Robert Sabatier

The ACT (STATIS method)

Abstract ACT (STATIS method) is a data analysis technique which computes Euclidean distances between configurations of the same observations obtained in K different circumstances, and thus handles three-way data as a set of K matrices. In this article, the recent developments of the ACT technique are fully described - concepts and theorems related to Euclidean scaling being discussed in the Appendix - and the software manipulation is illustrated on real data.

Neuroanatomical localisation and clinical correlates of white matter lesions in the elderly

Background: White matter lesions (WML) in elderly people co-occur with hypertension, depression, and cognitive impairment. Little is known about the density and distribution of WML in normal elderly people, whether they occur randomly in the aging brain or tend to cluster in certain areas, or whether patterns of WML aggregation are linked to clinical symptoms. Objectives: To describe patterns of WML distribution in a large representative population of elderly people using non-inferential cluster analysis; and to determine the extent to which such patterns are associated with clinical symptomatology. Method: A population sample of 1077 elderly people was recruited. Multiple analysis of correspondence followed by automatic classification methods was used to explore overall patterns of WML distribution. Correspondence was then sought between these patterns and a range of cerebrovascular, psychiatric, and neurological symptoms. Results: Three distinct patterns of spatial localisation within the brain were observed, corresponding to distinct clusters of clinical symptoms. In particular WML aggregation in temporal and occipital areas was associated with greater age, hypertension, late onset depressive disorder, poor global cognitive function, and overall WML frequency. Conclusions: WML localisation is not random in the aging brain, and their distribution is associated with age and the presence of clinical symptoms. Age differences suggest there may be patterns of progression across time; however, this requires confirmation from longitudinal imaging studies.

Refined approximations to permutation tests for multivariate inference

Various authors have proposed approximations to permutation tests of independence between two data tables. We develop approximations based on explicit expressions of the first three moments of three different test statistics under the permutation distribution. The rejection level is then determined by using a Pearson-type III distribution matching the values of the first three moments. We present three examples in which the relative merits of the test statistics are examined and the results of the approximation procedure are compared with explicit permutation tests.

Regional tests for trend detection in maximum precipitation series in the French Mediterranean region

Abstract Recent major floods in the French Mediterranean region prompted the investigation into whether they are the consequence of climatic change. Changes in monthly and annual maximum series from 92 precipitation gauges in the French Mediterranean region were examined. Each station provided a minimum of 56 years of daily measurements. Trends were searched for using the Mann-Kendall non-parametric test and the maximum-likelihood parametric method. The main aim was to build two tests to estimate the regional significance of the local trends detected to allow estimation of whether local changes are due to chance or to a real climatic change. In Languedoc Roussillon, it was observed that monthly maxima have been decreasing in March and increasing in April. This can be explained by a transfer of rainy days from March to April. Significant increases were also detected in annual maxima and in monthly maxima in October, in the southern part of the Massif Central.

Bounded optimal knots for regression splines

Using a B-spline representation for splines with knots seen as free variables, the approximation to data by splines improves greatly. The main limitations are the presence of too many local optima in the univariate regression context, and it becomes even worse in multivariate additive modeling. When the number of knots is a priori fixed, we present a simple algorithm to select their location subject to box constraints for computing least-squares spline approximations. Despite its simplicity, or perhaps because of it, the method is comparable with other more sophisticated techniques and is very attractive for a small number of variables, as shown in the examples. In a complete algorithm, the BIC and AIC criteria are evaluated for choosing the number of knots as well as the degree of the splines.

Two Approaches for Discriminant Partial Least Squares

In the medical sciences as well as in other contexts we often have to deal with the study of groups and with the research of their separation. The aim of this paper is to highlight how, in some situations, Partial Least Squares (PLS) Discriminant Analysis (Sjostrom et al., 1986) can lead to a solution that is not an answer to the given problem of discrimination. Within a PLS framework, the authors provide two extensions of it. The first is close to the Generalized PLS proposed by (1997) but used in the discrimination context. The second proposal, in the same framework, leads to consider the PLS Redundancy Analysis proposed by (1995) by using suitable metrics. Some examples of data treatment are given.

Spatial Distribution of Cerebral White Matter Lesions Predicts Progression to Mild Cognitive Impairment and Dementia

Context White matter lesions (WML) increase the risk of dementia. The relevance of WML location is less clear. We sought to determine whether a particular WML profile, based on the density and location of lesions, could be associated with an increased risk of mild cognitive impairment (MCI) or dementia over the following 7 years. Methods In 426 healthy subjects from a cohort of community-dwelling people aged 65 years and over (ESPRIT Project), standardized cognitive and neurological evaluations were repeated after 2, 4 and 7 years. Patterns of WML were computed with a supervised data mining approach (decision trees) using the regional WML volumes (frontal, parietal, temporal, and occipital regions) and the total WML volume estimated at baseline. Cox proportional hazard models were then constructed to study the association between WML patterns and risk of MCI/dementia. Results Total WML volume and percentage of WML in the temporal region proved to be the best predictors of progression to MCI and dementia. Specifically, severe total WML load with a high proportion of lesions in the temporal region was significantly associated with the risk of developing MCI or dementia. Conclusions Above a certain threshold of damage, a pattern of WML clustering in the temporal region identifies individuals at increased risk of MCI or dementia. As this WML pattern is observed before the onset of clinical symptoms, it may facilitate the detection of patients at risk of MCI/dementia.

A generalization of STATIS-ACT strategy: DO-ACT for two multiblocks tables

A new strategy is introduced for analyzing two multiblocks tables: DO-ACT. This method is closely related to the STATIS (or ACT) methodology and the Tucker inter-battery method. The length of two multiblocks are not necessarily the same and the optimal solution obtained is that of a global optimization problem. The advantage of using DO-ACT is that the first step provides a summary of the two multiblocks tables, in the second step two optimal representations (one for each multiblock) of the observations can be plotted and in the third step a global description of each table of each multiblock can be made. An example of DO-ACT performance is illustrated with a real data set. The program implementing the method has been developed using the S-Plus 6.0® (2000) language.

Effect of exposure to polycyclic aromatic hydrocarbons on basal ganglia and attention-deficit hyperactivity disorder symptoms in primary school children

BACKGROUND Polycyclic aromatic hydrocarbons (PAHs) have been proposed as environmental risk factors for attention deficit hyperactivity disorder (ADHD). The effects of these pollutants on brain structures potentially involved in the pathophysiology of ADHD are unknown. OBJECTIVE The aim of this study was to investigate the effects of PAHs on basal ganglia volumes and ADHD symptoms in school children. METHODS We conducted an imaging study in 242 children aged 8-12years, recruited through a set of representative schools of the city of Barcelona, Spain. Indoor and outdoor PAHs and benzo[a]pyrene (BPA) levels were assessed in the school environment, one year before the MRI assessment. Whole-brain volumes and basal ganglia volumes (caudate nucleus, globus pallidus, putamen) were derived from structural MRI scans using automated tissue segmentation. ADHD symptoms (ADHD/DSM-IV Scales, American Psychiatric Association 2002) were reported by teachers, and inattentiveness was evaluated with standard error of hit reaction time in the attention network computer-based test. RESULTS Total PAHs and BPA were associated with caudate nucleus volume (CNV) (i.e., an interquartile range increase in BPA outdoor level (67pg/m3) and indoor level (76pg/m3) was significantly linked to a decrease in CNV (mm3) (β=-150.6, 95% CI [-259.1, -42.1], p=0.007, and β=-122.4, 95% CI [-232.9, -11.8], p=0.030 respectively) independently of intracranial volume, age, sex, maternal education and socioeconomic vulnerability index at home). ADHD symptoms and inattentiveness increased in children with higher exposure to BPA, but these associations were not statistically significant. CONCLUSIONS Exposure to PAHs, and in particular to BPA, is associated with subclinical changes on the caudate nucleus, even below the legislated annual target levels established in the European Union. The behavioral consequences of this induced brain change were not identified in this study, but given the caudate nucleus involvement in many crucial cognitive and behavior processes, this volume reduction is concerning for the childrens neurodevelopment.

A new genetic algorithm in proteomics: Feature selection for SELDI-TOF data

Mass spectrometry from clinical specimens is used in order to identify biomarkers in a diagnosis. Thus, a reliable method for both feature selection and classification is required. A novel method is proposed to find biomarkers in SELDI-TOF in order to perform robust classification.The feature selection is based on a new genetic algorithm. Concerning the classification, a method which takes into account the great variability on intensity by using decision stumps has been developed. Moreover, as the samples are often small, it is more appropriate to use the decision stumps simultaneously than building a complete tree. The thresholds of the decision stumps are determined in the same genetic algorithm. Finally, the method was generalized to more than two groups based on pairwise coupling. The obtained algorithm was applied on two data sets: a publicly available one containing two groups allowing a comparison with other methods from the literature and a new one containing three groups.