From big data to big insights: statistical and bioinformatic approaches for exploring the lipidome

The goal of lipidomic studies is to provide a broad characterization of cellular lipids present and changing in a sample of interest. Recent lipidomic research has significantly contributed to revealing the multifaceted roles that lipids play in fundamental cellular processes, including signaling, energy storage, and structural support. Furthermore, these findings have shed light on how lipids dynamically respond to various perturbations. Continued advancement in analytical techniques has also led to improved abilities to detect and identify novel lipid species, resulting in increasingly large datasets. Statistical analysis of these datasets can be challenging not only because of their vast size, but also because of the highly correlated data structure that exists due to many lipids belonging to the same metabolic or regulatory pathways. Interpretation of these lipidomic datasets is also hindered by a lack of current biological knowledge for the individual lipids. These limitations can therefore make lipidomic data analysis a daunting task. To address these difficulties and shed light on opportunities and also weaknesses in current tools, we have assembled this review. Here, we illustrate common statistical approaches for finding patterns in lipidomic datasets, including univariate hypothesis testing, unsupervised clustering, supervised classification modeling, and deep learning approaches. We then describe various bioinformatic tools often used to biologically contextualize results of interest. Overall, this review provides a framework for guiding lipidomic data analysis to promote a greater assessment of lipidomic results, while understanding potential advantages and weaknesses along the way.

Statistical methods and resources for biomarker discovery using metabolomics

Metabolomics is a dynamic tool for elucidating biochemical changes in human health and disease. Metabolic profiles provide a close insight into physiological states and are highly volatile to genetic and environmental perturbations. Variation in metabolic profiles can inform mechanisms of pathology, providing potential biomarkers for diagnosis and assessment of the risk of contracting a disease. With the advancement of high-throughput technologies, large-scale metabolomics data sources have become abundant. As such, careful statistical analysis of intricate metabolomics data is essential for deriving relevant and robust results that can be deployed in real-life clinical settings. Multiple tools have been developed for both data analysis and interpretations. In this review, we survey statistical approaches and corresponding statistical tools that are available for discovery of biomarkers using metabolomics.

Pattern analysis of neuroimaging data reveals novel insights on threat learning and extinction in humans

Several decades of rodent neurobiology research have identified a network of brain regions that support Pavlovian threat conditioning and extinction, focused predominately on the amygdala, hippocampus, and medial prefrontal cortex (mPFC). Surprisingly, functional magnetic resonance imaging (fMRI) studies have shown inconsistent evidence for these regions while humans undergo threat conditioning and extinction. In this review, we suggest that translational neuroimaging efforts have been hindered by reliance on traditional univariate analysis of fMRI. Whereas univariate analyses average activity across voxels in a given region, multivariate pattern analyses (MVPA) leverage the information present in spatial patterns of activity. MVPA therefore provides a more sensitive analysis tool to translate rodent neurobiology to human neuroimaging. We review human fMRI studies using MVPA that successfully bridge rodent models of amygdala, hippocampus, and mPFC function during Pavlovian learning. We also highlight clinical applications of these information-sensitive multivariate analyses. In sum, we advocate that the field should consider adopting a variety of multivariate approaches to help bridge cutting-edge research on the neuroscience of threat and anxiety.

Univariate versus multivariate spectrophotometric methods for the simultaneous determination of omarigliptin and two of its degradation products

Six selective spectrophotometric techniques, four univariate and two multivariate ones were developed for the determination of the antidiabetic drug omarigliptin (OMR) along with its hydrolytic and oxidative degradation products. The proposed univariate spectrophotometric methods were ratio subtraction, first derivative, derivative ratio and ratio difference. Linearities were constructed in the range of 10.0-180.0 μg mL^-1 for both OMR & its hydrolytic degradation product and 10.0-110.0 μg mL^-1 for the oxidative degradation one. On the other hand, partial least squares and artificial neural networks were the chosen multivariate approaches. Their linearity ranges were 20.0-60.0 μg mL^-1 for OMR and 10.0-30.0 μg mL^-1 for the two degradation products. All the methods were validated, effectively applied for quantification of the intact drug in its tablet formulation and favorably compared to the reported one.

A Fast Method for Estimating Statistical Power of Multivariate GWAS in Real Case Scenarios: Examples from the Field of Imaging Genetics

In GWAS of imaging phenotypes (e.g., by the ENIGMA and CHARGE consortia), the growing number of phenotypes considered presents a statistical challenge that other fields are not experiencing (e.g. psychiatry and the Psychiatric Genetics Consortium). However, the multivariate nature of MRI measurements may also be an advantage as many of the MRI phenotypes are correlated and multivariate methods could be considered. Here, we compared the statistical power of a multivariate GWAS versus the current univariate approach, which consists of multiple univariate analyses. To do so, we used results from twin models to estimate pertinent vectors of SNP effect sizes on brain imaging phenotypes, as well as the residual correlation matrices, necessary to estimate analytically the statistical power. We showed that for subcortical structure volumes and hippocampal subfields, a multivariate GWAS yields similar statistical power to the current univariate approach. Our analytical approach is as accurate but ~ 1000 times faster than simulations and we have released the code to facilitate the investigation of other scenarios, may they be outside the field of imaging genetics.

New spectrophotometric/chemometric assisted methods for the simultaneous determination of imatinib, gemifloxacin, nalbuphine and naproxen in pharmaceutical formulations and human urine

In this paper, novel univariate and multivariate regression methods along with model-updating technique were developed and validated for the simultaneous determination of quaternary mixture of imatinib (IMB), gemifloxacin (GMI), nalbuphine (NLP) and naproxen (NAP). The univariate method is extended derivative ratio (EDR) which depends on measuring every drug in the quaternary mixture by using a ternary mixture of the other three drugs as divisor. Peak amplitudes were measured at 294nm, 250nm, 283nm and 239nm within linear concentration ranges of 4.0-17.0, 3.0-15.0, 4.0-80.0 and 1.0-6.0μgmL^-1 for IMB, GMI, NLP and NAB, respectively. Multivariate methods adopted are partial least squares (PLS) in original and derivative mode. These models were constructed for simultaneous determination of the studied drugs in the ranges of 4.0-8.0, 3.0-11.0, 10.0-18.0 and 1.0-3.0μgmL^-1 for IMB, GMI, NLP and NAB, respectively, by using eighteen mixtures as a calibration set and seven mixtures as a validation set. The root mean square error of predication (RMSEP) were 0.09 and 0.06 for IMB, 0.14 and 0.13 for GMI, 0.07 and 0.02 for NLP and 0.64 and 0.27 for NAP by PLS in original and derivative mode, respectively. Both models were successfully applied for analysis of IMB, GMI, NLP and NAP in their dosage forms. Updated PLS in derivative mode and EDR were applied for determination of the studied drugs in spiked human urine. The obtained results were statistically compared with those obtained by the reported methods giving a conclusion that there is no significant difference regarding accuracy and precision.

Antioxidants linked with physical, cognitive and psychological frailty: Analysis of candidate biomarkers and markers derived from the MARK-AGE study

Frailty among elderly people leads to an increased risk for negative health outcomes. To prevent frailty, we need a better understanding of the underlying mechanisms and early detection of individuals at risk. Both may be served by identifying candidate (bio)markers, i.e. biomarkers and markers, for the physical, cognitive, and psychological frailty domains. We used univariate (Rank-ANOVA) and multivariate (elastic net) approaches on the RASIG study population (age range: 35-74 years, n = 2220) of the MARK-AGE study to study up to 331 (bio)markers between individuals with and without frailty for each domain. Biomarkers and markers identified by both approaches were studied further regarding their association with frailty using logistic regression. Univariately, we found lower levels of antioxidants, including β-cryptoxanthin and zeaxanthin, in those who were physically, cognitively or psychologically frail. Additionally, self-reported health was worse in these three frail groups. Multivariately, we observed lower levels of β-cryptoxanthin and zeaxanthin in the cognitively frail. Levels of these carotenoids were inversely associated with the risk of being cognitively frail after adjusting for confounders. Antioxidants and self-reported health are potential (bio)markers to detect persons at risk of becoming frail. The biomarkers identified may indicate the involvement of inflammation in frailty, especially for physical and cognitive frailty.

Can Untargeted Metabolomics Be Utilized in Drug Discovery/Development?

Untargeted metabolomics is a promising approach for reducing the significant attrition rate for discovering and developing drugs in the pharmaceutical industry. This review aims to highlight the practical decision-making value of untargeted metabolomics for the advancement of drug candidates in drug discovery/development including potentially identifying and validating novel therapeutic targets, creating alternative screening paradigms, facilitating the selection of specific and translational metabolite biomarkers, identifying metabolite signatures for the drug efficacy mechanism of action, and understanding potential drug-induced toxicity. The review provides an overview of the pharmaceutical process workflow to discover and develop new small molecule drugs followed by the metabolomics process workflow that is involved in conducting metabolomics studies. The pros and cons of the major components of the pharmaceutical and metabolomics workflows are reviewed and discussed. Finally, selected untargeted metabolomics literature examples, from primarily 2010 to 2016, are used to illustrate why, how, and where untargeted metabolomics can be integrated into the drug discovery/preclinical drug development process.

Univariate and multivariate spatial models of health facility utilisation for childhood fevers in an area on the coast of Kenya

BACKGROUND

Precise quantification of health service utilisation is important for the estimation of disease burden and allocation of health resources. Current approaches to mapping health facility utilisation rely on spatial accessibility alone as the predictor. However, other spatially varying social, demographic and economic factors may affect the use of health services. The exclusion of these factors can lead to the inaccurate estimation of health facility utilisation. Here, we compare the accuracy of a univariate spatial model, developed only from estimated travel time, to a multivariate model that also includes relevant social, demographic and economic factors.

METHODS

A theoretical surface of travel time to the nearest public health facility was developed. These were assigned to each child reported to have had fever in the Kenya demographic and health survey of 2014 (KDHS 2014). The relationship of child treatment seeking for fever with travel time, household and individual factors from the KDHS2014 were determined using multilevel mixed modelling. Bayesian information criterion (BIC) and likelihood ratio test (LRT) tests were carried out to measure how selected factors improve parsimony and goodness of fit of the time model. Using the mixed model, a univariate spatial model of health facility utilisation was fitted using travel time as the predictor. The mixed model was also used to compute a multivariate spatial model of utilisation, using travel time and modelled surfaces of selected household and individual factors as predictors. The univariate and multivariate spatial models were then compared using the receiver operating area under the curve (AUC) and a percent correct prediction (PCP) test.

RESULTS

The best fitting multivariate model had travel time, household wealth index and number of children in household as the predictors. These factors reduced BIC of the time model from 4008 to 2959, a change which was confirmed by the LRT test. Although there was a high correlation of the two modelled probability surfaces (Adj R ² = 88%), the multivariate model had better AUC compared to the univariate model; 0.83 versus 0.73 and PCP 0.61 versus 0.45 values.

CONCLUSION

Our study shows that a model that uses travel time, as well as household and individual-level socio-demographic factors, results in a more accurate estimation of use of health facilities for the treatment of childhood fever, compared to one that relies on only travel time.

How to describe univariate data

Univariate analysis has the purpose to describe a single variable distribution in one sample. It is the first important step of every clinical trial. In this short review, we focus on this analysis, the methods that authors should use to report this type of data, information that they should not miss and mistakes that they must avoid.

Modeling individual heterogeneity in the acquisition of recurrent infections: an application to parvovirus B19

In recent years, it has been shown that individual heterogeneity in the acquisition of infectious diseases has a large impact on the estimation of important epidemiological parameters such as the (basic) reproduction number. Therefore, frailty modeling has become increasingly popular in infectious disease epidemiology. However, so far, using frailty models, it was assumed infections confer lifelong immunity after recovery, an assumption which is untenable for non-immunizing infections. Our work concentrates on refining the existing frailty models to encompass complexities of waning immunity and consequently recurrent infections while accounting for individual heterogeneity. Univariate and shared gamma frailty models, frequently used in practice, and correlated gamma frailty models that have proven to be a valuable alternative are considered. We show that incorrectly assuming lifelong immunity when applying frailty models introduces substantial bias in the estimation of both the baseline hazard and the frailty parameters, and consequently of the basic and effective reproduction number. We illustrate our work using cross-sectional serological data on parvovirus B19 (PVB19) from Belgium for which the link with varicella zoster virus is exploited.

On the interpretation of weight vectors of linear models in multivariate neuroimaging

The increase in spatiotemporal resolution of neuroimaging devices is accompanied by a trend towards more powerful multivariate analysis methods. Often it is desired to interpret the outcome of these methods with respect to the cognitive processes under study. Here we discuss which methods allow for such interpretations, and provide guidelines for choosing an appropriate analysis for a given experimental goal: For a surgeon who needs to decide where to remove brain tissue it is most important to determine the origin of cognitive functions and associated neural processes. In contrast, when communicating with paralyzed or comatose patients via brain-computer interfaces, it is most important to accurately extract the neural processes specific to a certain mental state. These equally important but complementary objectives require different analysis methods. Determining the origin of neural processes in time or space from the parameters of a data-driven model requires what we call a forward model of the data; such a model explains how the measured data was generated from the neural sources. Examples are general linear models (GLMs). Methods for the extraction of neural information from data can be considered as backward models, as they attempt to reverse the data generating process. Examples are multivariate classifiers. Here we demonstrate that the parameters of forward models are neurophysiologically interpretable in the sense that significant nonzero weights are only observed at channels the activity of which is related to the brain process under study. In contrast, the interpretation of backward model parameters can lead to wrong conclusions regarding the spatial or temporal origin of the neural signals of interest, since significant nonzero weights may also be observed at channels the activity of which is statistically independent of the brain process under study. As a remedy for the linear case, we propose a procedure for transforming backward models into forward models. This procedure enables the neurophysiological interpretation of the parameters of linear backward models. We hope that this work raises awareness for an often encountered problem and provides a theoretical basis for conducting better interpretable multivariate neuroimaging analyses.