Ewa Szymańska

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.

Gender-dependent associations of metabolite profiles and body fat distribution in a healthy population with central obesity: towards metabolomics diagnostics

Obesity is a risk factor for cardiovascular diseases and type 2 diabetes especially when the fat is accumulated to central depots. Novel biomarkers are crucial to develop diagnostics for obesity and related metabolic disorders. We evaluated the associations between metabolite profiles (136 lipid components, 12 lipoprotein subclasses, 17 low-molecular-weight metabolites, 12 clinical markers) and 28 phenotype parameters (including different body fat distribution parameters such as android (A), gynoid (G), abdominal visceral (VAT), subcutaneous (SAT) fat) in 215 plasma/serum samples from healthy overweight men (n=32) and women (n=83) with central obesity. (Partial) correlation analysis and partial least squares (PLS) regression analysis showed that only specific metabolites were associated to A:G ratio, VAT, and SAT, respectively. These association patterns were gender dependent. For example, insulin, cholesterol, VLDL, and certain triacylglycerols (TG 54:1-3) correlated to VAT in women, while in men VAT was associated with TG 50:1-5, TG 55:1, phosphatidylcholine (PC 32:0), and VLDL ((X)L). Moreover, multiple regression analysis revealed that waist circumference and total fat were sufficient to predict VAT and SAT in women. In contrast, only VAT but not SAT could be predicted in men and only when plasma metabolites were included, with PC 32:0 being most strongly associated with VAT. These findings collectively highlight the potential of metabolomics in obesity and that gender differences need to be taken into account for novel biomarker and diagnostic discovery for obesity and metabolic disorders.

Between Metabolite Relationships: an essential aspect of metabolic change

Not only the levels of individual metabolites, but also the relations between the levels of different metabolites may indicate (experimentally induced) changes in a biological system. Component analysis methods in current ‘standard’ use for metabolomics, such as Principal Component Analysis (PCA), do not focus on changes in these relations. We therefore propose the concept of ‘Between Metabolite Relationships’ (BMRs): common changes in the covariance (or correlation) between all metabolites in an organism. Such structural changes may indicate metabolic change brought about by experimental manipulation but which are lost with standard data analysis methods. These BMRs can be analysed by the INdividual Differences SCALing (INDSCAL) method. First the BMR quantification is described and subsequently the INDSCAL method. Finally, two studies illustrate the power and the applicability of BMRs in metabolomics. The first study is about the induced plant response of cabbage to herbivory, of which BMRs are a considerable part. In the second study—a human nutritional intervention study of green tea extract—standard data analysis tools did not reveal any metabolic change, although the BMRs were considerably affected. The presented results show that BMRs can be easily implemented in a wide variety of metabolomic studies. They provide a new source of information to describe biological systems in a way that fits flawlessly into the next generation of systems biology questions, dealing with personalized responses.

Data size reduction strategy for the classification of breath and air samples using multicapillary column-ion mobility spectrometry.

Ion mobility spectrometry combined with multicapillary column separation (MCC-IMS) is a well-known technology for detecting volatile organic compounds (VOCs) in gaseous samples. Due to their large data size, processing of MCC-IMS spectra is still the main bottleneck of data analysis, and there is an increasing need for data analysis strategies in which the size of MCC-IMS data is reduced to enable further analysis. In our study, the first untargeted chemometric strategy is developed and employed in the analysis of MCC-IMS spectra from 264 breath and ambient air samples. This strategy does not comprise identification of compounds as a primary step but includes several preprocessing steps and a discriminant analysis. Data size is significantly reduced in three steps. Wavelet transform, mask construction, and sparse-partial least squares-discriminant analysis (s-PLS-DA) allow data size reduction with down to 50 variables relevant to the goal of analysis. The influence and compatibility of the data reduction tools are studied by applying different settings of the developed strategy. Loss of information after preprocessing is evaluated, e.g., by comparing the performance of classification models for different classes of samples. Finally, the interpretability of the classification models is evaluated, and regions of spectra that are related to the identification of potential analytical biomarkers are successfully determined. This work will greatly enable the standardization of analytical procedures across different instrumentation types promoting the adoption of MCC-IMS technology in a wide range of diverse application fields.

Chemometrics for ion mobility spectrometry data: recent advances and future prospects

Historically, advances in the field of ion mobility spectrometry have been hindered by the variation in measured signals between instruments developed by different research laboratories or manufacturers. This has triggered the development and application of chemometric techniques able to reveal and analyze precious information content of ion mobility spectra. Recent advances in multidimensional coupling of ion mobility spectrometry to chromatography and mass spectrometry has created new, unique challenges for data processing, yielding high-dimensional, megavariate datasets. In this paper, a complete overview of available chemometric techniques used in the analysis of ion mobility spectrometry data is given. We describe the current state-of-the-art of ion mobility spectrometry data analysis comprising datasets with different complexities and two different scopes of data analysis, i.e. targeted and non-targeted analyte analyses. Two main steps of data analysis are considered: data preprocessing and pattern recognition. A detailed description of recent advances in chemometric techniques is provided for these steps, together with a list of interesting applications. We demonstrate that chemometric techniques have a significant contribution to the recent and great expansion of ion mobility spectrometry technology into different application fields. We conclude that well-thought out, comprehensive data analysis strategies are currently emerging, including several chemometric techniques and addressing different data challenges. In our opinion, this trend will continue in the near future, stimulating developments in ion mobility spectrometry instrumentation even further.

Individual differences in metabolomics: individualised responses and between-metabolite relationships

Many metabolomics studies aim to find ‘biomarkers’: sets of molecules that are consistently elevated or decreased upon experimental manipulation. Biological effects, however, often manifest themselves along a continuum of individual differences between the biological replicates in the experiment. Such differences are overlooked or even diminished by methods in standard use for metabolomics, although they may contain a wealth of information on the experiment. Properly understanding individual differences is crucial for generating knowledge in fields like personalised medicine, evolution and ecology. We propose to use simultaneous component analysis with individual differences constraints (SCA-IND), a data analysis method from psychology that focuses on these differences. This method constructs axes along the natural biochemical differences between biological replicates, comparable to principal components. The model may shed light on changes in the individual differences between experimental groups, but also on whether these differences correspond to, e.g., responders and non-responders or to distinct chemotypes. Moreover, SCA-IND reveals the individuals that respond most to a manipulation and are best suited for further experimentation. The method is illustrated by the analysis of individual differences in the metabolic response of cabbage plants to herbivory. The model reveals individual differences in the response to shoot herbivory, where two ‘response chemotypes’ may be identified. In the response to root herbivory the model shows that individual plants differ strongly in response dynamics. Thereby SCA-IND provides a hitherto unavailable view on the chemical diversity of the induced plant response, that greatly increases understanding of the system.

Increasing conclusiveness of clinical breath analysis by improved baseline correction of multi capillary column – ion mobility spectrometry (MCC-IMS) data

Current challenges of clinical breath analysis include large data size and non-clinically relevant variations observed in exhaled breath measurements, which should be urgently addressed with competent scientific data tools. In this study, three different baseline correction methods are evaluated within a previously developed data size reduction strategy for multi capillary column - ion mobility spectrometry (MCC-IMS) datasets. Introduced for the first time in breath data analysis, the Top-hat method is presented as the optimum baseline correction method. A refined data size reduction strategy is employed in the analysis of a large breathomic dataset on a healthy and respiratory disease population. New insights into MCC-IMS spectra differences associated with respiratory diseases are provided, demonstrating the additional value of the refined data analysis strategy in clinical breath analysis.

Assessing the metabolic effects of prednisolone in healthy volunteers using urine metabolic profiling

BackgroundGlucocorticoids, such as prednisolone, are widely used anti-inflammatory drugs, but therapy is hampered by a broad range of metabolic side effects including skeletal muscle wasting and insulin resistance. Therefore, development of improved synthetic glucocorticoids that display similar efficacy as prednisolone but reduced side effects is an active research area. For efficient development of such new drugs, in vivo biomarkers, which can predict glucocorticoid metabolic side effects in an early stage, are needed. In this study, we aim to provide the first description of the metabolic perturbations induced by acute and therapeutic treatments with prednisolone in humans using urine metabolomics, and to derive potential biomarkers for prednisolone-induced metabolic effects.MethodsA randomized, double blind, placebo-controlled trial consisting of two protocols was conducted in healthy men. In protocol 1, volunteers received placebo (n = 11) or prednisolone (7.5 mg (n = 11), 15 mg (n = 13) or 30 mg (n = 12)) orally once daily for 15 days. In protocol 2, volunteers (n = 6) received placebo at day 0 and 75 mg prednisolone at day 1. We collected 24 h urine and serum samples at baseline (day 0), after a single dose (day 1) and after prolonged treatment (day 15) and obtained mass-spectrometry-based urine and serum metabolic profiles.ResultsAt day 1, high-dose prednisolone treatment increased levels of 13 and 10 proteinogenic amino acids in urine and serum respectively, as well as levels of 3-methylhistidine, providing evidence for an early manifestation of glucocorticoid-induced muscle wasting. Prednisolone treatment also strongly increased urinary carnitine derivatives at day 1 but not at day 15, which might reflect adaptive mechanisms under prolonged treatment. Finally, urinary levels of proteinogenic amino acids at day 1 and of N-methylnicotinamide at day 15 significantly correlated with the homeostatic model assessment of insulin resistance and might represent biomarkers for prednisolone-induced insulin resistance.ConclusionThis study provides evidence that urinary metabolomics represents a noninvasive way of monitoring the effect of glucocorticoids on muscle protein catabolism after a single dose and can derive new biomarkers of glucocorticoid-induced insulin resistance. It might, therefore, help the development of improved synthetic glucocorticoids.Trial RegistrationClinicalTrials.gov NCT00971724

Weight randomization test for the selection of the number of components in PLS models

The selection of the optimal number of components remains a difficult but essential task in partial least squares (PLS). Randomization tests have the advantage of being automatic and they make use of the entire dataset, in contrary with the widely used cross‐validation approaches. Partial least squares modeling may include component(s) with a large amount of irrelevant data variation, and this might affect the model, depending on the assigned y‐loading (which is the regression coefficient in the latent domain). This has recently been indicated by us in the basic sequence framework with respect to the underlying theory of the PLS algorithm and presented to the chemometrics society. We will show in this work that this irrelevant data variation is the root cause of the difficulty in current methods for selecting the optimal number of components. For randomization tests, PLS models with nonsignificant components may result in false positive tests because of the incorrect assumption that “the components enter the model in a natural order”.

Boosting model performance and interpretation by entangling preprocessing selection and variable selection

The aim of data preprocessing is to remove data artifacts-such as a baseline, scatter effects or noise-and to enhance the contextually relevant information. Many preprocessing methods exist to deliver one or more of these benefits, but which method or combination of methods should be used for the specific data being analyzed is difficult to select. Recently, we have shown that a preprocessing selection approach based on Design of Experiments (DoE) enables correct selection of highly appropriate preprocessing strategies within reasonable time frames. In that approach, the focus was solely on improving the predictive performance of the chemometric model. This is, however, only one of the two relevant criteria in modeling: interpretation of the model results can be just as important. Variable selection is often used to achieve such interpretation. Data artifacts, however, may hamper proper variable selection by masking the true relevant variables. The choice of preprocessing therefore has a huge impact on the outcome of variable selection methods and may thus hamper an objective interpretation of the final model. To enhance such objective interpretation, we here integrate variable selection into the preprocessing selection approach that is based on DoE. We show that the entanglement of preprocessing selection and variable selection not only improves the interpretation, but also the predictive performance of the model. This is achieved by analyzing several experimental data sets of which the true relevant variables are available as prior knowledge. We show that a selection of variables is provided that complies more with the true informative variables compared to individual optimization of both model aspects. Importantly, the approach presented in this work is generic. Different types of models (e.g. PCR, PLS, …) can be incorporated into it, as well as different variable selection methods and different preprocessing methods, according to the taste and experience of the user. In this work, the approach is illustrated by using PLS as model and PPRV-FCAM (Predictive Property Ranked Variable using Final Complexity Adapted Models) for variable selection.