Integrated methodologies to study human transcriptional regulation from functional genomics data

Abstract : The human body is composed of a huge number of cells acting together in a concerted manner. The current understanding is that proteins perform most of the necessary activities in keeping a cell alive. The DNA, on the other hand, stores the information on how to produce the different proteins in the genome. Regulating gene transcription is the first important step that can thus affect the life of a cell, modify its functions and its responses to the environment. Regulation is a complex operation that involves specialized proteins, the transcription factors. Transcription factors (TFs) can bind to DNA and activate the processes leading to the expression of genes into new proteins. Errors in this process may lead to diseases. In particular, some transcription factors have been associated with a lethal pathological state, commonly known as cancer, associated with uncontrolled cellular proliferation, invasiveness of healthy tissues and abnormal responses to stimuli. Understanding cancer-related regulatory programs is a difficult task, often involving several TFs interacting together and influencing each other's activity. This Thesis presents new computational methodologies to study gene regulation. In addition we present applications of our methods to the understanding of cancer-related regulatory programs. The understanding of transcriptional regulation is a major challenge. We address this difficult question combining computational approaches with large collections of heterogeneous experimental data. In detail, we design signal processing tools to recover transcription factors binding sites on the DNA from genome-wide surveys like chromatin immunoprecipitation assays on tiling arrays (ChIP-chip). We then use the localization about the binding of TFs to explain expression levels of regulated genes. In this way we identify a regulatory synergy between two TFs, the oncogene C-MYC and SP1. C-MYC and SP1 bind preferentially at promoters and when SP1 binds next to C-NIYC on the DNA, the nearby gene is strongly expressed. The association between the two TFs at promoters is reflected by the binding sites conservation across mammals, by the permissive underlying chromatin states 'it represents an important control mechanism involved in cellular proliferation, thereby involved in cancer. Secondly, we identify the characteristics of TF estrogen receptor alpha (hERa) target genes and we study the influence of hERa in regulating transcription. hERa, upon hormone estrogen signaling, binds to DNA to regulate transcription of its targets in concert with its co-factors. To overcome the scarce experimental data about the binding sites of other TFs that may interact with hERa, we conduct in silico analysis of the sequences underlying the ChIP sites using the collection of position weight matrices (PWMs) of hERa partners, TFs FOXA1 and SP1. We combine ChIP-chip and ChIP-paired-end-diTags (ChIP-pet) data about hERa binding on DNA with the sequence information to explain gene expression levels in a large collection of cancer tissue samples and also on studies about the response of cells to estrogen. We confirm that hERa binding sites are distributed anywhere on the genome. However, we distinguish between binding sites near promoters and binding sites along the transcripts. The first group shows weak binding of hERa and high occurrence of SP1 motifs, in particular near estrogen responsive genes. The second group shows strong binding of hERa and significant correlation between the number of binding sites along a gene and the strength of gene induction in presence of estrogen. Some binding sites of the second group also show presence of FOXA1, but the role of this TF still needs to be investigated. Different mechanisms have been proposed to explain hERa-mediated induction of gene expression. Our work supports the model of hERa activating gene expression from distal binding sites by interacting with promoter bound TFs, like SP1. hERa has been associated with survival rates of breast cancer patients, though explanatory models are still incomplete: this result is important to better understand how hERa can control gene expression. Thirdly, we address the difficult question of regulatory network inference. We tackle this problem analyzing time-series of biological measurements such as quantification of mRNA levels or protein concentrations. Our approach uses the well-established penalized linear regression models where we impose sparseness on the connectivity of the regulatory network. We extend this method enforcing the coherence of the regulatory dependencies: a TF must coherently behave as an activator, or a repressor on all its targets. This requirement is implemented as constraints on the signs of the regressed coefficients in the penalized linear regression model. Our approach is better at reconstructing meaningful biological networks than previous methods based on penalized regression. The method is tested on the DREAM2 challenge of reconstructing a five-genes/TFs regulatory network obtaining the best performance in the "undirected signed excitatory" category. Thus, these bioinformatics methods, which are reliable, interpretable and fast enough to cover large biological dataset, have enabled us to better understand gene regulation in humans.

Preoperative percutaneous portal vein embolization: evaluation of adverse events in 188 patients.

PURPOSE: To retrospectively assess the frequency of adverse events related to percutaneous preoperative portal vein embolization (PPVE). MATERIALS AND METHODS: Institutional review board did not require its approval or patient informed consent for this study. The adverse events that occurred during PPVE or until planned hepatic surgery was performed or cancelled were retrospectively obtained from clinical, imaging, and laboratory data files in 188 patients (109 male and 79 female patients; mean age, 60 years; range, 16-78 years). Liver resection was planned for metastases (n = 137), hepatocarcinoma (n = 31), cholangiocarcinoma (n = 15), fibrolamellar hepatoma (n = 1), and benign disease (n = 4). PPVE was performed with a single-lumen 5-F catheter and a contralateral approach with n-butyl cyanoacrylate mixed with iodized oil as the main embolic agent. The rate of complications in patients with cirrhosis was compared with that in patients without cirrhosis by using the chi(2) test. RESULTS: Adverse events occurred in 24 (12.8%) of 188 patients, including 12 complications and 12 incidental imaging findings. Complications included thrombosis of the portal vein feeding the future remnant liver (n = 1); migration of emboli in the portal vein feeding the future remnant liver, which necessitated angioplasty (n = 2); hemoperitoneum (n = 1); rupture of a metastasis in the gallbladder (n = 1); transitory hemobilia (n = 1); and transient liver failure (n = 6). Incidental findings were migration of small emboli in nontargeted portal branches (n = 10) and subcapsular hematoma (n = 2). Among the 187 patients in whom PPVE was technically successful, there was a significant difference (P < .001) between the occurrence of liver failure after PPVE in patients with cirrhosis (five of 30) and those without (one of 157). Sixteen liver resections were cancelled due to cancer progression (n = 12), insufficient hypertrophy of the nonembolized liver (n = 3), and complete portal thrombosis (n = 1). CONCLUSION: PPVE is a safe adjuvant technique for hypertrophy of the initially insufficient liver reserve. Post-PPVE transient liver failure is more common in patients with cirrhosis than in those without cirrhosis.

Firm's network capability and innovation performance: evidences from China hi-tech industry

ABSTRACT : A firm's competitive advantage can arise from internal resources as well as from an interfirm network. -This dissertation investigates the competitive advantage of a firm involved in an innovation network by integrating strategic management theory and social network theory. It develops theory and provides empirical evidence that illustrates how a networked firm enables the network value and appropriates this value in an optimal way according to its strategic purpose. The four inter-related essays in this dissertation provide a framework that sheds light on the extraction of value from an innovation network by managing and designing the network in a proactive manner. The first essay reviews research in social network theory and knowledge transfer management, and identifies the crucial factors of innovation network configuration for a firm's learning performance or innovation output. The findings suggest that network structure, network relationship, and network position all impact on a firm's performance. Although the previous literature indicates that there are disagreements about the impact of dense or spare structure, as well as strong or weak ties, case evidence from Chinese software companies reveals that dense and strong connections with partners are positively associated with firms' performance. The second essay is a theoretical essay that illustrates the limitations of social network theory for explaining the source of network value and offers a new theoretical model that applies resource-based view to network environments. It suggests that network configurations, such as network structure, network relationship and network position, can be considered important network resources. In addition, this essay introduces the concept of network capability, and suggests that four types of network capabilities play an important role in unlocking the potential value of network resources and determining the distribution of network rents between partners. This essay also highlights the contingent effects of network capability on a firm's innovation output, and explains how the different impacts of network capability depend on a firm's strategic choices. This new theoretical model has been pre-tested with a case study of China software industry, which enhances the internal validity of this theory. The third essay addresses the questions of what impact network capability has on firm innovation performance and what are the antecedent factors of network capability. This essay employs a structural equation modelling methodology that uses a sample of 211 Chinese Hi-tech firms. It develops a measurement of network capability and reveals that networked firms deal with cooperation between, and coordination with partners on different levels according to their levels of network capability. The empirical results also suggests that IT maturity, the openness of culture, management system involved, and experience with network activities are antecedents of network capabilities. Furthermore, the two-group analysis of the role of international partner(s) shows that when there is a culture and norm gap between foreign partners, a firm must mobilize more resources and effort to improve its performance with respect to its innovation network. The fourth essay addresses the way in which network capabilities influence firm innovation performance. By using hierarchical multiple regression with data from Chinese Hi-tech firms, the findings suggest that there is a significant partial mediating effect of knowledge transfer on the relationships between network capabilities and innovation performance. The findings also reveal that the impacts of network capabilities divert with the environment and strategic decision the firm has made: exploration or exploitation. Network constructing capability provides a greater positive impact on and yields more contributions to innovation performance than does network operating capability in an exploration network. Network operating capability is more important than network constructing capability for innovative firms in an exploitation network. Therefore, these findings highlight that the firm can shape the innovation network proactively for better benefits, but when it does so, it should adjust its focus and change its efforts in accordance with its innovation purposes or strategic orientation.

THE INFLUENCE OF THE REGULATORY ENVIRONMENT ON THE DEFINITION OF PERFORMANCE: THE EXAMPLE OF THE SPORT BETTING AND LOTTERY SECTORS

This thesis examines the interplay between state regulation and the way organisations define performance. Performance is generally understood to be a multidimensional concept, but the extent to which its different facets are shaped by regulation remains an understudied question. This thesis aims to address this question and provide at least a partial answer to it. To do so, it examines whether the level of regulation amplifies or abates the multidimensionality of regulated entities' performance definition, i.e. the way they define the concept of performance. The leading question is whether an organisation's performance definition can be associated with the regulatory intensity its environment confronts it with. Moreover, the study explores whether the type of ownership-public or private-plays a role in regard to how a regulated entity defines performance. In order to undertake this investigation, the thesis focuses on the performance definitions of organisations in six different sport betting and lottery regulations. Qualitative data is gathered from primary and secondary documents as well as through semi-structured interviews with chief executive officers (CEO), members of executive management and gambling experts in each of these countries. The thesis concludes that the performance definitions of the organisations under study are indeed multidimensional, as well as clearly influenced by their respective regulatory environments. However, not all performance dimensions identified in the literature are present, nor can they all be estimated to be part of the performance definition. In addition, the public-private difference in defining performance-as conceptualised in the literature- seems to be abated in a regulated environment. The central role played by regulation in regard to the multidimensionality of the performance definition partially outweighs the effect of the nature of ownership.

Factors favoring a degradation or an improvement in activities of daily living (ADL) performance among nursing home (NH) residents: A survival analysis.

Different factors influence ADL performance among nursing home (NH) residents in long term care. The aim was to investigate which factors were associated with a significant change of ADL performance in NH residents, and whether or not these factors were gender-specific. The design was a survival analysis. The 10,199 participants resided in ninety Swiss NHs. Their ADL performance had been assessed by the Resident Assessment Instrument Minimum Data Set (RAI-MDS) in the period from 1997 to 2007. Relevant change in ADL performance was defined as 2 levels of change on the ADL scale between two successive assessments. The occurrence of either an improvement or a degradation of the ADL status) was analyzed using the Cox proportional hazard model. The analysis included a total of 10,199 NH residents. Each resident received between 2 and 23 assessments. Poor balance, incontinence, impaired cognition, a low BMI, impaired vision, no daily contact with proxies, impaired hearing and the presence of depression were, by hierarchical order, significant risk factors for NH residents to experience a degradation of ADL performance. Residents, who were incontinent, cognitively impaired or had a high BMI were significantly less likely to improve their ADL abilities. Male residents with cancer were prone to see their ADL improve. The year of NH entry was significantly associated with either degradation or improvement of ADL performance. Measures aiming at improving balance and continence, promoting physical activity, providing appropriate nourishment and cognitive enhancement are important for ADL performance in NH residents.

Correlation between microtensile bond strength data and clinical outcome of Class V restorations.

OBJECTIVE: To determine if the results of resin-dentin microtensile bond strength (µTBS) is correlated with the outcome parameters of clinical studies on non-retentive Class V restorations. METHODS: Resin-dentin µTBS data were obtained from one test center; the in vitro tests were all performed by the same operator. The µTBS testing was performed 8h after bonding and after 6 months of storing the specimens in water. Pre-test failures (PTFs) of specimens were included in the analysis, attributing them a value of 1MPa. Prospective clinical studies on cervical restorations (Class V) with an observation period of at least 18 months were searched in the literature. The clinical outcome variables were retention loss, marginal discoloration and marginal integrity. Furthermore, an index was formulated to be better able to compare the laboratory and clinical results. Estimates of adhesive effects in a linear mixed model were used to summarize the clinical performance of each adhesive between 12 and 36 months. Spearman correlations between these clinical performances and the µTBS values were calculated subsequently. RESULTS: Thirty-six clinical studies with 15 adhesive/restorative systems for which µTBS data were also available were included in the statistical analysis. In general 3-step and 2-step etch-and-rinse systems showed higher bond strength values than the 2-step/3-step self-etching systems, which, however, produced higher values than the 1-step self-etching and the resin modified glass ionomer systems. Prolonged water storage of specimens resulted in a significant decrease of the mean bond strength values in 5 adhesive systems (Wilcoxon, p<0.05). There was a significant correlation between µTBS values both after 8h and 6 months of storage and marginal discoloration (r=0.54 and r=0.67, respectively). However, the same correlation was not found between µTBS values and the retention rate, clinical index or marginal integrity. SIGNIFICANCE: As µTBS data of adhesive systems, especially after water storage for 6 months, showed a good correlation with marginal discoloration in short-term clinical Class V restorations, longitudinal clinical trials should explore whether early marginal staining is predictive for future retention loss in non-carious cervical restorations.

Performance of a chromatographic procedure used in the certification of reference material for polycyclic aromatic hydrocarbons in sewage sludge

Semi-automatic capillary gas chromatographic method with classical flame ionization detection, which satisfies the conditions for required performance and gave acceptable results within the framework of an interlaboratory certification programme for PAHs in sewage sludge, is described. The interesting feature of the procedure is that it incorporates automatic operations such as sample fractionation by semi-preparative HPLC, fraction collection at signal level recognition and evaporation under nitrogen flow. Multiple injections in the GC capillary column are performed in the on-column mode via an autosampler with temperature-programmable injector. Automatic data acquisition and chromatogram treatment are made via computer software. This partially automatic procedure releases personnel from tedious and time-consuming tasks and its robust character was validated through the certification of reference material for PAHs in sewage sludge, demonstrating its reliable performance.

Ensemble methods for environmental data modelling with support vector regression

This paper investigates the use of ensemble of predictors in order to improve the performance of spatial prediction methods. Support vector regression (SVR), a popular method from the field of statistical machine learning, is used. Several instances of SVR are combined using different data sampling schemes (bagging and boosting). Bagging shows good performance, and proves to be more computationally efficient than training a single SVR model while reducing error. Boosting, however, does not improve results on this specific problem.

A Note on Interviewer Performance Measures in centralised CATI Surveys

Interviewer performance with respect to convincing sample members to participate in surveys is an important dimension of survey quality. However, unlike in CAPI surveys where each sample case 'belongs' to one interviewer, there are hardly any good measures of interview performance for centralised CATI surveys, where even single contacts are assigned to interviewers at random. If more than one interviewer works one sample case, it is not clear how to attribute success or failure to the interviewers involved. In this article, we propose two correlated methods to measure interviewer contact performance in centralised CATI surveys. Their modelling must take complex multilevel clustering effects, which need not be hierarchical, into account. Results are consistent with findings from CAPI data modelling, and we find that when comparing effects with a direct ('naive') measure of interviewer contact results, interviewer random effects are largely underestimated using the naive measure.

The effect of mountain bike suspensions on vibrations and off-road uphill performance.

AIM: This study evaluates the effect of front suspension (FS) and dual suspension (DS) mountain-bike on performance and vibrations during off-road uphill riding. METHODS: Thirteen male cyclists (27+/-5 years, 70+/-6 kg, VO(2max)59+/-6 mL.kg(-1).min(-1), mean+/-SD) performed, in a random sequence, at their lactate threshold, an off-road uphill course (1.69 km, 212 m elevation gain) with both type of bicycles. Variable measured: a) VO(2) consumption (K4b2 analyzer, Cosmed), b) power output (SRM) c) gain in altitude and d) 3-D accelerations under the saddle and at the wheel (Physilog, EPFL, Switzerland). Power spectral analy- sis (Fourier) was performed from the vertical acceleration data. RESULTS: Respectively for the FS and DS mountain bike: speed amounted to 7.5+/-0.7 km.h(-1) and 7.4+/-0.8 km.h(-1), (NS), energy expenditure 1.39+/-0.16 kW and 1.38+/-0.18, (NS), gross efficiency 0.161+/-0.013 and 0.159+/-0.013, (NS), peak frequency of vibration under the saddle 4.78+/-2.85 Hz and 2.27+/-0.2 Hz (P<0.01) and median-frequency of vertical displacements of the saddle 9.41+/-1.47 Hz and 5.78+/-2.27 Hz (P<0.01). CONCLUSION: Vibrations at the saddle level of the DS bike are of low frequencies whereas those of the FS bike are mostly of high frequencies. In the DS bike, the torque produced by the cyclist at the pedal level may generate low frequency vibrations. We conclude that the DS bike absorbs more high frequency vibrations, is more comfortable and performs as well as the FS bicycle.

Use of the Cognitive Performance Scale (CPS) to detect cognitive impairment in the acute care setting: concurrent and predictive validity.

The Cognitive Performance Scale (CPS) was initially designed to assess cognition in long term care residents. Subsequently, the CPS has also been used among in-home, post-acute, and acute care populations even though CPS' clinimetric performance has not been studied in these settings. This study aimed to determine CPS agreement with the Mini Mental Status Exam (MMSE) and its predictive validity for institutionalization and death in a cohort (N=401) of elderly medical inpatients aged 75 years and over. Medical, physical and mental status were assessed upon admission. The same day, the patient's nurse completed the CPS by interview. Follow-up data were gathered from the central billing system (nursing home stay) and proxies (death). Cognitive impairment was present in 92 (23%) patients according to CPS (score >or= 2). Agreement with MMSE was moderate (kappa 0.52, P<.001). Analysis of discordant results suggested that cognitive impairment was overestimated by the CPS in dependent patients with comorbidities and depressive symptoms, and underestimated in older ones. During follow-up, subjects with abnormal CPS had increased risks of death (adjusted hazard ratio (adjHR) 1.7, 95% CI 1.0-2.8, P=.035) and institutionalization (adjHR 2.7, 95% CI 1.3-5.3, P=.006), independent of demographic, health and functional status. Interestingly, subjects with abnormal CPS were at increased risk of death only if they also had abnormal MMSE. The CPS predicted death and institutionalization during follow-up, but correlated moderately well with the MMSE. Combining CPS and MMSE provided additional predictive information, suggesting that domains other than cognition are assessed by professionals when using the CPS in elderly medical inpatients.

Machine learning for geospatial data : algorithms, software tools and case studies

Résumé Cette thèse est consacrée à l'analyse, la modélisation et la visualisation de données environnementales à référence spatiale à l'aide d'algorithmes d'apprentissage automatique (Machine Learning). L'apprentissage automatique peut être considéré au sens large comme une sous-catégorie de l'intelligence artificielle qui concerne particulièrement le développement de techniques et d'algorithmes permettant à une machine d'apprendre à partir de données. Dans cette thèse, les algorithmes d'apprentissage automatique sont adaptés pour être appliqués à des données environnementales et à la prédiction spatiale. Pourquoi l'apprentissage automatique ? Parce que la majorité des algorithmes d'apprentissage automatiques sont universels, adaptatifs, non-linéaires, robustes et efficaces pour la modélisation. Ils peuvent résoudre des problèmes de classification, de régression et de modélisation de densité de probabilités dans des espaces à haute dimension, composés de variables informatives spatialisées (« géo-features ») en plus des coordonnées géographiques. De plus, ils sont idéaux pour être implémentés en tant qu'outils d'aide à la décision pour des questions environnementales allant de la reconnaissance de pattern à la modélisation et la prédiction en passant par la cartographie automatique. Leur efficacité est comparable au modèles géostatistiques dans l'espace des coordonnées géographiques, mais ils sont indispensables pour des données à hautes dimensions incluant des géo-features. Les algorithmes d'apprentissage automatique les plus importants et les plus populaires sont présentés théoriquement et implémentés sous forme de logiciels pour les sciences environnementales. Les principaux algorithmes décrits sont le Perceptron multicouches (MultiLayer Perceptron, MLP) - l'algorithme le plus connu dans l'intelligence artificielle, le réseau de neurones de régression généralisée (General Regression Neural Networks, GRNN), le réseau de neurones probabiliste (Probabilistic Neural Networks, PNN), les cartes auto-organisées (SelfOrganized Maps, SOM), les modèles à mixture Gaussiennes (Gaussian Mixture Models, GMM), les réseaux à fonctions de base radiales (Radial Basis Functions Networks, RBF) et les réseaux à mixture de densité (Mixture Density Networks, MDN). Cette gamme d'algorithmes permet de couvrir des tâches variées telle que la classification, la régression ou l'estimation de densité de probabilité. L'analyse exploratoire des données (Exploratory Data Analysis, EDA) est le premier pas de toute analyse de données. Dans cette thèse les concepts d'analyse exploratoire de données spatiales (Exploratory Spatial Data Analysis, ESDA) sont traités selon l'approche traditionnelle de la géostatistique avec la variographie expérimentale et selon les principes de l'apprentissage automatique. La variographie expérimentale, qui étudie les relations entre pairs de points, est un outil de base pour l'analyse géostatistique de corrélations spatiales anisotropiques qui permet de détecter la présence de patterns spatiaux descriptible par une statistique. L'approche de l'apprentissage automatique pour l'ESDA est présentée à travers l'application de la méthode des k plus proches voisins qui est très simple et possède d'excellentes qualités d'interprétation et de visualisation. Une part importante de la thèse traite de sujets d'actualité comme la cartographie automatique de données spatiales. Le réseau de neurones de régression généralisée est proposé pour résoudre cette tâche efficacement. Les performances du GRNN sont démontrées par des données de Comparaison d'Interpolation Spatiale (SIC) de 2004 pour lesquelles le GRNN bat significativement toutes les autres méthodes, particulièrement lors de situations d'urgence. La thèse est composée de quatre chapitres : théorie, applications, outils logiciels et des exemples guidés. Une partie importante du travail consiste en une collection de logiciels : Machine Learning Office. Cette collection de logiciels a été développée durant les 15 dernières années et a été utilisée pour l'enseignement de nombreux cours, dont des workshops internationaux en Chine, France, Italie, Irlande et Suisse ainsi que dans des projets de recherche fondamentaux et appliqués. Les cas d'études considérés couvrent un vaste spectre de problèmes géoenvironnementaux réels à basse et haute dimensionnalité, tels que la pollution de l'air, du sol et de l'eau par des produits radioactifs et des métaux lourds, la classification de types de sols et d'unités hydrogéologiques, la cartographie des incertitudes pour l'aide à la décision et l'estimation de risques naturels (glissements de terrain, avalanches). Des outils complémentaires pour l'analyse exploratoire des données et la visualisation ont également été développés en prenant soin de créer une interface conviviale et facile à l'utilisation. Machine Learning for geospatial data: algorithms, software tools and case studies Abstract The thesis is devoted to the analysis, modeling and visualisation of spatial environmental data using machine learning algorithms. In a broad sense machine learning can be considered as a subfield of artificial intelligence. It mainly concerns with the development of techniques and algorithms that allow computers to learn from data. In this thesis machine learning algorithms are adapted to learn from spatial environmental data and to make spatial predictions. Why machine learning? In few words most of machine learning algorithms are universal, adaptive, nonlinear, robust and efficient modeling tools. They can find solutions for the classification, regression, and probability density modeling problems in high-dimensional geo-feature spaces, composed of geographical space and additional relevant spatially referenced features. They are well-suited to be implemented as predictive engines in decision support systems, for the purposes of environmental data mining including pattern recognition, modeling and predictions as well as automatic data mapping. They have competitive efficiency to the geostatistical models in low dimensional geographical spaces but are indispensable in high-dimensional geo-feature spaces. The most important and popular machine learning algorithms and models interesting for geo- and environmental sciences are presented in details: from theoretical description of the concepts to the software implementation. The main algorithms and models considered are the following: multi-layer perceptron (a workhorse of machine learning), general regression neural networks, probabilistic neural networks, self-organising (Kohonen) maps, Gaussian mixture models, radial basis functions networks, mixture density networks. This set of models covers machine learning tasks such as classification, regression, and density estimation. Exploratory data analysis (EDA) is initial and very important part of data analysis. In this thesis the concepts of exploratory spatial data analysis (ESDA) is considered using both traditional geostatistical approach such as_experimental variography and machine learning. Experimental variography is a basic tool for geostatistical analysis of anisotropic spatial correlations which helps to understand the presence of spatial patterns, at least described by two-point statistics. A machine learning approach for ESDA is presented by applying the k-nearest neighbors (k-NN) method which is simple and has very good interpretation and visualization properties. Important part of the thesis deals with a hot topic of nowadays, namely, an automatic mapping of geospatial data. General regression neural networks (GRNN) is proposed as efficient model to solve this task. Performance of the GRNN model is demonstrated on Spatial Interpolation Comparison (SIC) 2004 data where GRNN model significantly outperformed all other approaches, especially in case of emergency conditions. The thesis consists of four chapters and has the following structure: theory, applications, software tools, and how-to-do-it examples. An important part of the work is a collection of software tools - Machine Learning Office. Machine Learning Office tools were developed during last 15 years and was used both for many teaching courses, including international workshops in China, France, Italy, Ireland, Switzerland and for realizing fundamental and applied research projects. Case studies considered cover wide spectrum of the real-life low and high-dimensional geo- and environmental problems, such as air, soil and water pollution by radionuclides and heavy metals, soil types and hydro-geological units classification, decision-oriented mapping with uncertainties, natural hazards (landslides, avalanches) assessments and susceptibility mapping. Complementary tools useful for the exploratory data analysis and visualisation were developed as well. The software is user friendly and easy to use.

Effects of sample size on the performance of species distribution models

A wide range of modelling algorithms is used by ecologists, conservation practitioners, and others to predict species ranges from point locality data. Unfortunately, the amount of data available is limited for many taxa and regions, making it essential to quantify the sensitivity of these algorithms to sample size. This is the first study to address this need by rigorously evaluating a broad suite of algorithms with independent presence-absence data from multiple species and regions. We evaluated predictions from 12 algorithms for 46 species (from six different regions of the world) at three sample sizes (100, 30, and 10 records). We used data from natural history collections to run the models, and evaluated the quality of model predictions with area under the receiver operating characteristic curve (AUC). With decreasing sample size, model accuracy decreased and variability increased across species and between models. Novel modelling methods that incorporate both interactions between predictor variables and complex response shapes (i.e. GBM, MARS-INT, BRUTO) performed better than most methods at large sample sizes but not at the smallest sample sizes. Other algorithms were much less sensitive to sample size, including an algorithm based on maximum entropy (MAXENT) that had among the best predictive power across all sample sizes. Relative to other algorithms, a distance metric algorithm (DOMAIN) and a genetic algorithm (OM-GARP) had intermediate performance at the largest sample size and among the best performance at the lowest sample size. No algorithm predicted consistently well with small sample size (n < 30) and this should encourage highly conservative use of predictions based on small sample size and restrict their use to exploratory modelling.

What matters for predicting spatial distributions of trees: techniques, data, or species' characteristics?

Data characteristics and species traits are expected to influence the accuracy with which species' distributions can be modeled and predicted. We compare 10 modeling techniques in terms of predictive power and sensitivity to location error, change in map resolution, and sample size, and assess whether some species traits can explain variation in model performance. We focused on 30 native tree species in Switzerland and used presence-only data to model current distribution, which we evaluated against independent presence-absence data. While there are important differences between the predictive performance of modeling methods, the variance in model performance is greater among species than among techniques. Within the range of data perturbations in this study, some extrinsic parameters of data affect model performance more than others: location error and sample size reduced performance of many techniques, whereas grain had little effect on most techniques. No technique can rescue species that are difficult to predict. The predictive power of species-distribution models can partly be predicted from a series of species characteristics and traits based on growth rate, elevational distribution range, and maximum elevation. Slow-growing species or species with narrow and specialized niches tend to be better modeled. The Swiss presence-only tree data produce models that are reliable enough to be useful in planning and management applications.

Superior diagnostic performance of perfusion-cardiovascular magnetic resonance versus SPECT to detect coronary artery disease: The secondary endpoints of the multicenter multivendor MR-IMPACT II (Magnetic Resonance Imaging for Myocardial Perfusion Assessment in Coronary Artery Disease Trial).

ABSTRACT: BACKGROUND: Perfusion-cardiovascular magnetic resonance (CMR) is generally accepted as an alternative to SPECT to assess myocardial ischemia non-invasively. However its performance vs gated-SPECT and in sub-populations is not fully established. The goal was to compare in a multicenter setting the diagnostic performance of perfusion-CMR and gated-SPECT for the detection of CAD in various populations using conventional x-ray coronary angiography (CXA) as the standard of reference. METHODS: In 33 centers (in US and Europe) 533 patients, eligible for CXA or SPECT, were enrolled in this multivendor trial. SPECT and CXA were performed within 4 weeks before or after CMR in all patients. Prevalence of CAD in the sample was 49% and 515 patients received MR contrast medium. Drop-out rates for CMR and SPECT were 5.6% and 3.7%, respectively (ns). The study was powered for the primary endpoint of non-inferiority of CMR vs SPECT for both, sensitivity and specificity for the detection of CAD (using a single-threshold reading), the results for the primary endpoint were reported elsewhere. In this article secondary endpoints are presented, i.e. the diagnostic performance of CMR versus SPECT in subpopulations such as multi-vessel disease (MVD), in men, in women, and in patients without prior myocardial infarction (MI). For diagnostic performance assessment the area under the receiver-operator-characteristics-curve (AUC) was calculated. Readers were blinded versus clinical data, CXA, and imaging results. RESULTS: The diagnostic performance (= area under ROC = AUC) of CMR was superior to SPECT (p = 0.0004, n = 425) and to gated-SPECT (p = 0.018, n = 253). CMR performed better than SPECT in MVD (p = 0.003 vs all SPECT, p = 0.04 vs gated-SPECT), in men (p = 0.004, n = 313) and in women (p = 0.03, n = 112) as well as in the non-infarct patients (p = 0.005, n = 186 in 1-3 vessel disease and p = 0.015, n = 140 in MVD). CONCLUSION: In this large multicenter, multivendor study the diagnostic performance of perfusion-CMR to detect CAD was superior to perfusion SPECT in the entire population and in sub-groups. Perfusion-CMR can be recommended as an alternative for SPECT imaging. TRIAL REGISTRATION: ClinicalTrials.gov, Identifier: NCT00977093.