Search

In nutrition and health research, untargeted metabolomics is actually analyzed simultaneously with clinical data to improve prediction and better understand pathological status. This can be modeled using a multiblock supervised model with several input data blocks (metabolomics, clinical data) being potential predictors of the outcome to be explained. Alternatively, this configuration can be represented with a path diagram where the input blocks are each connected by links directed to the outcome—as in multiblock supervised modeling—and are also related to each other, thus allowing one to account for block effects. On the basis of a path model, we show herein how to estimate the effect of an input block, either on its own or conditionally to other(s), on the output response, respectively called “global” and “partial” effects, by percentages of explained variance in dedicated PLS regression models. These effects have been computed in two different path diagrams in a case study relative to metabolic syndrome, involving metabolomics and clinical data from an older men′s cohort (NuAge). From the two effects associated with each path, the results highlighted the complementary information provided by metabolomics to clinical data and, reciprocally, in the metabolic syndrome exploration.

Check-all-that-apply (CATA) has become a popular method for obtaining a consumer-based sensory characterization. In most case studies, consumers are also asked to evaluate the set of products according to a liking scale with the aim to identify the key sensory attributes associated with the most liked, or disliked, products. The common approach consists, first, in the identification of consumer segments based on the preference profiles. Thereafter, the analysis of the CATA responses is performed within each segment. Our purpose herein is to investigate different ways to simultaneously identify clusters of preference profiles while taking into account the CATA attributes. These approaches are derived from strategies already proposed by the different co-authors, namely: Fuzzy Clusterwise Regression (FCR), Clustering around Latent Variables (CLV) approach with external data, CLUSCATA-liking and CLV3W. The first two approaches involve the aggregation of the individual CATA data into a contingency table, while the last two ones deal with the combination of liking and CATA data at the individual level. These four strategies are illustrated on the basis of a real case study. Results are compared with respect to cluster stability together with interpretability of liking profiles within each segment. The stability of the results, assessed by bootstrapping, differed according to the strategy used. Moreover, working at the individual level or with combined data lead to a somewhat different segmentation of the panel of consumers.

International audience; A new approach for the analysis of the data and the clustering of the subjects in a Check All That Apply (CATA) experiment is outlined. It encompasses indices to assess the agreements among the subjects. These indices are taken into account for the analysis of the CATA data and the segmentation of the subjects. The analysis of the CATA data bears some similarities to the STATIS method and the cluster analysis of the subjects follows the same pattern as a clustering approach, called CLV, for the cluster analysis of variables, and a clustering approach, called CLUSTATIS, for clustering datasets.

Humans are exposed daily to complex mixtures of chemical pollutants through their environment and diet, some of which have the potential to disrupt the bodies’ natural endocrine functions and contribute to reproductive diseases like endometriosis. Increasing epidemiological and experimental evidence supports the association between endometriosis and certain persistent organic pollutants (POPs) like dioxins; however, little is known about the underlying linking mechanisms. The main objective of this study is to proof the methodological applicability and discovery potential of integrating ultra-trace mass spectrometry (MS) profiling of POP biomarkers and endogenous biomarker profiling (MS metabolomics and cytokines) in a case-control study for the etiological research of endometriosis. The approach is applied in a pilot clinical-based study conducted in France where women with and without surgically confirmed endometriosis were recruited. Serum samples were analysed with high-resolution MS for about 30 polychlorinated biphenyls (PCBs), organochlorinated pesticides and perfluoroalkyl substances (PFAS). About 600 serum metabolites and lipids were identified with targeted metabolomics using tandem MS with the Biocrates MxP® Quant 500 Kit. A panel of 4 pro-inflammatory cytokines were analysed using ELISA-based 4-PLEX analyser. Statistical analysis included a battery of variable selection approaches, multivariate logistic regression for single-chemical associations, Bayesian kernel machine regressions (BKMR) to identify mixture effects of POPs and a multiblock approach to identify shared biomarker signatures among high risk clusters. The results showed the positive associations between some POPs and endometriosis risk, including the pesticide trans-nonachlor Odds Ratio (95% Confidence Interval) 3.38 (2.06–5.98), p

International audience; In the field of clustering techniques, little attention has been paid to the recovery of a set of clusters from the structure of an additive tree. To bridge this gap, this work presents an original partitioning technique which aims to reveal clusters from an additive tree that represents a large set of objects. Specifically, an algorithm that splits a tree into successive subtrees was developed, based on a ratio of the lengths of edges. The stability of the clusters obtained with this technique was then evaluated using measurements of cohesion and isolation that were generated using a bootstrapping strategy. Finally, the degree of association of each object to clusters was analyzed to gain insight into their internal structure. This analysis was performed on the results of a sorting task conducted by 156 subjects, who were asked to sort 96 terms associated with the odor of wine. The methodology developed in this paper represents an innovative way to highlight groups of terms within a large set of wine odor attributes, with the ultimate goal being to improve the structure of the lexicon.

International audience; The set-up of comprehensive studies in life sciences involving a longitudinal dimension-as appears in time-scale metabolomics-calls for the use of dimension reduction techniques for three-way data structures (e.g., samples by variables by time points). For this purpose, a clustering around latent variables for three-way data approach,CLV3W, has been proposed.CLV3Waims at both partitioning the variables into nonoverlapping clusters and estimating within each cluster a rank-one Parafac model consisting of a latent component (resp. a weighting system) associated with the first mode (resp. third mode) and a vector of loadings reflecting the degree of closeness of each variable of the second mode to its cluster. In this paper, two constrainedCLV3Wmodels are discussed. First, a nonnegativity constraint is defined implying that clusters are composed of positively correlated variables. Second, it is proposed to constrain the weighting system to be the same for all clusters. These two constraints aim at providing more parsimonious models with configurations that are easier to interpret. The appropriateness of both constraints is evaluated in a simulation study and illustrated on two case studies pertaining to sensory evaluation and metabolomics data. Regarding the first case study,CLV3Wyields the identification of two consumer segments together with one common emotional pleasantness dimension associated with coffee aromas.CLV3Wanalysis of human preterm breast milk metabolomics data provided three clusters of lipid species that are responsible for specific functions (i.e., milk fat globules membrane-constituents, fatty acid oxidation-products, lipid mediators as eicosanoids and endocannabinoids).

Microbiological spoilage of meat is considered as a process which involves mainly bacterial metabolism leading to degradation of meat sensory qualities. Studying spoilage requires the collection of different types of experimental data encompassing microbiological, physicochemical and sensorial measurements. Within this framework, the objective herein was to carry out a multiblock path modelling workflow to decipher causality relationships between different types of spoilage-related responses: composition of microbiota, volatilome and off-odour profiles. Analyses were performed with the Path-ComDim approach on a large-scale dataset collected on fresh turkey sausages. This approach enabled to quantify the importance of causality relationships determined a priori between each type of responses as well as to identify important responses involved in spoilage, then to validate causality assumptions. Results were very promising: the data integration confirmed and quantified the causality between data blocks, exhibiting the dynamical nature of spoilage, mainly characterized by the evolution of off-odour profiles caused by the production of volatile organic compounds such as ethanol or ethyl acetate. This production was possibly associated with several bacterial species like Lactococcus piscium, Leuconostoc gelidum, Psychrobacter sp. or Latilactobacillus fuchuensis. Likewise, the production of acetoin and diacetyl in meat spoilage was highlighted. The Path-ComDim approach illustrated here with meat spoilage can be applied to other large-scale and heterogeneous datasets associated with pathway scenarios and represents a promising key tool for deciphering causality in complex biological phenomena.

Endometriosis is a gynaecological disease characterised by the presence of endometriotic tissue outside of the uterus impacting a significant fraction of women of childbearing age. Evidence from epidemiological studies suggests a relationship between risk of endometriosis and exposure to some organochlorine persistent organic pollutants (POPs). However, these chemicals are numerous and occur in complex and highly correlated mixtures, and to date, most studies have not accounted for this simultaneous exposure. Linear and logistic regression models are constrained to adjusting for multiple exposures when variables are highly intercorrelated, resulting in unstable coefficients and arbitrary findings. Advanced machine learning models, of emerging use in epidemiology, today appear as a promising option to address these limitations. In this study, different machine learning techniques were compared on a dataset from a case-control study conducted in France to explore associations between mixtures of POPs and deep endometriosis. The battery of models encompassed regularised logistic regression, artificial neural network, support vector machine, adaptive boosting, and partial least-squares discriminant analysis with some additional sparsity constraints. These techniques were applied to identify the biomarkers of internal exposure in adipose tissue most associated with endometriosis and to compare model classification performance. The five tested models revealed a consistent selection of most associated POPs with deep endometriosis, including octachlorodibenzofuran, cis-heptachlor epoxide, polychlorinated biphenyl 77 or trans-nonachlor, among others. The high classification performance of all five models confirmed that machine learning may be a promising complementary approach in modelling highly correlated exposure biomarkers and their associations with health outcomes. Regularised logistic regression provided a good compromise between the interpretability of traditional statistical approaches and the classification capacity of machine learning approaches. Applying a battery of complementary algorithms may be a strategic approach to decipher complex exposome-health associations when the underlying structure is unknown.

Salt reduction is becoming a major concern for public authorities, especially in cereal products. As childhood is important for the development of healthy eating habits, this study aimed to formulate salt‐reduced breads with satisfying sensory properties for children. Sourdough and an artisanal bread‐making process were used to compensate the flavor loss due to salt reduction. French breads (FBs) made with sourdough and artisanal processing were compared with white breads (WBs). Two salt levels were applied (1.2 and 1.8 g /100 g flour). To determine their acceptability and characterization, the four breads were assessed (i) by an adult panel (n= 39) according to cohesiveness, overall odor intensity, overall aroma in the mouth and saltiness intensity and (ii) a panel of children (n= 100, aged 6 to 11 years) according to overall liking and saltiness intensity. Finally, consumption by children (n= 89, aged 6 to 11 years) was measured during school lunch to evaluate the acceptability of salt reduction in a real consumption context. Both formulation and salt level induced physical and sensory changes in breads perceived by adults. They described WB as less dense, cohesive, and aromatic but more odorant than FB. Saltiness differences were perceived by adults but not by children. Children showed a preference for the saltiest breads and the FB but these drivers of preference were not confirmed during consumption measurements. These results shed new light on how natural solutions to enhance the flavor of bread can reduce its salt level while maintaining acceptability. Salt reduction in bread could be compensated by the use of sourdough and an artisanal bread‐making process. These methods allow an improvement of the nutritional quality of breads while maintaining their acceptance by young consumers by favoring the development of appealing organoleptic characteristics (aroma, texture). These methods are natural, easy to implement, and could be adapted to other fermented products in order to improve their nutritional quality.

International audience; ComDim (Common Dimensions) analysis was initially introduced within the context of sensometrics to analyze conventional and free choice sensory profiling data, and more generally multiblock datasets. Thereafter, it has gained some popularity in chemometrics and has been extended in different ways to meet specific needs. Recently, this strategy of analysis has been adapted to the supervised case, under the name of P-ComDim. Going further, we propose herein to extend ComDim to Path-ComDim where the datasets at hand are assumed to have a specific pattern of directed relations among them reflecting, for instance, a chain of influence. The aim of Path-ComDim is to analyze these datasets taking into account the structural connections among them. After a brief review of alternative path modeling approaches, Path-ComDim is detailed encompassing both methodological and algorithmic aspects. In the particular case of a single block to be predicted, it is shown that Path-ComDim is equivalent to P-ComDim analysis. Path-ComDim analysis is illustrated on the basis of a case study involving instrumental, sensory and preference data. Finally, the outcomes are compared to those obtained from alternative path modeling methods.

International audience; Abstract In a large variety of fields such as epidemiology, process monitoring, chemometrics, marketing, and social sciences among others, many research questions pertain to regression analysis from large data sets. Although in some cases standard regression will suffice, modeling is sometimes more challenging for various reasons (i) explain several variables; (ii) with a large number of explanatory variables organized into meaningful, usually ill?conditioned, multidimensional matrices; (iii) where observations come from different subpopulations; and (iv) with the opportunity to predict new observations. Although some developed methods partially meet these challenges, none of them covers all these aspects. To fill this gap, a new method, called regularized clusterwise multiblock regression (CW.rMBREG), is proposed. The method CW.rMBREG combines clustering and a component?based (multiblock) model associated with a well?defined criterion to optimize. It provides simultaneously a partition of the observations into clusters along with the regression coefficients associated with each cluster. To go further, we propose to investigate a key feature generally neglected in clusterwise regression, ie, the prediction of new observations. The usefulness of CW.rMBREG is illustrated on the basis of both a simulation study and a real example in the field of indoor air quality. It results that CW.rMBREG improves the quality of the prediction and facilitates the interpretation of complex ill?conditioned data. The proposed method is available for users through the R package mbclusterwise.

International audience; The aging process is associated with physiological, sensory, psychological, and sociological changes likely to have an impact on food intake and the nutritional status. The present study aimed to explore the heterogeneity of the French older population (>65 years old) using a multidisciplinary approach. More specifically, the study aimed to highlight different typologies (i.e. clusters of individuals with similar characteristics) within the older population. We conducted face-to-face interviews and tests with 559 French older people, recruited from different categories of dependency (at home without help, at home with help, in nursing homes). Clustering analysis highlighted seven clusters. Clusters 1-3 contained ‘young’ older people (

International audience; The STATIS method has been successfully applied to the analysis of sensory profiling data and other kinds data in sensometrics. We discuss its use and benefits and compare its outcomes to alternative methods for the analysis of multiblock data arising in situations such as projective mapping and free sorting experiments. More importantly, a method of clustering a collection of datasets measured on the same individuals, called CLUSTATIS, is introduced. It is based on the optimization of a criterion and consists in a hierarchical cluster analysis and a partitioning algorithm akin to the K-means algorithm. The procedure of analysis can be seen as an extension of the cluster analysis of variables around latent components (CLV, Vigneau & Qannari, 2003) to the case of blocks of variables. Alongside the determination of the clusters, a latent configuration is determined by the STATIS method. The interest of CLUSTATIS in sensometrics is discussed and illustrated on the basis of two case studies pertaining to the projective mapping also called Napping and the free sorting tasks, respectively.

ComDim analysis was designed to assess the relationships between individuals and variables within a multiblock setting where several variables, organized in blocks, are measured on the same individuals. An overview of this method is presented together with some of its properties. Furthermore, we discuss a new extension of the method of analysis to the case of (K+1) datasets. More precisely, the aim is to explore the relationships between a response dataset and K other datasets. An illustration of this latter strategy of analysis on the basis of a case study involving Time Domain - Nuclear Magnetic Resonance data is outlined and the outcomes are compared with those of Multiblock Partial Least Squares regression.

International audience; Aromatic characterization is a key element of enhancing one's knowledge of wine. While several studies have investigated the importance of wine expertise in the ability to perform odor-related sensory tasks, little attention has been paid to the influence of expertise on the semantic categorization of wine odors. To bridge this gap, this study aimed to explore the influence of a subject's expertise on the semantic representation of wine odors by means of a free sorting task. For this purpose, 156 subjects were recruited. Their level of expertise was measured using a questionnaire and the data analysis revealed four clusters of subjects with a gradual level of expertise. Subjects also performed a sorting task on 96 odor terms. From the number and the size of odor groups formed, as well as the additive tree representation and the consensus partition between the terms for each expertise level, we observed that all subjects, regardless of their experience, had largely the same semantic categorization of wine-odor attributes, which was mainly shaped by the sources of the odorants. It appeared that level of wine expertise played a minor role in creating the semantic representation of wine odors, affecting mainly the knowledge of specialized terms.

Data fusion refers to gathering different kinds of information together into a procedure yielding a single model. Among the different methods that exist to combine data from different sources, multiblock (or multitable) methods are a pertinent choice. These methods can be used to combine different data matrices obtained using different analytical techniques. Of the many multiblock methods available, this chapter presents the common components and specific weights analysis method, also called ComDim. After a thorough presentation of the basic ComDim algorithm, two recent extensions of ComDim are presented and compared with other existing methods using examples: Predictive ComDim is used when one of the data tables is dependent on all the others; Path modeling ComDim is used when there is a network of relationships among some of the data tables and where some tables may be both predictive and response blocks.

ISI Document Delivery No.: JQ2BB Times Cited: 0 Cited Reference Count: 37 Managmana, Essomanda Tchandao Cariou, Veronique Vigneau, Evelyne Kakai, Romain Lucas Glele Qannari, El Mostafa Glele Kakai, Romain/0000-0002-6965-4331 France overseas cultural and cooperation network of the Embassy of France in Togo; African Excellence Center in Mathematical Sciences and Applications of the University of Abomey-Calavi (Benin) The first author is very grateful to France overseas cultural and cooperation network of the Embassy of France in Togo and the African Excellence Center in Mathematical Sciences and Applications of the University of Abomey-Calavi (Benin) for their financial support. 0 3 Elsevier Amsterdam 1873-3239; International audience; For the analysis of multiblock data, a unified approach of several strategies such as Generalized Canonical Correlation Analysis (GCCA), Multiblock Principal Components Analysis (MB-PCA), Hierarchical Principal Components Analysis (H-PCA) and ComDim is outlined. These methods are based on the determination of global and block components. The unified approach postulates, on the one hand, two link functions that relate the block components to their associated global components and, on the other hand, two summing up expressions to compute the global components from their associated block components. Not only several well-known methods are retrieved but we also introduce a variant of GCCA. More generally, we hint to other possibilities of extensions thus emphasizing the fact that the unified approach, besides being simple, is versatile. We also show how this approach of analysis although basically unsupervised could be adapted to yield a supervised method to be used for a prediction purpose. Illustrations on the basis of simulated and real case studies are discussed.

International audience; Salt reduction is becoming a major concern for public authorities, especially in cereal products. As childhood is important for the development of healthy eating habits, this study aimed to formulate salt-reduced breads with satisfying sensory properties for children. Sourdough and an artisanal bread-making process were used to compensate the flavor loss due to salt reduction. French breads (FBs) made with sourdough and artisanal processing were compared with white breads (WBs). Two salt levels were applied (1.2 and 1.8 g /100 g flour). To determine their acceptability and characterization, the four breads were assessed (i) by an adult panel (n = 39) according to cohesiveness, overall odor intensity, overall aroma in the mouth and saltiness intensity and (ii) a panel of children (n = 100, aged 6 to 11 years) according to overall liking and saltiness intensity. Finally, consumption by children (n = 89, aged 6 to 11 years) was measured during school lunch to evaluate the acceptability of salt reduction in a real consumption context. Both formulation and salt level induced physical and sensory changes in breads perceived by adults. They described WB as less dense, cohesive, and aromatic but more odorant than FB. Saltiness differences were perceived by adults but not by children. Children showed a preference for the saltiest breads and the FB but these drivers of preference were not confirmed during consumption measurements. These results shed new light on how natural solutions to enhance the flavor of bread can reduce its salt level while maintaining acceptability. Practical Application Salt reduction in bread could be compensated by the use of sourdough and an artisanal bread-making process. These methods allow an improvement of the nutritional quality of breads while maintaining their acceptance by young consumers by favoring the development of appealing organoleptic characteristics (aroma, texture). These methods are natural, easy to implement, and could be adapted to other fermented products in order to improve their nutritional quality.

In consumer studies, segmentation has been widely applied to identify consumer subsets on the basis of their preference for a set of products. From the last decade onwards, a more comprehensive evaluation of product performance has led to take into account various information such as consumer emotion assessment or hedonic measures on several aspects, like taste, visual and flavor. This multi-attribute evaluation of products naturally yields a three-way (products by consumers by attributes) data structure. In order to identify segments of consumers on the basis of such three-way data, the Three-Way Cluster analysis around Latent Variables (CLV3W) approach (Wilderjans & Cariou, 2016) is considered. This method groups the consumers into clusters and estimates for each cluster an associated latent product variable and attribute weights, along with a set of consumer loadings, which may be used for the purpose of cluster-specific product characterization. As consumers who rate the products along the attributes in an opposite way (i.e., raters’ disagreement) should not be in the same cluster, in this paper, we propose to add a non-negativity constraint on the consumer loadings and to integrate this constraint within the versatile CLV3W approach. This non-negatively constrained criterion implies that the latent variable for each cluster is determined such that consumers within each cluster are as much related – in terms of a positive covariance – as possible with this latent product component. This approach is applied to a consumer emotion ratings dataset related to coffee aromas. ispartof: Food Quality and Preference vol:67 pages:18-26 ispartof: location:ENGLAND, Brighton status: published

International audience; Several statistical procedures have been proposed for the analysis of the data from a free sorting task. A straightforward strategy of analysis based on correspondence analysis and cluster analysis performed on the co-occurrence matrix is proposed herein. More specifically, two situations are considered depending on whether the aim is to depict the relationships among the stimuli or to investigate the agreement among the subjects. The approach of analysis is illustrated on the basis of free sorting data characterizing chocolate products.

Human milk is recommended for feeding preterm infant.Yet the potential impact of specific breast-milk lipid components on the initial growth rate of very-preterm infants has received scant attention. The current pilot study aims to determine whether breast-milk lipidome had any impact on the early growth pattern of preterm infants fed their own mother’s milk. A prospective monocentric observational birth cohort was established, enrolling 147 preterm infants, who received their own mother’s breast-milk throughout hospital stay. Among that cohort, infants who experienced slow (n=15) or fast (n=11) growth were selected, based on the change in their weight Z-score between birth and hospital discharge (-1.54± 0.42 and -0.48± 0.19 Z-score, respectively). Liquid chromatography-high resolution-mass spectrometry was used to obtain lipidomic signatures in breast-milk. Multivariate analyses made it possible to identify breast-milk lipid species that allowed clear-cut discrimination between the 2 infants’ groups. Validation of the selected biomarkers was performed by means of various multidimensional statistical techniques, false-discovery rate and ROC curve computation. Breast-milk associated with fast growth contained more medium chain-saturated fatty acid and -sphingomyelin, dihomo-γ-linolenic acid (DGLA)-containing phosphethanolamine, and less oleic acid-containing triglyceride and DGLA-oxylipin. Their predictive ability of preterm early-growth rate was validated in presence of confounding clinical factors.&nbsp

National audience

International audience; The classification of microorganisms by high-dimensional phenotyping methods such as FTIR spectroscopy is often a complicated process due to the complexity of microbial phylogenetic taxonomy. A hierarchical structure developed for such data can often facilitate the classification analysis. The hierarchical tree structure can either be imposed to a given set of phenotypic data by integrating the phylogenetic taxonomic structure or set up by revealing the inherent clusters in the phenotypic data. In this study, we wanted to compare different approaches to hierarchical classification of microorganisms based on high-dimensional phenotypic data. A set of 19 different species of molds (filamentous fungi) obtained from the mycological strain collection of the Norwegian Veterinary Institute (Oslo, Norway) is used for the study. Hierarchical cluster analysis is performed for setting up the classification trees. Classification algorithms such as artificial neural networks (ANN), partial least-squared discriminant analysis and random forest (RF) are used and compared. The 2 methods ANN and RF outperformed all the other approaches even though they did not utilize predefined hierarchical structure. To our knowledge, the RF approach is used here for the first time to classify microorganisms by FTIR spectroscopy.

Within the framework of nonlinear partial least squares (PLS), the quadratic PLS regression approach, involving both linear and quadratic terms in the criterion, is discussed. A new algorithm for the determination of the components is proposed, and its advantages over the original algorithm are outlined. The approach of analysis is illustrated on the basis of simulated and real data. Copyright © 2012 John Wiley & Sons, Ltd.

Within the framework of sensory conventional profiling, we propose a general strategy of weight assignment to the various statistical units (products × assessors). This means that for each product, an assessor who does not agree with the other assessors on the evaluation of the product under consideration will be down-weighted thus limiting their impact on the outcome of the analysis. The weights are derived from similarity measures between assessors which reflect the extent to which the assessors agree on the position of each product in the perceptual space. The weights are used to compute the usual statistical parameters such as means and variance–covariance matrices. They are also used to compute the PLS-discriminant components which reflect directions of discrimination among the products. The usefulness of the strategy of weight assignment is demonstrated on the basis of a case study where it is shown that it improves the stability of the representation of the products to small perturbations.

Document Type : Proceedings Paper Conference Date : JUL 25-28, 2010 Conference Location : Rotterdam, NETHERLANDS Conference Sponsor : Sensometr Soc; Several methods have been proposed in the literature to analyze conventional sensory profiling data. We focus on factor analytical methods which have been extensively used due to their ability to produce graphical displays which are both useful and easy to interpret. Available factor analytical methods include principal components analysis on averaged assessors data or on the data matrix obtained by stacking the assessors' datasets one on top of the others, and canonical variates analysis; each method has advantages and drawbacks. As an alternative it is advocated the use of PLS discriminant analysis, which is at the intersection of the methods mentioned above. This method of analysis takes account of the within and between product variations while minimizing the impact of multicolinearity. It provides statistical tools to assess on the one hand the agreement among assessors and the discrimination among products by means of the between to total variance ratio, and on the other hand the relative importance of variables by means of VIP (variable importance in the projection) indices. The VIP indices may also be useful to guide the selection of a subset of relevant attributes from the complete set of attributes. In this paper, PIS discriminant analysis is compared with other methods, and results are illustrated through a case study. In particular, the stability of the various methods is investigated using assessors' re-sampling (bootstrap) and confidence ellipses. (C) 2011 Elsevier Ltd. All rights reserved.

International audience; To detect panel disagreement, we propose the clustering around latent variables for three-way data (CLV3W) approach which extends the clustering of variables around latent components (CLV) approach to three-way data typically obtained from a conventional sensory profiling procedure (i.e., assessors rating products on various descriptors). The CLV3W method groups the descriptors into Q clusters and estimates for each cluster an associated latent sensory component such that the attributes within each cluster are as much related (i.e., highest squared covariance) as possible with the latent component. Simultaneously, for each latent sensory component separately, a system of weights is estimated that yields information regarding the extent to which an assessor (dis)agrees with the rest of the panel according to the latent sensory component under study. Our new approach is illustrated with a dataset pertaining to Quantitative Descriptive Analysis applied to cider varieties. It is shown that CLV3W, as opposed to related approaches, is able to detect differential panel disagreement on various latent sensory components.

Les viticulteurs disposent de plusieurs moyens de mesure de la qualite de leurs raisins lorsqu'ils sont arrives a maturite pour la vendange. La prediction de la qualite du raisin, vise a formaliser le savoir-faire des experts (viticulteurs et conseillers) pour developper un modele mathematique qui definira les relations entre la qualite du raisin et les facteurs (pratiques, sol, climat) ayant permis d'aboutir a cette qualite. La presente etude a pour objet de developper une premiere etude exploratoire permettant d'investiguer chaque parametre de donnees et chaque facteur precedent la construction effective d'un modele explicatif, voire predictif de la qualite du raisin obtenu a la vendange a partir de pratiques culturales et des donnees du milieu. Le jeu de donnees utilise pour construire le modele provient de parcelles produisant du raisin en AOP (Appellation d'Origine Protegee) a partir du cepage Chenin situe en moyenne Vallee de la Loire (France), destine a produire un vin blanc sec pendant lors des millesimes 2010 a 2013. Les pratiques, le milieu et la qualite du raisin ont ete analyses pour chacune de ces parcelles et constituent la base de donnees du modele. Ce dernier s'appuiera par la suite sur une analyse statistique nommee Regression PLS (Partial Least Square).

International audience; In sensory analysis, preference mapping covers several modeling techniques applied to hedonic data for a better understanding of consumers' liking and product optimization. External preference mapping aims at relating consumers' hedonic data to sensory data in order to identify liking drivers. Classically, preference mapping proceeds in two steps: the first step consists in defining the perceptual space on which preference data are regressed and, the second step, identifies the predictive model according to this perceptual space. The strategy of analysis (quadratic PLS regression) presented herein fits within the framework of PLS regression. An optimal perceptual space is sought by taking account of the linear and the quadratic relationships between hedonic and sensory data. Quadratic PLS is compared to other methods of analysis on the basis of a case study related to coffee data. (C) 2013 Elsevier Ltd. All rights reserved.

Le déclin de la perception sensorielle des personnes âgées cache une grande variabilité : est-il lié aux préférences alimentaires et au statut nutritionnel ?. Le Sensolier, 10. journée : Des différences de perception entre individus

Le processus de vieillissement, même lorsqu’il se déroule normalement, est associé à de nombreuses modifications d’ordre physiologiques (ralentissement du transit intestinal, du métabolisme basal, perte de dents, altérations des capacités chimiosensorielles), psychologiques (solitude, dépression, démence) et sociologiques (retraite, veuvage, déménagement…) susceptibles d’avoir un impact sur la prise alimentaire et le statut nutritionnel de la personne âgée. Au-delà de ces changements, l’apparition d’une dépendance pour l’alimentation, qui fait généralement suite à une perte d’autonomie de la personne (apparition d’une incapacité physique ou psychique, veuvage…), modifie en profondeur le rapport de la personne à son alimentation. Dés lors qu’une personne âgée doit déléguer tout ou partie de son alimentation à une aide ménagère ou à un service de restauration collective, elle doit s’adapter aux pratiques culinaires, aux habitudes alimentaires, voire aux préférences alimentaires de tierces personnes. Au regard de ces changements, l’âge et la dépendance s’accompagne fréquemment d’une diminution de l’appétit et d’une dégradation du statut nutritionnel. Selon les estimations de la HAS (2007), 4 à 10% des personnes âgées vivant à domicile seraient dénutries et cette proportion atteindrait entre 15 et 38% des personnes âgées vivant en EHPAD (Etablissement d’Hébergement pour Personnes Agées Dépendantes). A l’heure actuelle, la prévention et la lutte contre la dénutrition s’appuient essentiellement sur une prise en charge nutritionnelle – mise en place de régimes alimentaires adaptés, formulation d’aliments enrichis, prescription de compléments alimentaires. Or, comme l’exprime la grande majorité des personnes âgées, l’acte alimentaire – manger – est et reste avant tout un plaisir. « C’est même le dernier plaisir qu’il nous reste » pour reprendre l’expression d’une personne interrogée dans le cadre de nos travaux. Afin de mieux comprendre le lien entre plaisir sensoriel et état nutritionnel chez les personnes âgées, une enquête multidisciplinaire a été réalisée auprès de 559 personnes âgées de plus de 65 ans. Cette enquête comprenait près de 200 questions portant entre autre sur les habitudes et les préférences alimentaires des seniors, leur perception des odeurs et des saveurs des aliments ainsi que sur leur état de santé et leur état nutritionnel. Cette enquête a été menée dans quatre villes françaises (Angers, Brest, Nantes et Dijon), auprès de seniors vivant en EHPAD ou à domicile, avec ou sans aide extérieure. Les résultats ont mis en évidence une grande variabilité au sein de la population âgée, tant au niveau de la perception sensorielle que des préférences alimentaires. Alors que la communauté scientifique admet que les capacités chimiosensorielles diminuent avec l’âge, l’enquête a montré qu’environ 45% de la population âgée présentait des capacités sensorielles préservées. Le reste de la population se partage entre ceux qui perçoivent plutôt moins bien les odeurs et ceux qui perçoivent plutôt moins bien les saveurs. En parallèle, les résultats ont mis en évidence trois styles de mangeurs. Les amateurs de “mitonné” préfèrent la viande au poisson, le beurre à l’huile et ne peuvent pas se passer d’un dessert à la fin de leur repas. Les amateurs de “bonne chair” apprécient un bon rôti, saignant de préférence. Ces personnes apprécient un bon verre de vin pour accompagner leur repas mais n’apprécient pas particulièrement les produits sucrés. Enfin, les amateurs de “naturalité” préfèrent le poisson à la viande, l’huile au beurre. Ils apprécient les légumes, en particulier cuits à la vapeur, et rejettent les plats tout préparés. Si peu de liens ont été observés entre perception sensorielle et préférences alimentaires, les résultats ont en revanche montré que les personnes présentant une forte altération olfactive tendent à avoir un statut nutritionnel plus fragile. A l’inverse, les amateurs de “naturalité” tendent à avoir un meilleur statut nutritionnel. Fort de ces constats, l’enjeu des travaux à venir est de développer des aliments répondant à la fois aux attentes en matière d’habitudes alimentaires et de préférences des personnes âgées, tout en tenant compte de leurs capacités chimiosensorielles et de leurs besoins nutritionnels. In fine, l’objectif est d’augmenter le plaisir associé au repas pour cette population et de contribuer ainsi à la lutte contre la dénutrition. Remerciements. Les travaux susmentionnés s’inscrivent dans le cadre du programme AUPALESENS (Amélioration du plaisir alimentaire du senior pour bien vieillir et lutter contre la dénutrition) financé par l’Agence Nationale de la Recherche (ANR – 2010-2013). Ce programme est coordonné par Virginie Van Wymelbeke du CHU Dijon (Unité de recherche service de médecine gériatrique) et implique l’INRA (UMR Centre des Sciences du Goût et de l’Alimentation, Dijon et l’UMR ALISS, Ivry Sur Seine), l’Ecole Supérieure d’Agriculture (UPSP GRAPPE, LUNAM Université, Groupe ESA, Angers), le CHU Angers (Service d’endocrinologie et diabétologie), ONIRIS (Unité de Sensométrie et Chimiométrie, Nantes), l’Université François Rabelais (Laboratoire Psychologie des Ages de la Vie, Tours) ainsi que les partenaires du monde économique Lactalis (Retiers), Les Repas Santé (Beaune), Cecab d’Aucy (Saint Thurien), Livrac (Haute Goulaine), Frutarom (Dijon), Entremont Alliance (Malestroit).

National audience; Fish or rare steak? Fruit or pastry? Soup and ham or roast and dessert?... In order to explore the favourite meal of the elderly, a large survey was conducted including 559 people older than 65 (AUPALESENS survey). Different categories of elderly were interviewed, including people who are living at home either without any assistance or with assistance, and people who are living in a nursing home.Each respondent completed a questionnaire designed to explore three dimensions, namely meal content, eater style and pickiness. Regarding meal content, respondents were asked to choose among broad categories of dishes (e.g., salad, soup, roast, stew, veggies, cheese, fruit, pastry…) the ones they would like to have for a daily meal. Regarding eater style, respondents’ preference was explored through 21 items such as “I prefer eating meat rather than eating fish”; “I would rather use butter than oil”… Finally, pickiness was assessed through the number of disliked products among a list of 72 products including raw foods (beef, veal, zucchinis, rice…) and dishes (paella, gratin Dauphinois…). In parallel, respondents completed a questionnaire on their health and nutritional state.Results revealed three main styles of eaters. The “meat & pastries” eaters prefer meat to fish, butter to oil, and can not do without a dessert at the end of their meal. The “roast & wine” eaters are close to the previous ones except that they are not fond of sweet foods. Rather, they welcome a glass of wine with their meal. The “fish & veggies” eaters prefer fish to meat, oil to butter. They appreciate vegetables, especially when steamed, and they reject ready-to-eat dishes. Similarly, result on meal content revealed that respondents are shared between those who favoured a copious meal (delicatessen, meat stew and potatoes, cake) and those who favoured a light meal (salad, ham or poultry, vegetables, possibly a fruit), with intermediate stages between. Finally, results showed that “fish & veggie” eaters tend to have a better nutritional state that the others and that pickiness is inversely correlated with nutritional state. The number of picky respondents also increases with dependency.To conclude, the questionnaires developed in this study proved to be an interesting tool to gather information about what elderly people favour in their meal in order to propose meals adapted to their preferences with a positive impact on their nutritional state.

This chapter is in the scope of static and dynamic discovery-driven explorations of a data cube. It presents different methods to facilitate the whole process of data exploration. Each kind of analysis (static or dynamic) is developed for either a count measure or a quantitative measure. Both are based on the calculation, on the fly, of specific statistical built-in indicators. Firstly, a global methodology is proposed to help a dynamic discovery-driven exploration. It aims at identifying the most relevant dimensions to expand. A built-in rank on dimensions is restituted interactively, at each step of the process. Secondly, to help a static discovery-driven exploration, generalized statistical criteria are detailed to detect and highlight interesting cells within a cube slice. The cell’s degree of interest is determined by the calculation of either test-value or Chi-Square contribution. Their display is done by a color-coding system. A proof of concept implementation on the ORACLE 10g system is described at the end of the chapter.

Self-organizing maps (SOM) have been widely used in different data analysis fields for both their clustering and visualisation properties. However, dealing with spectral data, artificial neural networks (ANN) have generally been applied within a supervised context rather than unsupervised one. In this chapter, we present how the use of self-organizing maps may help end-users to visualise spectral data. While representing the Kohonen map, external characteristics associated with spectra can be projected on the map. Dealing with high-dimensional data, a dimension reduction is proposed to provide synthetic representation of the map to the most relevant variables.

International audience; A statistical approach for the analysis of free sorting data is discussed. In a first stage, the sorting data from each subject are arranged into a dataset consisting of indicator variables which reflect the memberships of the stimuli to the groups formed by the subject under consideration. Thereafter, an appropriate standardization is applied on these data and a three way statistical method, namely Common Components and Specific Weights Analysis, is performed on the datasets thus obtained. This makes it possible to take account of the individual differences among the subjects and to depict graphical displays showing the relationships among the stimuli on the one hand and among subjects on the other hand. The general strategy of analysis is illustrated on the basis of a dataset and the outcomes are compared to those of other strategies of analysis of sorting data.

In large companies, Online Analytical Processing (OLAP) technologies are widely used by business analysts as a decision-support tool. The exploration of the data is performed using operators such as drill-down, roll-up or slice. While exploring the cube, end-users are rapidly confronted with analysing a huge number of drill-paths according to the different dimensions. Generally, analysts are only interested in a small part of them, which corresponds to either high statistical associations between dimensions or atypical cell values. Moreover, identifying the most interesting cells is a matter for business analysts. Coupling OLAP technologies and mining methods may help them by the automation of this tedious task. This paper, in the scope of discovery-driven exploration, presents a method to facilitate the whole process of exploration of the data cube by identifying the most relevant dimensions to expand. A built-in rank on dimensions is displayed, at each step of the process, to the users, who are still free to choose the right dimension to expand for their analysis. Built-in rank on dimensions is performed through indicators computed on the fly according to the user-defined data selection. We present how this methodology offers a support to the decision-making, directly integrated to a commercial OLAP management system. A proof of concept implementation on the ORACLE 10g system is described at the end of the paper.