine extract consumption. The variations in the urine metabolic profiles are studied between and within the human subjects using the multilevel analysis. After variation splitting, multilevel PCA is used to investigate the experimental and biological differences between the subjects, whereas a multilevel PLS-DA model is used to reveal the net treatment effect within the subjects. The observed treatment effect is validated with cross model validation and permutations. It is shown that the statistical significance of the multilevel classification model ( p << 0.0002) is a major improvement compared to a ordinary PLS-DA model ( p = 0.058) without variation splitting. Finally, rank products are used to determine which NMR signals are most important in the multilevel classification model.

The purpose of this study was to compare the effects of black and green tea consumption on human metabolism. Seventeen healthy male volunteers consumed black tea, green tea, or caffeine in a randomized crossover study. Twenty-four-hour urine and blood plasma samples were analyzed by NMR-based metabonomics, that is, high-resolution 1H NMR metabolic profiling combined with multivariate statistics. Green and black tea consumption resulted in similar increases in urinary excretion of hippuric acid and 1,3-dihydroxyphenyl-2-O-sulfate, both of which are end products of tea flavonoid degradation by colonic bacteria. Several unidentified aromatic metabolites were detected in urine specifically after green tea intake. Interestingly, green and black tea intake also had a different impact on endogenous metabolites in urine and plasma. Green tea intake caused a stronger increase in urinary excretion of several citric acid cycle intermediates, which suggests an effect of green tea flavanols on human oxidative energy metabolism and/or biosynthetic pathways.