Search

Contains fulltext : 293314.pdf (Publisher’s version ) (Open Access)

The rapid evolution of the flow cytometry field, currently allowing the measurement of 30-50 parameters per cell, has led to a marked increase in deep multivariate information. Manual gating is insufficient to extract all this information. Therefore, multivariate analysis (MVA) methods have been developed to extract information and efficiently analyze the high-density multicolour flow cytometry (MFC) data. To aid interpretation, MFC data are often logarithmically transformed before MVA. We studied the consequences of different transformations of flow cytometry data in datasets containing negative intensities caused by background subtractions and spreading error, as logarithmic transformation of negative data is impossible. Transformations such as logicle or hyperbolic arcsine transformations allow linearity around zero, whereas higher (positive and negative) intensities are logarithmically transformed. To define the linear range, a parameter (or cofactor) must be chosen. We show how the chosen transformation parameter has great impact on the MVA results. In some cases, peak splitting is observed, producing two distributions around zero in an actual homogeneous population. This may be misinterpreted as the presence of multiple cell populations. Moreover, when performing arbitrary transformation before MVA analysis, biologically relevant and statistically significant information might be missed. We present a new algorithm, Optimal Transformation for flow cytometry data (OTflow), which uses various statistical methods to optimally choose the parameter of the transformation and prevent artifacts such as peak splitting. Arbitrary or unconsidered transformation can lead to wrong conclusions for the MVA cluster methods, dimensionality reduction methods, and classification methods. We recommend transformation of flow cytometry data by using OTflow-defined parameters estimated per channel, in order to prevent peak splitting and other artifacts in the data.

The long-term prediction performance of spectroscopic calibration models is a critical factor to monitor or control many production processes. Over time, new variations may emerge that deteriorate prediction performance. Therefore, models have to be maintained to retain or improve their prediction performance through time, requiring considerable resources and data. Maintenance should improve relevant predictions but also needs to be resource and cost efficient. Current approaches do not consider these trade-offs. We propose a new method to quantify the effectiveness and cost of model maintenance strategies based on historical data. Model performance over time for past, imminent and future samples is evaluated as these may react differently to maintenance. The model performance and required updating resources are translated into relative cost and benefit to compare strategies and determine optimal maintenance parameters. We used this method to evaluate a maintenance strategy that combines adding incoming samples to the calibration data with re-optimization of spectral preprocessing and modelling parameters. Continuously adding samples to the calibration data is shown to improve prediction performance and leads to more robust and generic models for emerging variations in all investigated data streams. Selectively adding incoming sample variations showed a reduced prediction performance but saves considerably in resources. Comparing model performance on the different sampling windows can also be used to determine an optimal updating frequency. This novel strategy to evaluate the expected performance and determine an optimal maintenance strategy is generally applicable and should lead to robust and consistently high prospective and/or retrospective model performance through time, which can be crucial for optimal operation and fault detection in industrial processes.

The use of soft-sensors in industry is becoming more popular, as they allow for the prediction of critical product qualities from process variables in real-time. The requirement for this that all process variables are dynamically synchronized is often not met. Although different methods for dynamically synchronizing process variables are reported, no critical comparison of these methods is available. In this study we show that the choice in synchronization method significantly influences a soft-sensor's accuracy. From the methods studied, median filtering using a moving window with a width of 168 minutes placed before the target times led to the highest sensor accuracy for the production plant studied, a method not reported in literature. This optimal width is remarkable, as the total processing time of the plant is 30 minutes. This suggests that changes in the physical state of the plant can affect the production quality than one might expect.

Understanding how different units of an industrial production plant are operationally related is key to improving production quality and sustainability. Data science has proven indispensable in obtaining such understanding from vast amounts of historical process data. Path modelling is a valuable statistical tool to obtain such information from historical production data. Investigating how relationships within a process are affected by multiple production conditions and their interactions can however provide an even deeper understanding of the plant’s daily operation. We therefore propose conditional path modelling as an approach to obtain such improved understanding, demonstrated for a milk protein powder production plant. For this plant we studied how the relationships between different production units and steps are dependent on factors like production line, different seasons and product quality range. We show how the interaction of such factors can be quantified and interpreted in context of daily plant operation. This analysis revealed an augmented insight into the process that can be readily placed in the context of the plant’s structure and behavior. Such insights can be vital to identify and improve upon shortcomings in current plant-wide monitoring and control routines.

Multicolour flow cytometry (MFC) is used to measure multiple cellular markers at the single-cell level. Cellular markers may be coloured with different panels of fluorescently-labelled antibodies to enable cell identification or the detection of activated cells in pre-defined, gated’ specific cell subsets. The number of markers that can be used per measurement is technologically limited however, requiring every panel to be analysed in a separate aliquot measurement. The combined analyses of these dedicated panels may enhance the predictive ability of these measurements and could enrich the interpretation of the immunological information. Here we introduce a fusion method for MFC data, based on DAMACY (Discriminant Analysis of Multi-Aspect Cytometry data), which can combine information from complementary panels. This approach leads to both enhanced predictions and clearer interpretations in comparison with the analysis of separate measurements. We illustrate this method using two datasets: the response of neutrophils evoked by a systemic endotoxin challenge and the activated immune status of the innate cells, T cells and B cells in obese versus lean individuals. The data fusion approach was able to detect cells that do not individually show a difference between clinical phenotypes but do play a role in combination with other cells. This research received funding from the Netherlands Organization for Scientific Research (NWO) in the framework of the Technology Area COAST of the Fund New Chemical Innovations. Functions for Matlab are available at http://www.ru.nl/science/analyticalchemistry/research/software/.

Chemical production processes benefit from intelligent data analysis. Previous work showed how process knowledge can be included in a structural equation modelling framework. While predictive models increase process value, currently available methods have limitations that hinder applicability to many (industrial) processes. This paper describes the Process PLS algorithm which can analyze multi-block, multistep and/or multidimensional processes. Process PLS was benchmarked on a simulated crude oil distillation process. Analysis of 22 empirical data sets from a production process at Nouryon illustrated how Process PLS solves limitations of PLS path modelling. In the analysis of the benchmark Val de Loire data, Process PLS revealed substantially meaningful effects which the recently proposed Sequential Orthogonalized PLS path modelling completely missed. Process PLS is a promising approach that enables data-driven analysis of process data using information on the complex process structure, to demonstrably increase insight in the underlying system, making model-based predictions much more valuable.

(Bio)chemical industrial batch reactions have to be terminated timely, to prevent waste of resources and decreased production quality due to prolonging the production when the (primary) reaction has already finished. Approaches for detecting endpoints on off-line and/or on-line analysis exist, but may be inaccurate for productions with high batch-to-batch variations. In this study, we present a novel multi-step strategy for endpoint detection named ENDBOSS (ENdpoint Detection using Batch-specific cOntrol Spaces of Spectroscopic data). This strategy is designed to have higher robustness against batch-to-batch variation than endpoint detection methods reported in literature and to be implemented for on-line monitoring. We demonstrate ENDBOSS on three industrially relevant reactions with high batch variations. A method for optimizing the settings of ENDBOSS for a given production process is proposed and demonstrated. The correlations between detected and reference endpoint were validated to be 0.96, 0.80 and 0.18 for the three demonstrator reactions. ENDBOSS has high performance for two reactions. For the third reaction, the size of the dataset was too limited, indicating that ENDBOSS does benefit from quantitative integration with strategic data collection. ENDBOSS is furthermore shown to outperform endpoint detection methods currently reported in literature for all demonstrator reactions.

Statistical modelling of industrial production data can lead to improved understanding of the process to benefit process monitoring and control routines. The production data required for such models need however to be synchronized in time, a topic sparsely covered in literature. We propose a strategy for data-driven automated optimization of dynamic synchronization of industrial production data, that optimizes the synchronization per process variable and can be applied for on-line monitoring in real-time. The strategy is tested and validated for two relevant production facilities, each of which has multiple production lines or configurations. For all lines and configurations, models predicting the production quality from process variables improved in accuracy using the presented per-variable optimization strategy. Although the prediction accuracy for two models would still be insufficient for real-time monitoring and control, process operators and engineers may still obtain novel process understanding from applying the presented strategy on these models. • Process variables need to be dynamically synchronized before multivariate statistical analysis • Different synchronization methods are optimal per process variables • A automated strategy to find the best synchronization method per process variable has been developed and tested • This per-variable optimization increases modelling accuracy for all five demonstrator processes

The 2000 European Union Water Framework Directive (WFD) states that ‘Member States shall ensure the necessary protection for the bodies of water identified with the aim of avoiding deteriorat...

Background Drug mass spectrometry imaging (MSI) data contain knowledge about drug and several other molecular ions present in a biological sample. However, a proper approach to fully explore the potential of such type of data is still missing. Therefore, a computational pipeline that combines different spatial and non-spatial methods is proposed to link the observed drug distribution profile with tumor heterogeneity in solid tumor. Our data analysis steps include pre-processing of MSI data, cluster analysis, drug local indicators of spatial association (LISA) map, and ions selection. Results The number of clusters identified from different tumor tissues. The spatial homogeneity of the individual cluster was measured using a modified version of our drug homogeneity method. The clustered image and drug LISA map were simultaneously analyzed to link identified clusters with observed drug distribution profile. Finally, ions selection was performed using the spatially aware method. Conclusions In this paper, we have shown an approach to correlate the drug distribution with spatial heterogeneity in untargeted MSI data. Our approach is freely available in an R package 'CorrDrugTumorMSI'.

By combining portable, handheld near-infrared (NIR) spectroscopy with state-of-the-art classification algorithms, we developed a powerful method to test chicken meat authenticity. The research presented shows that it is both possible to discriminate fresh from thawed meat, based on NIR spectra, as well as to correctly classify chicken fillets according to the growth conditions of the chickens with good accuracy. In all cases, the random subspace discriminant ensemble (RSDE) method significantly outperformed other common classification methods such as partial least squares-discriminant analysis (PLS-DA), artificial neural network (ANN) and support vector machine (SVM) with classification accuracy of >95%. This study shows that handheld NIR coupled with machine learning algorithms is a useful, fast, non-destructive tool to identify the authenticity of chicken meat. By comparing and combining different protocols to measure the NIR spectra (i.e., through packaging and directly on meat), we show the possibilities for both consumers and food inspection authorities to check the authenticity and origin of packaged chicken fillet.

Diffuse reflectance near-infrared (NIR) data (908–1676 nm) of chicken breast fillets was recorded in a non-destructive way using a portable miniaturised NIR spectrometer. The NIR data was used to discriminate between fresh and thawed breast fillets and to determine the birds’ growth conditions. NIR data was recorded of 153 commercial supermarket chicken fillet samples by applying the NIR device equipped with the standard issue collar on the samples in three different ways: (i) directly on the meat (ii) through the top foil of the package (i.e. with an air pocket between the foil and the breast fillet), and (iii) through the top foil with the packaging turned bottom up (i.e. no air pocket between the foil and the breast fillet). In order to generate thawed samples, the fresh samples were frozen and subsequently thawed. The freshness of the fillets was checked using β-hydroxyacyl-CoA-dehydrogenase of 13% of the sample set. Five NIR spectra were collected per measurement mode from each sample resulting in 4590 raw NIR spectra. Multivariate statistics was applied and the interpretation of these calculations can be found in Parastar et al. [1]. The NIR data has a reuse potential for follow-up studies of chicken breast fillet authentication using a similar brand NIR device or to serve as calibration transfer data. Keywords: Handheld near-infrared, Chicken breast fillet, Meat authenticity, Growth conditions, Chemometrics, Ensemble learning

Multicolor Flow Cytometry (MFC)-based gating allows the selection of cellular (pheno)types based on their unique marker expression. Current manual gating practice is highly subjective and may remove relevant information to preclude discovery of cell populations with specific co-expression of multiple markers. Only multivariate approaches can extract such aspects of cell variability from multi-dimensional MFC data. We describe the novel method ECLIPSE (Elimination of Cells Lying in Patterns Similar to Endogeneity) to identify and characterize aberrant cells present in individuals out of homeostasis. ECLIPSE combines dimensionality reduction by Simultaneous Component Analysis with Kernel Density Estimates. A Difference between Densities (DbD) is used to eliminate cells in responder samples that overlap in marker expression with cells of controls. Thereby, subsequent data analyses focus on the immune response-specific cells, leading to more informative and focused models. To prove the power of ECLIPSE, we applied the method to study two distinct datasets: the in vivo neutrophil response induced by systemic endotoxin challenge and in studying the heterogeneous immune-response of asthmatics. ECLIPSE described the well-characterized common response in the LPS challenge insightfully, while identifying slight differences between responders. Also, ECLIPSE enabled characterization of the immune response associated to asthma, where the co-expressions between all markers were used to stratify patients according to disease-specific cell profiles.

[EN] This article explores the potential of Kernel-Partial Least Squares (K-PLS) regression for the analysis of data proceeding from mixture designs of experiments. Gower's idea of pseudo-sample trajectories is exploited for interpretation purposes. The results show that, when the datasets under study are affected by severe nonlinearities and comprise few observations, the proposed approach can represent a feasible lternative to classical methodologies (i.e. Scheffe polynomial fitting by means of Ordinary Least Squares - OLS - and Cox polynomial fitting by means of Partial Least Squares - PLS). Furthermore, a way of recovering the parameters of a Scheffe model (provided that it holds and has the same complexity as the K-PLS one) from the trend of the aforementioned pseudo-sample trajectories is illustrated via a simulated case-study., This research work was partially supported by the Spanish Ministry of Economy and Competitiveness under the project DPI2014-55276-C5-1R and Shell Global Solutions International B.V. (Amsterdam, The Netherlands).

Chronic kidney disease (CKD) is a progressive pathological condition in which renal function deteriorates in time. The first diagnosis of CKD is often carried out in general care attention by general practitioners by means of serum creatinine (CNN) levels. However, it lacks sensitivity and thus, there is a need for new robust biomarkers to allow the detection of kidney damage particularly in early stages. Multivariate data analysis of plasma concentrations obtained from LC-QTOF targeted metabolomics method may reveal metabolites suspicious of being either up-regulated or down-regulated from urea cycle, arginine methylation and arginine-creatine metabolic pathways in CKD pediatrics and controls. The results show that citrulline (CIT), symmetric dimethylarginine (SDMA) and S-adenosylmethionine (SAM) are interesting biomarkers to support diagnosis by CNN: early CKD samples and controls were classified with an increase in classification accuracy of 18% when using these 4 metabolites compared to CNN alone. These metabolites together allow classification of the samples into a definite stage of the disease with an accuracy of 74%, being the 90% of the misclassifications one level above or below the CKD stage set by the nephrologists. Finally, sex-related, age-related and treatment-related effects were studied, to evaluate whether changes in metabolite concentration could be attributable to these factors, and to correct them in case a new equation is developed with these potential biomarkers for the diagnosis and monitoring of pediatric CKD.

The covariance matrix (or its inverse, the precision matrix) is central to many chemometric techniques. Traditional sample estimators perform poorly for high-dimensional data such as metabolomics data. Because of this, many traditional inference techniques break down or produce unreliable results. In this paper, we selectively review several modern estimators of the covariance and precision matrix that improve upon the traditional sample estimator. We focus on 3 general techniques: eigenvalue-shrinkage estimation, ridge-type estimation, and structured estimation. These methods rely on different assumptions regarding the structure of the covariance or precision matrix. Various examples, in particular using metabolomics data, are used to compare these techniques and to demonstrate that in concert with, eg, principal component analysis, multivariate analysis of variance, and Gaussian graphical models, better results are obtained.We selectively review modern estimators of the covariance and precision matrix focusing on 3 general techniques, namely, eigenvalue-shrinkage estimation, ridge-type estimation, and structured estimation. These methods rely on different structural assumptions of the covariance or precision matrix. Various examples, in particular using metabolomics data, are used to compare these techniques and to demonstrate that in concert with, eg, principal component analysis, multivariate analysis of variance, and Gaussian graphical models, better results are obtained.

The -omics technologies are becoming increasingly important in health care and are expected to contribute to personalized health care. In a typical experiment, cases and controls are compared as a two-class classification problem. This approach is often unsuitable, for example, because the classes are not well defined due to associated populations being biologically too heterogeneous. Recently, statistical health monitoring (SHM) was introduced as a complementary approach to allow for predictions at the individual level. This approach could be of use in all sorts of applications such as diagnosis of rare diseases, analysis of individual patterns in disease manifestation, disease monitoring, or personalized therapy.SHM uses the framework of statistical process monitoring (SPM) in a clinical setting. The method essentially combines estimation of Mahalanobis distances (MD) with principal component analysis (PCA) to evaluate the difference in the -omics data of an individual subject to a normal reference range (normal operating conditions). It is well known from SPM, however, that reliable identification of the variables primarily responsible for this difference is hampered by the smearing effect, which is a result of the PCA step. To avoid this problem, we propose to combine estimation of the MD with variable selection via an 11-norm penalty instead of using dimension reduction. This way a sparse MD metric is obtained.The effectiveness of this method is illustrated by several simulation studies and its application to urine H-1-NMR metabolomics data for diagnosis of multiple inborn errors of metabolism.

Combining the individual analytical strengths of mass spectrometry and infrared spectroscopy, infrared ion spectroscopy is increasingly recognized as a powerful tool for small-molecule identification in a wide range of analytical applications. Mass spectrometry is itself a leading analytical technique for small-molecule identification on the merit of its outstanding sensitivity, selectivity and versatility. The foremost shortcoming of the technique, however, is its limited ability to directly probe molecular structure, especially when contrasted against spectroscopic techniques. In infrared ion spectroscopy, infrared vibrational spectra are recorded for mass-isolated ions and provide a signature that can be matched to reference spectra, either measured from standards or predicted using quantum-chemical calculations. Here we present an overview of the potential for this technique to develop into a versatile analytical method for identifying molecular structures in mass spectrometry-based analytical workflows. In this tutorial perspective, we introduce the reader to the technique of infrared ion spectroscopy and highlight a selection of recent experimental advances and applications in current analytical challenges, in particular in the field of untargeted metabolomics. We report on the coupling of infrared ion spectroscopy with liquid chromatography and present experiments that serve as proof-of-principle ex- amples of strategies to address outstanding challenges.

Flow Cytometry is an analytical technology to simultaneously measure multiple markers per single cell. Ten thousands to millions of single cells can be measured per sample and each sample may contain a different number of cells. All samples may be bundled together, leading to a ‘multi-set’ structure. Many multivariate methods have been developed for Flow Cytometry data but none of them considers this structure in their quantitative handling of the data. The standard pre-processing used by existing multivariate methods provides models mainly influenced by the samples with more cells, while such a model should provide a balanced view of the biomedical information within all measurements. We propose an alternative ‘multi-set’ preprocessing that corrects for the difference in number of cells measured, balancing the relative importance of each multi-cell sample in the data while using all data collected from these expensive analyses. Moreover, one case example shows how multi-set pre-processing may benefit removal of undesired measurement-to-measurement variability and another where class-based multi-set pre-processing enhances the studied response upon comparison to the control reference samples. Our results show that adjusting data analysis algorithms to consider this multi-set structure may greatly benefit immunological insight and classification performance of Flow Cytometry data.

Platforms like metabolomics provide an unprecedented view on the chemical versatility in biomedical samples. Many diseases reflect themselves as perturbations in specific metabolite combinations. Multivariate analyses are essential to detect such combinations and associate them to specific diseases. For this, usually targeted discriminations of samples associated to a specific disease from non-diseased control samples are used. Such targeted data interpretation may not respect the heterogeneity of metabolic responses, both between diseases and within diseases. Here we show that multivariate methods that find any set of perturbed metabolites in a single patient, may be employed in combination with data collected with a single metabolomics technology to simultaneously investigate a large array of diseases. Several such untargeted data analysis approaches have been already proposed in other fields to find both expected and unexpected perturbations, e.g. in Statistical Process Control. We have critically compared several of these approaches for their sensitivity and their correct identification of the specifically perturbed metabolites. Also a new approach is introduced for this purpose. The newly introduced Sparse Mean approach, which we find here as most sensitive and best able to identify the specifically perturbed metabolites, turns metabolomics into an untargeted diagnostic platform. Aside from metabolomics, the proposed approach may greatly benefit fault diagnosis with untargeted analyses in many other fields, such as Industrial Process Control, food Adulteration Detection, and Intrusion Detection.

The constant development of analytical techniques leads to an increase in the amount of information available to describe phenomena in life science. In parallel, the inherent complexity of life science makes it almost impossible to obtain a comprehensive description using only one technical modality. Therefore, it became very popular to combine several biological or technical platforms/modalities to obtain a better understanding of the underlying problems. Merging different types of measurements/platforms into a single analysis is, however, a complex topic. Combining various platforms into single analysis is defined as data fusion. We describe here different types of data fusion strategies: the well-established low-, mid-, and high-level data fusion and the more recently introduced sustainable mid-level data fusion and kernel-based data fusion. For each type, we provide a detailed description. To illustrate these various data fusion approaches, we rely on four real data sets, namely, exhaled breath data of patients with Crohn disease (CD) obtained by gas chromatography–mass spectrometry (GC-MS), 454 pyrosequencing microbiome data of patients with CD, and metabolic profiling of beer brands by GC-MS and positive and negative ion modes of liquid chromatography.

Contains fulltext : 161386.pdf (Publisher’s version ) (Open Access) 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.

Background Genomic prediction (GP) allows breeders to select plants and animals based on their breeding potential for desirable traits, without lengthy and expensive field trials or progeny testing. We have proposed to use Dissimilarity-based Partial Least Squares (DPLS) for GP. As a case study, we use the DPLS approach to predict Bacterial wilt (BW) in tomatoes using SNPs as predictors. The DPLS approach was compared with the Genomic Best-Linear Unbiased Prediction (GBLUP) and single-SNP regression with SNP as a fixed effect to assess the performance of DPLS. Results Eight genomic distance measures were used to quantify relationships between the tomato accessions from the SNPs. Subsequently, each of these distance measures was used to predict the BW using the DPLS prediction model. The DPLS model was found to be robust to the choice of distance measures; similar prediction performances were obtained for each distance measure. DPLS greatly outperformed the single-SNP regression approach, showing that BW is a comprehensive trait dependent on several loci. Next, the performance of the DPLS model was compared to that of GBLUP. Although GBLUP and DPLS are conceptually very different, the prediction quality (PQ) measured by DPLS models were similar to the prediction statistics obtained from GBLUP. A considerable advantage of DPLS is that the genotype-phenotype relationship can easily be visualized in a 2-D scatter plot. This so-called score-plot provides breeders an insight to select candidates for their future breeding program. Conclusions DPLS is a highly appropriate method for GP. The model prediction performance was similar to the GBLUP and far better than the single-SNP approach. The proposed method can be used in combination with a wide range of genomic dissimilarity measures and genotype representations such as allele-count, haplotypes or allele-intensity values. Additionally, the data can be insightfully visualized by the DPLS model, allowing for selection of desirable candidates from the breeding experiments. In this study, we have assessed the DPLS performance on a single trait.

To make industrial processes lean, inclusion of technical process information is required into statistical modelling. Understanding how parts of a process are related to other parts and to output quality is key to understanding and controlling processes. In this work, we show how PLS path modelling can be used to incorporate process knowledge into predictive chemical process analysis. The result is a wealth of information which is not obtained by standard data analytic techniques commonly used by analytical chemists or process engineers. By comparing model parameters across multiple data sets from different batches of the same process, model parameters could be used as soft sensors. Some variables which would normally be discarded as uninformative were highly predictive of production costs. The methodology reported here improves chemical process understanding through the analysis of complementary historical process data, which may serve as the basis for development of improved process conditions and control.

Spectral-spatial classification of hyperspectral images has been the subject of many studies in recent years. When there are only a few labeled pixels for training and a skewed class label distribution, this task becomes very challenging because of the increased risk of overfitting when training a classifier. In this paper, we show that in this setting, a convolutional neural network with a single hidden layer can achieve state-of-the-art performance when three tricks are used: a spectral-locality-aware regularization term and smoothing- and label-based data augmentation. The shallow network architecture prevents overfitting in the presence of many features and few training samples. The locality-aware regularization forces neighboring wavelengths to have similar contributions to the features generated during training. The new data augmentation procedure favors the selection of pixels in smaller classes, which is beneficial for skewed class label distributions. The accuracy of the proposed method is assessed on five publicly available hyperspectral images, where it achieves state-of-the-art results. As other spectral-spatial classification methods, we use the entire image (labeled and unlabeled pixels) to infer the class of its unlabeled pixels. To investigate the positive bias induced by the use of the entire image, we propose a new learning setting where unlabeled pixels are not used for building the classifier. Results show the beneficial effect of the proposed tricks also in this setting and substantiate the advantages of using labeled and unlabeled pixels from the image for hyperspectral image classification.

The interpretation of estimates of model parameters in terms of biological information is often just as important as the predictions of the model itself. In this study we consider the identification of metabolites in a possibly biologically heterogeneous case group that show abnormal patterns with respect to a set of (healthy) control observations. For this purpose, we filter normal (baseline) natural variation from the data by projection of the data on a control sample model: the residual approach. This step should more easily highlight the abnormal metabolites. Interpretation is, however, hindered by a problem we named the ‘residual bias’ effect, which may lead to the identification of the wrong metabolites as ‘abnormal’. This effect is related to the smearing effect. We propose to alleviate residual bias by considering a weighted average of the filtered and raw data. This way, a compromise is found between excluding irrelevant natural variation from the data and the amount of residual bias that occurs. We show for simulated and real-world examples that this compromise may outperform inspection of the raw or filtered data. The method holds promise in numerous applications such as disease diagnoses, personalized healthcare, and industrial process control.

GC-MS/O (gas chromatography-mass spectrometry/olfactometry) is an indispensable technique to associate individual volatile odorants to odors perceived by human assessors. Interpretation of GC-MS/O data is, however, hampered in practice by different factors related to the instrumental set-up and by heterogeneity among odor descriptions given by the assessors (olfactometer). In this paper, a novel automated approach is presented, which deals with these GC-MS/O challenges and enables visualization and interpretation of GC-MS/O data. It includes signal warping via COW (correlation optimized warping), synchronizing MS and O data via detection of odor areas and construction of a TOC (total odor count) to visualize odor heterogeneity, respectively. Our approach is implemented in practice, and we successfully associated odors to compounds in data sets of two alcoholic beverages with different flavor compositions. It leads to a faster and less biased association of odors to compounds compared to current practice, reducing the time and effort needed for interpreting GC-MS/O data.

Principal Component Analysis (PCA) is widely used in analytical chemistry, to reduce the dimensionality of a multivariate data set in a few Principal Components (PCs) that summarize the predominant patterns in the data. An accurate estimate of the number of PCs is indispensable to provide meaningful interpretations and extract useful information. We show how existing estimates for the number of PCs may fall short for datasets with considerable coherence, noise or outlier presence. We present here how Angle Distribution of the Loading Subspaces (ADLS) can be used to estimate the number of PCs based on the variability of loading subspace across bootstrap resamples. Based on comprehensive comparisons with other well-known methods applied on simulated dataset, we show that ADLS (1) may quantify the stability of a PCA model with several numbers of PCs simultaneously; (2) better estimate the appropriate number of PCs when compared with the cross-validation and scree plot methods, specifically for coherent data, and (3) facilitate integrated outlier detection, which we introduce in this manuscript. We, in addition, demonstrate how the analysis of different types of real-life spectroscopic datasets may benefit from these advantages of ADLS.

Introduction Metabolomics investigates biochemical processes directly, potentially complementing transcriptomics and proteomics in providing insight into treatment outcome. Objectives This study aimed to use magnetic resonance (MR) spectroscopy on breast tumor tissue to explore the effect of neoadjuvant therapy on metabolic profiles, determine metabolic effects of the antiangiogenic drug bevacizumab, and investigate metabolic differences between responders and non-responders. Methods Breast tumors from 122 patients were profiled using high resolution magic angle spinning MR spectroscopy. All patients received neoadjuvant chemotherapy, and were randomized to receive bevacizumab or not. Tumors were biopsied prior, during, and after treatment. Results Principal component analysis showed clear metabolic changes indicating a decline in glucose consumption and a transition to normal breast adipose tissue as an effect of chemotherapy. Partial least squares-discriminant analysis revealed metabolic differences between pathological minimal residual disease patients and pathological non-responders after treatment (accuracy of 77%, p

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". In this work, we introduce a new randomization test, weight randomization test, selection of the optimal number of components in PLS in light of the underlying theory of the PLS algorithm. In the proposed method the null distribution is well characterized and efficiently determined taking into account a newly defined model quality metric: the number of consecutive non-significant components (CNC). We illustrate the effectiveness of weight randomization test in optimization of preprocessing as well as in classification models, where results are compared with the double cross-validation procedure for the latter. This is an important step towards the full automation of PLS model development and routine updates.

Multicolour Flow Cytometry (MFC) produces multidimensional analytical data on the quantitative expression of multiple markers on single cells. This data contains invaluable biomedical information on (1) the marker expressions per cell, (2) the variation in such expression across cells, (3) the variability of cell marker expression across samples that (4) may vary systematically between cells collected from donors and patients. Current conventional and even advanced data analysis methods for MFC data explore only a subset of these levels. The Discriminant Analysis of MultiAspect CYtometry (DAMACY) we present here provides a comprehensive view on health and disease responses by integrating all four levels. We validate DAMACY by using three distinct datasets: in vivo response of neutrophils evoked by systemic endotoxin challenge, the clonal response of leukocytes in bone marrow of acute myeloid leukaemia (AML) patients, and the complex immune response in blood of asthmatics. DAMACY provided good accuracy 91–100% in the discrimination between health and disease, on par with literature values. Additionally, the method provides figures that give insight into the marker expression and cell variability for more in-depth interpretation, that can benefit both physicians and biomedical researchers to better diagnose and monitor diseases that are reflected by changes in blood leukocytes.

Revealing the biochemistry associated to micro-organismal interspecies interactions is highly relevant for many purposes. Each pathogen has a characteristic metabolic fingerprint that allows identification based on their unique multivariate biochemistry. When pathogen species come into mutual contact, their co-culture will display a chemistry that may be attributed both to mixing of the characteristic chemistries of the mono-cultures and to competition between the pathogens. Therefore, investigating pathogen development in a polymicrobial environment requires dedicated chemometric methods to untangle and focus upon these sources of variation. The multivariate data analysis method Projected Orthogonalised Chemical Encounter Monitoring (POCHEMON) is dedicated to highlight metabolites characteristic for the interaction of two micro-organisms in co-culture. However, this approach is currently limited to a single time-point, while development of polymicrobial interactions may be highly dynamic. A well-known multivariate implementation of Analysis of Variance (ANOVA) uses Principal Component Analysis (ANOVA-PCA). This allows the overall dynamics to be separated from the pathogen-specific chemistry to analyse the contributions of both aspects separately. For this reason, we propose to integrate ANOVA-PCA with the POCHEMON approach to disentangle the pathogen dynamics and the specific biochemistry in interspecies interactions. Two complementary case studies show great potential for both liquid and gas chromatography - mass spectrometry to reveal novel information on chemistry specific to interspecies interaction during pathogen development.

The calibration performance of Partial Least Squares regression (PLS) can be improved by eliminating uninformative variables. For PLS, many variable elimination methods have been developed. One is the Uninformative-Variable Elimination for PLS (UVE-PLS). However, the number of variables retained by UVE-PLS is usually still large. In UVE-PLS, variable elimination is repeated as long as the root mean squared error of cross validation (RMSECV) is decreasing. The set of variables in this first local minimum is retained. In this paper, a modification of UVE-PLS is proposed and investigated, in which UVE is repeated until no further reduction in variables is possible, followed by a search for the global RMSECV minimum. The method is called Global-Minimum Error Uninformative-Variable Elimination for PLS, denoted as GME-UVE-PLS or simply GME-UVE. After each iteration, the predictive ability of the PLS model, built with the remaining variable set, is assessed by RMSECV. The variable set with the global RMSECV minimum is then finally selected. The goal is to obtain smaller sets of variables with similar or improved predictability than those from the classical UVE-PLS method. The performance of the GME-UVE-PLS method is investigated using four data sets, i.e. a simulated set, NIR and NMR spectra, and a theoretical molecular descriptors set, resulting in twelve profile-response (X-y) calibrations. The selective and predictive performances of the models resulting from GME-UVE-PLS are statistically compared to those from UVE-PLS and 1-step UVE, one-sided paired t-tests. The results demonstrate that variable reduction with the proposed GME-UVE-PLS method, usually eliminates significantly more variables than the classical UVE-PLS, while the predictive abilities of the resulting models are better. With GME-UVE-PLS, a lower number of uninformative variables, without a chemical meaning for the response, may be retained than with UVE-PLS. The selectivity of the classical UVE method thus can be improved by the application of the proposed GME-UVE method resulting in more parsimonious models.

An important part of the current hazard identification of novel plant varieties is comparative targeted analysis of the novel and reference varieties. Comparative analysis will become much more informative with unbiased analytical approaches, e.g. omics profiling. Data analysis estimating the similarity of new varieties to a reference baseline class of known safe varieties would subsequently greatly facilitate hazard identification. Further biological and eventually toxicological analysis would then only be necessary for varieties that fall outside this reference class. For this purpose, a one-class classifier tool was explored to assess and classify transcriptome profiles of potato (Solanum tuberosum) varieties in a model study. Profiles of six different varieties, two locations of growth, two year of harvest and including biological and technical replication were used to build the model. Two scenarios were applied representing evaluation of a 'different' variety and a 'similar' variety. Within the model higher class distances resulted for the 'different' test set compared with the 'similar' test set. The present study may contribute to a more global hazard identification of novel plant varieties.

Item does not contain fulltext

Despite gaining popularity and success in many modeling applications, Partial Least Squares (PLS) regression continues to provide challenges in the evaluation of important variables. This article describes the relationship between the regression coefficients and orthogonally decomposed variances in PLS. The relation between prediction, model interpretation, and important variable determination is described using the theory of the basic sequence presented here as a special case of the famous Krylov sequence (or the power method). Variable selection methods e.g. Selectivity Ratio (SR) and Variable Importance in the Projection (VIP) are also described in this framework. We show that the interpretation can be affected by unnecessary rotation toward the main source of variance in the X -block. Significance Multivariate Correlation (sMC) is developed using the knowledge obtained from the basic sequence to minimize the effect of irrelevant X -structures. Simultaneously sMC highlights the variables most correlated to the response. The performance of sMC is demonstrated, using simulated and real datasets, against commonly used variable selection methods, such as the Variable Importance in the Projection and Selectivity Ratio.

Contains fulltext : 128372.pdf (Publisher’s version ) (Closed access)