Edoardo Saccenti

Double-check: validation of diagnostic statistics for PLS-DA models in metabolomics studies

Partial Least Squares-Discriminant Analysis (PLS-DA) is a PLS regression method with a special binary ‘dummy’ y-variable and it is commonly used for classification purposes and biomarker selection in metabolomics studies. Several statistical approaches are currently in use to validate outcomes of PLS-DA analyses e.g. double cross validation procedures or permutation testing. However, there is a great inconsistency in the optimization and the assessment of performance of PLS-DA models due to many different diagnostic statistics currently employed in metabolomics data analyses. In this paper, properties of four diagnostic statistics of PLS-DA, namely the number of misclassifications (NMC), the Area Under the Receiver Operating Characteristic (AUROC), Q2 and Discriminant Q2 (DQ2) are discussed. All four diagnostic statistics are used in the optimization and the performance assessment of PLS-DA models of three different-size metabolomics data sets obtained with two different types of analytical platforms and with different levels of known differences between two groups: control and case groups. Statistical significance of obtained PLS-DA models was evaluated with permutation testing. PLS-DA models obtained with NMC and AUROC are more powerful in detecting very small differences between groups than models obtained with Q2 and Discriminant Q2 (DQ2). Reproducibility of obtained PLS-DA models outcomes, models complexity and permutation test distributions are also investigated to explain this phenomenon. DQ2 and Q2 (in contrary to NMC and AUROC) prefer PLS-DA models with lower complexity and require higher number of permutation tests and submodels to accurately estimate statistical significance of the model performance. NMC and AUROC seem more efficient and more reliable diagnostic statistics and should be recommended in two group discrimination metabolomic studies.

Reflections on univariate and multivariate analysis of metabolomics data

AbstractMetabolomics experiments usually result in a large quantity of data. Univariate and multivariate analysis techniques are routinely used to extract relevant information from the data with the aim of providing biological knowledge on the problem studied. Despite the fact that statistical tools like the t test, analysis of variance, principal component analysis, and partial least squares discriminant analysis constitute the backbone of the statistical part of the vast majority of metabolomics papers, it seems that many basic but rather fundamental questions are still often asked, like: Why do the results of univariate and multivariate analyses differ? Why apply univariate methods if you have already applied a multivariate method? Why if I do not see something univariately I see something multivariately? In the present paper we address some aspects of univariate and multivariate analysis, with the scope of clarifying in simple terms the main differences between the two approaches. Applications of the t test, analysis of variance, principal component analysis and partial least squares discriminant analysis will be shown on both real and simulated metabolomics data examples to provide an overview on fundamental aspects of univariate and multivariate methods.

Individual human phenotypes in metabolic space and time.

Differences between individual phenotypes are due both to differences in genotype and to exposure to different environmental factors. A fundamental contribution to the definition of the individual phenotype for clinical and therapeutic applications would come from a deeper understanding of the metabolic phenotype. The existence of unique individual metabolic phenotypes has been hypothesized, but the experimental evidence has been only recently collected. Analysis of individual phenotypes over the timescale of years shows that the metabolic phenotypes are largely invariant. The present work also supports the idea that the individual metabolic phenotype can also be considered a metagenomic entity that is strongly affected by both gut microbiome and host metabolic phenotype, the latter defined by both genetic and environmental contributions.

Geochemical and microbial community determinants of reductive dechlorination at a site biostimulated with glycerol.

Biostimulation is widely used to enhance reductive dechlorination of chlorinated ethenes in contaminated aquifers. However, the knowledge on corresponding biogeochemical responses is limited. In this study, glycerol was injected in an aquifer contaminated with cis-dichloroethene (cDCE), and geochemical and microbial shifts were followed for 265 days. Consistent with anoxic conditions and sulfate reduction after biostimulation, MiSeq 16S rRNA gene sequencing revealed temporarily increased relative abundance of Firmicutes, Bacteriodetes and sulfate reducing Deltaproteobacteria. In line with 13 C cDCE enrichment and increased Dehalococcoides mccartyi (Dcm) numbers, dechlorination was observed toward the end of the field experiment, albeit being incomplete with accumulation of vinyl chloride. This was concurrent with (i) decreased concentrations of dissolved organic carbon (DOC), reduced relative abundances of fermenting and sulfate reducing bacteria that have been suggested to promote Dcm growth by providing electron donor (H2 ) and essential corrinoid cofactors, (ii) increased sulfate concentration and increased relative abundance of Epsilonproteobacteria and Deferribacteres as putative oxidizers of reduced sulfur compounds. Strong correlations of DOC, relative abundance of fermenters and sulfate reducers, and dechlorination imply the importance of syntrophic interactions to sustain robust dechlorination. Tracking microbial and environmental parameters that promote/preclude enhanced reductive dechlorination should aid development of sustainable bioremediation strategies.

Impact of a wastewater treatment plant on microbial community composition and function in a hyporheic zone of a eutrophic river

The impact of the installation of a technologically advanced wastewater treatment plant (WWTP) on the benthic microbial community of a vinyl chloride (VC) impacted eutrophic river was examined two years before, and three and four years after installation of the WWTP. Reduced dissolved organic carbon and increased dissolved oxygen concentrations in surface water and reduced total organic carbon and total nitrogen content in the sediment were recorded in the post-WWTP samples. Pyrosequencing of bacterial 16S rRNA gene fragments in sediment cores showed reduced relative abundance of heterotrophs and fermenters such as Chloroflexi and Firmicutes in more oxic and nutrient poor post-WWTP sediments. Similarly, quantitative PCR analysis showed 1–3 orders of magnitude reduction in phylogenetic and functional genes of sulphate reducers, denitrifiers, ammonium oxidizers, methanogens and VC-respiring Dehalococcoides mccartyi. In contrast, members of Proteobacteria adapted to nutrient-poor conditions were enriched in post-WWTP samples. This transition in the trophic state of the hyporheic sediments reduced but did not abolish the VC respiration potential in the post-WWTP sediments as an important hyporheic sediment function. Our results highlight effective nutrient load reduction and parallel microbial ecological state restoration of a human-stressed urban river as a result of installation of a WWTP.

Allostasis and Resilience of the Human Individual Metabolic Phenotype

The urine metabotype of 12 individuals was followed over a period of 8-10 years, which provided the longest longitudinal study of metabolic phenotypes to date. More than 2000 NMR metabolic profiles were analyzed. The majority of subjects have a stable metabotype. Subjects who were exposed to important pathophysiological stressful conditions had a significant metabotype drift. When the stress conditions ceased, the original metabotypes were regained, while an irreversible stressful condition resulted in a permanent metabotype change. These results suggest that each individual occupies a well-defined region in the broad metabolic space, within which a limited degree of allostasis is permitted. The insurgence of significant stressful conditions causes a shift of the metabotype to another distinct region. The spontaneous return to the original metabolic region when the stressful conditions are removed suggests that the original metabotype has some degree of resilience. In this picture, precision medicine should aim at reinforcing the patients metabolic resilience, that is, his or her ability to revert to his or her specific metabotype rather than to a generic healthy one.

Probabilistic networks of blood metabolites in healthy subjects as indicators of latent cardiovascular risk.

The complex nature of the mechanisms behind cardiovascular diseases prevents the detection of latent early risk conditions. Network representations are ideally suited to investigate the complex interconnections between the individual components of a biological system that underlies complex diseases. Here, we investigate the patterns of correlations of an array of 29 metabolites identified and quantified in the plasma of 864 healthy blood donors and use a systems biology approach to define metabolite probabilistic networks specific for low and high latent cardiovascular risk. We adapted methods based on the likelihood of correlation and methods from information theory and combined them with resampling techniques. Our results show that plasma metabolite networks can be defined that associate with latent cardiovascular disease risk. The analysis of the networks supports our previous finding of a possible association between cardiovascular risk and impaired mitochondrial activity and highlights post-translational modifications (glycosilation and oxidation) of lipoproteins as a possible target-mechanism for early detection of latent cardiovascular risk.

Correlation patterns in experimental data are affected by normalization procedures: consequences for data analysis and network inference

Normalization is a fundamental step in data processing to account for the sample-to-sample variation observed in biological samples. However, data structure is affected by normalization. In this paper, we show how, and to what extent, the correlation structure is affected by the application of 11 different normalization procedures. We also discuss the consequences for data analysis and interpretation, including principal component analysis, partial least-squares discrimination, and the inference of metabolite-metabolite association networks.

Simplivariate models: uncovering the underlying biology in functional genomics data

One of the first steps in analyzing high-dimensional functional genomics data is an exploratory analysis of such data. Cluster Analysis and Principal Component Analysis are then usually the method of choice. Despite their versatility they also have a severe drawback: they do not always generate simple and interpretable solutions. On the basis of the observation that functional genomics data often contain both informative and non-informative variation, we propose a method that finds sets of variables containing informative variation. This informative variation is subsequently expressed in easily interpretable simplivariate components. We present a new implementation of the recently introduced simplivariate models. In this implementation, the informative variation is described by multiplicative models that can adequately represent the relations between functional genomics data. Both a simulated and two real-life metabolomics data sets show good performance of the method.

Comparison of 432 Pseudomonas strains through integration of genomic, functional, metabolic and expression data

Pseudomonas is a highly versatile genus containing species that can be harmful to humans and plants while others are widely used for bioengineering and bioremediation. We analysed 432 sequenced Pseudomonas strains by integrating results from a large scale functional comparison using protein domains with data from six metabolic models, nearly a thousand transcriptome measurements and four large scale transposon mutagenesis experiments. Through heterogeneous data integration we linked gene essentiality, persistence and expression variability. The pan-genome of Pseudomonas is closed indicating a limited role of horizontal gene transfer in the evolutionary history of this genus. A large fraction of essential genes are highly persistent, still non essential genes represent a considerable fraction of the core-genome. Our results emphasize the power of integrating large scale comparative functional genomics with heterogeneous data for exploring bacterial diversity and versatility.