Die chemische Zusammensetzung pharmazeutisch verwendeter Gläser (Ampullen, Infusionsglas, Fertigspritzenbehältnisse) wurde mit Laser-Ablation-ICP-Massenspektrometrie (LA-ICP-MS) untersucht. Dazu kam ein Nd:YAG-Laser mit einer Frequenz von 10 Hz im q-switch mode mit der Grundwellenlänge sowie der 4. Harmonischen (1064 und 266 nm) zum Einsatz. Ziel war die Identifizierung verschiedener Arten pharmazeutisch eingesetzter Gläser. Folgende Isotope wurden zur Charakterisierung bestimmt: 7Li, 11B, 23Na, 24Mg, 27Al, 28Si, 29Si, 30Si, 39K, 42Ca, 47Ti, 57Fe, 90Zr, 121Sb, 137Ba. Die quantitativen Ergebnisse zeigten relative Standardabweichungen von 1.8 % bis 8.0 %. Die Normierung der Daten erfolgte unter Bezug auf 29Si. Mit Referenzmaterial verschiedener Zusammensetzung (Natron-Kalk-Glas, Borosilikatglas und Bleiglas) wurden externe Mehrpunktkalibrationen erstellt. Für alle gemessenen Isotope konnten lineare Kalibrierfunktionen festgestellt werden. Die Richtigkeit des Verfahrens wurde durch Aufschluß des Standardmaterials wie auch der untersuchten Proben mit Flußsäure/Salpetersäure und Messung der Aufschlußlösungen nach externer Kalibration durch Multielement-Standards gezeigt. Aufschluß- und Laserablation-Ergebnisse zeigten gute Übereinstimmung und wichen um bis zu 8 % voneinander ab. Zur Prüfung repräsentativer Ablation der Proben wurden die Isotopenverhältnisse bestimmt. Auch hier zeigte sich eine ausreichende Übereinstimmung mit den theoretischen Werten. Weiterhin wurden Präzision, Empfindlichkeit, Selektivität, Nachweisgrenze, Bestimmungsgrenze und Robustheit des Verfahrens bestimmt. Im Unterschied zu den Ergebnissen unter der Grundwellenlänge führte die Verwendung der Laserwellenlänge 266 nm zu keinen signifikanten Verbesserungen der Resultate. Die Aufnahme transienter Signale sollte zur Erkennung von Elementverteilungen in Glasoberflächen dienen. Am Beispiel eines Natron-Kalk-Glases wurden in einzelnen Schichten verschiedene Elementkonzentrationen festgestellt. Weiterhin konnten mit dieser Technik Glasampullen erkannt werden, deren Oberflächen durch ein Silikonisierungsverfahren vergütet waren. Zum Abschluß der Untersuchungen wurde versucht, auf Grundlage der unnormierten Intensitätsdaten der LA-Messungen Gläser zu klassifizieren. Hierzu wurde die lineare Diskriminanzanalyse eingesetzt, mit deren Hilfe nach Variablenreduktion (8 von 13 Isotopen) alle eingesetzten Glasarten korrekt identifiziert werden konnten. Durch das entwickelte und validierte LA-ICP-MS-Verfahren steht eine leistungsfähige Technik zur quantitativen Untersuchung der chemischen Zusammensetzung von Gläsern sowie zur Identifizierung der jeweiligen Glassorte zur Verfügung. / The chemical composition of pharmaceutical glasses (ampoules, infusion bottles, plunger) has been determined by laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS). A Nd:YAG laser with 10 Hz repetition rate in the q-switch mode at its fundamental wavelength and its 4th harmonic (1064 and 266 nm) was used to identify common types of pharmaceutical glasses. The following isotopes were used for measurements: 7Li, 11B, 23Na, 24Mg, 27Al, 28Si, 29Si, 30Si, 39K, 42Ca, 47Ti, 57Fe, 90Zr, 121Sb, 137Ba. Relative standard deviations between 1.8 % and 8.0 % of the quantitative results were obtained. 29Si was used as internal standard. Standard reference materials (soda-lime-, lead- and borosilicate glasses) were used for external calibration of laser sampling. Linear calibration functions for each isotope were found. The accuracy was determined by digestion of all samples and standard materials in a two-step-procedure by nitric/fluoric acid, measurement and external calibration by ICP-MS with multi-elemental standard solutions. Digestion and laser ablation results agreed within 8 % with the certified values. The proof of representative ablation was given by sufficient agreement of intensity ratios of most isotopes with the corresponding theoretical values. Further precision, sensititivity, selectivity, limit of detection, limit of quantitation and robustness were determined. In contrast to the fundamental wavelength the laserablation technique with 266 nm showed no significant improvement in the quality of the results. By measuring transient signals element concentrations of surface layers should be detected. Differing concentrations could be determined in soda-lime-glass samples. Also different surface-treated borosilicate-glass ampoules were examined by this method. Silicon-treated glass surfaces could be identified. Finally raw data of LA measurements were used for applying linear discriminant analysis. After reduction of the used variables (8 of 13 isotopes) all types of glasses could be distinguished only by their intensity data of LA measurements. The developed LA-ICP-MS method is a powerful technique to distinguish different types of pharmaceutical glasses and to examine their chemical composition.

In this thesis, improving the performance of the automatic recognition of the Doppler radar targets is studied. The radar used in this study is a ground-surveillance doppler radar. Target types are car, truck, bus, tank, helicopter, moving man and running man. The input of this thesis is the output of the real doppler radar signals which are normalized and preprocessed (TRP vectors: Target Recognition Pattern vectors) in the doctorate thesis by Erdogan (2002). TRP vectors are normalized and homogenized doppler radar target signals with respect to target speed, target aspect angle and target range. Some target classes have repetitions in time in their TRPs. By the use of these repetitions, improvement of the target type classification performance is studied. K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) algorithms are used for doppler radar target classification and the results are evaluated. Before classification PCA (Principal Component Analysis), LDA (Linear Discriminant Analysis), NMF (Nonnegative Matrix Factorization) and ICA (Independent Component Analysis) are implemented and applied to normalized doppler radar signals for feature extraction and dimension reduction in an efficient way. These techniques transform the input vectors, which are the normalized doppler radar signals, to another space. The effects of the implementation of these feature extraction algoritms and the use of the repetitions in doppler radar target signals on the doppler radar target classification performance are studied.

In the last decade, there has been a growing interest in the use of differential sensing for molecular recognition. Inspired by the mammalian olfactory system, differential sensing employs an array of non-selective receptors, which through cross-reactive interactions, create a distinct pattern for each analyte tested. The unique fingerprints obtained for each analyte with differential sensing are studied with statistical analysis techniques, such as principal component analysis and linear discriminant analysis. It was postulated that serum albumin proteins would be applicable to differential sensing schemes due to significant differences in sequence identity between different serum albumin species, and due to the wide range of hydrophobic molecules which are known to bind to these proteins. Consequently, cross-reactive serum albumin arrays were developed, utilizing hydrophobic fluorescent indicators to detect hydrophobic molecules. As such, serum albumin cross-reactive arrays were employed to discriminate subtly different hydrophobic analytes, and mixtures of these analytes, in the form of terpenes and perfumes, plasticizers and plastic explosive mixtures, and glycerides and adipocyte extracts. In this doctoral work, a detailed review of the field of differential sensing, and a thorough study of principal component analysis and linear discriminant analysis in various differential sensing scenarios, are given. These introductory chapters aid in better understanding the methods and techniques applied in later experimental chapters. In chapter 3, serum albumins, a PRODAN indicator, and an additive are shown to discriminate five terpene analytes and terpene doped perfumes. Chapter 4 describes an array with serum albumins, two dansyl fluorophores, and an additive which successfully differentiate the plasticizers found within the plastic explosives C4 and Semtex and simulated C4 and Semtex mixtures. Discrimination of these simulated mixtures was also achieved with this array in the presence of soil contaminants, demonstrating the potential real-world applicability of this sensing ensemble. Finally, chapter 5 details an array consisting of serum albumins, several fluorescent indicators, and a Grubb's olefin metathesis reaction, to differentiate saturated and unsaturated triglycerides, diglycerides, and monoglycerides. Mixtures of glycerides in adipocyte extracts taken from rats with different health states were then successfully discriminated, showing promise for clinical applications in differentiating adipoctyes from pre-diabetic, type 2 diabetic, and non-diabetic individuals. / text

When we develop and use species distribution models to predict species' current or potential distributions, we are faced with the trade-offs between model generality, precision, and realism. It is important to know how to improve and validate model generality while maintaining good model precision and realism. However, it is difficult for ecologists to evaluate species distribution models using field-sampled data alone because the true species response function to environmental or ecological factors is unknown. Species distribution models should be able to approximate the true characteristics and distributions of species if ecologists want to use them as reliable tools. Simulated data provide the advantage of being able to know the true species-environment relationships and control the causal factors of interest to obtain insights into the effects of these factors on model performance. I used a case study on Bythotrephes longimanus distributions from several hundred Ontario lakes and a simulation study to explore the effects on model performance caused by several factors: the choice of predictor variables, the model evaluation methods, the quantity and quality of the data used for developing models, and the strengths and weaknesses of different species distribution models. Linear discriminant analysis, multiple logistic regression, random forests, and artificial neural networks were compared in both studies. Results based on field data sampled from lakes indicated that the predictive performance of the four models was more variable when developed on abiotic (physical and chemical) conditions alone, whereas the generality of these models improved when including biotic (relevant species) information. When using simulated data, although the overall performance of random forests and artificial neural networks was better than linear discriminant analysis and multiple logistic regression, linear discriminant analysis and multiple logistic regression had relatively good and stable model sensitivity at different sample size and detection limit levels, which may be useful for predicting species presences when data are limited. Random forests performed consistently well at different sample size levels, but was more sensitive to high detection limit. The performance of artificial neural networks was affected by both sample size and detection limit, and it was more sensitive to small sample size.

When we develop and use species distribution models to predict species' current or potential distributions, we are faced with the trade-offs between model generality, precision, and realism. It is important to know how to improve and validate model generality while maintaining good model precision and realism. However, it is difficult for ecologists to evaluate species distribution models using field-sampled data alone because the true species response function to environmental or ecological factors is unknown. Species distribution models should be able to approximate the true characteristics and distributions of species if ecologists want to use them as reliable tools. Simulated data provide the advantage of being able to know the true species-environment relationships and control the causal factors of interest to obtain insights into the effects of these factors on model performance. I used a case study on Bythotrephes longimanus distributions from several hundred Ontario lakes and a simulation study to explore the effects on model performance caused by several factors: the choice of predictor variables, the model evaluation methods, the quantity and quality of the data used for developing models, and the strengths and weaknesses of different species distribution models. Linear discriminant analysis, multiple logistic regression, random forests, and artificial neural networks were compared in both studies. Results based on field data sampled from lakes indicated that the predictive performance of the four models was more variable when developed on abiotic (physical and chemical) conditions alone, whereas the generality of these models improved when including biotic (relevant species) information. When using simulated data, although the overall performance of random forests and artificial neural networks was better than linear discriminant analysis and multiple logistic regression, linear discriminant analysis and multiple logistic regression had relatively good and stable model sensitivity at different sample size and detection limit levels, which may be useful for predicting species presences when data are limited. Random forests performed consistently well at different sample size levels, but was more sensitive to high detection limit. The performance of artificial neural networks was affected by both sample size and detection limit, and it was more sensitive to small sample size.

This thesis deals with the development of estimation algorithms with embedded feature selection the context of high dimensional data, in the supervised and unsupervised frameworks. The contributions of this work are materialized by two algorithms, GLOSS for the supervised domain and Mix-GLOSS for unsupervised counterpart. Both algorithms are based on the resolution of optimal scoring regression regularized with a quadratic formulation of the group-Lasso penalty which encourages the removal of uninformative features. The theoretical foundations that prove that a group-Lasso penalized optimal scoring regression can be used to solve a linear discriminant analysis bave been firstly developed in this work. The theory that adapts this technique to the unsupervised domain by means of the EM algorithm is not new, but it has never been clearly exposed for a sparsity-inducing penalty. This thesis solidly demonstrates that the utilization of group-Lasso penalized optimal scoring regression inside an EM algorithm is possible. Our algorithms have been tested with real and artificial high dimensional databases with impressive resuits from the point of view of the parsimony without compromising prediction performances.

Náplní této práce je teoretický úvod a následné praktické zpracování tématu Molekulární signatura jako optimální multi-objektivní funkce s aplikací v predikci v onkogenomice. Úvodní kapitoly jsou zaměřeny na téma rakovina, zejména pak rakovina prsu a její podtyp triple negativní rakovinu prsu. Následuje literární přehled z oblasti optimalizačních metod, zejména se zaměřením na metaheuristické metody a problematiku strojového učení. Část se odkazuje na onkogenomiku a principy microarray a také na statistiku a s důrazem na výpočet p-hodnoty a bimodálního indexu. Praktická část je pak zaměřena na konkrétní průběh výzkumu a nalezené závěry, vedoucí k dalším krokům výzkumu. Implementace vybraných metod byla provedena v programech Matlab a R, s využitím dalších programovacích jazyků a to konkrétně programů Java a Python.

This thesis deals with the increasingly developing field of biometric systems which is the identification of faces. The thesis deals with the possibilities of face localization in pictures and their normalization, which is necessary due to external influences and the influence of different scanning techniques. It describes various techniques of localization of dominant features of the face such as eyes, mouth or nose. Not least, it describes different approaches to the identification of faces. Furthermore a it deals with an implementation of the Dominant Face Features Recognition application, which demonstrates chosen methods for localization of the dominant features (Hough Transform for Circles, localization of mouth using the location of the eyes) and for identification of a face (Linear Discriminant Analysis, Kernel Discriminant Analysis). The last part of the thesis contains a summary of achieved results and a discussion.

Tato práce se zabývá pravděpodobnostními modely pro automatické rozpoznávání řečníka. Podrobně analyzuje zejména pravděpodobnostní lineární diskriminační analýzu (PLDA), která modeluje nízkodimenzionální reprezentace promluv ve formě \acronym{i--vektorů}.  Práce navrhuje dvě rozšíření v současnosti požívaného PLDA modelu. Nově navržený PLDA model s plným posteriorním rozložením  modeluje neurčitost při generování i--vektorů. Práce také navrhuje nový diskriminativní přístup k trénování systému pro verifikaci řečníka, který je založený na PLDA. Pokud srovnáváme původní PLDA s modelem rozšířeným o modelování  neurčitosti i--vektorů, výsledky dosažené s rozšířeným modelem dosahují až 20% relativního zlepšení při testech s krátkými nahrávkami. Pro delší  testovací segmenty  (více než jedna minuta) je zisk v přesnosti  menší, nicméně přesnost nového modelu není nikdy menší než přesnost výchozího systému.  Trénovací data jsou ale obvykle dostupná ve formě dostatečně dlouhých segmentů, proto v těchto případech použití nového modelu neposkytuje žádné výhody při trénování. Při trénování může být použit původní PLDA model a jeho rozšířená verze může být využita pro získání skóre v  případě, kdy se bude provádět testování na krátkých segmentech řeči. Diskriminativní model je založen na klasifikaci dvojic i--vektorů do dvou tříd představujících oprávněný a neoprávněný soud (target a non-target trial). Funkcionální forma pro získání skóre pro každý pár je odvozena z PLDA a trénování je založeno na logistické regresi, která minimalizuje vzájemnou entropii mezi správným označením všech soudů a pravděpodobnostním označením soudů, které navrhuje systém. Výsledky dosažené s diskriminativně trénovaným klasifikátorem jsou podobné výsledkům generativního PLDA, ale diskriminativní systém prokazuje schopnost produkovat lépe kalibrované skóre. Tato schopnost vede k lepší skutečné přesnosti na neviděné evaluační sadě, což je důležitá vlastnost pro reálné použití.

L’objectif de ce travail est de mettre au point une nouvelle approche automatique pour identifier les réseaux de gènes concourant à une même fonction biologique. Ceci permet une meilleure compréhension des phénomènes biologiques et notamment des processus impliqués dans les maladies telles que les cancers. Différentes stratégies ont été développées pour essayer de regrouper les gènes d’un organisme selon leurs relations fonctionnelles : génétique classique et génétique moléculaire. Ici, nous utilisons une propriété connue des réseaux de gènes fonctionnellement liés à savoir que ces gènes sont généralement co-régulés et donc co-exprimés. Cette co-régulation peut être mise en évidence par des méta-analyses de données de puces à ADN (micro-arrays) telles que Gemma ou COXPRESdb. Dans un travail précédent [Al Adhami et al., 2015], la topologie d’un réseau de co-expression de gènes a été caractérisé en utilisant deux paramètres de description des réseaux qui discriminent des groupes de gènes sélectionnés aléatoirement (modules aléatoires, RM) de groupes de gènes avec des liens fonctionnels connus (modules fonctionnels, FM), c’est-à-dire des gènes appartenant au même processus biologique GO. Dans le présent travail, nous avons cherché à généraliser cette approche et à proposer une méthode, appelée TopoFunc, pour améliorer l’annotation existante de la fonction génique. Nous avons d’abord testé différents descripteurs topologiques du réseau de co-expression pour sélectionner ceux qui identifient le mieux des modules fonctionnels. Puis, nous avons constitué une base de données rassemblant des modules fonctionnels et aléatoires, pour lesquels, sur la base des descripteurs sélectionnés, nous avons construit un modèle de discrimination LDA [Friedman et al., 2001] permettant, pour un sous-ensemble de gènes donné, de prédire son type (fonctionnel ou non). Basée sur la méthode de similarité de gènes travaillée par Wang et ses collègues [Wang et al., 2007], nous avons calculé un score de similarité fonctionnelle entre les gènes d’un module. Nous avons combiné ce score avec celui du modèle LDA dans une fonction de fitness implémenté dans un algorithme génétique (GA). À partir du processus biologique d’ontologie de gènes donné (GO-BP), AG visait à éliminer les gènes faiblement co-exprimés avec la plus grande clique de GO-BP et à ajouter des gènes «améliorant» la topologie et la fonctionnalité du module. Nous avons testé TopoFunc sur 193 GO-BP murins comprenant 50-100 gènes et avons montré que TopoFunc avait agrégé un certain nombre de nouveaux gènes avec le GO-BP initial tout en améliorant la topologie des modules et la similarité fonctionnelle. Ces études peuvent être menées sur plusieurs espèces (homme, souris, rat, et possiblement poulet et poisson zèbre) afin d’identifier des modules fonctionnels conservés au cours de l’évolution. / The aim of this work is to develop a new automatic approach to identify networks of genes involved in the same biological function. This allows a better understanding of the biological phenomena and in particular of the processes involved in diseases such as cancers. Various strategies have been developed to try to cluster genes of an organism according to their functional relationships : classical genetics and molecular genetics. Here we use a well-known property of functionally related genes mainly that these genes are generally co-regulated and therefore co-expressed. This co-regulation can be detected by microarray meta-analyzes databases such as Gemma or COXPRESdb. In a previous work [Al Adhami et al., 2015], the topology of a gene coexpression network was characterized using two description parameters of networks that discriminate randomly selected groups of genes (random modules, RM) from groups of genes with known functional relationship (functional modules, FM), e.g. genes that belong to the same GO Biological Process. We first tested different topological descriptors of the co-expression network to select those that best identify functional modules. Then, we built a database of functional and random modules for which, based on the selected descriptors, we constructed a discrimination model (LDA)[Friedman et al., 2001] allowing, for a given subset of genes, predict its type (functional or not). Based on the similarity method of genes worked by Wang and co-workers [Wang et al., 2007], we calculated a functional similarity score between the genes of a module. We combined this score with that of the LDA model in a fitness function implemented in a genetic algorithm (GA). Starting from a given Gene Ontology Biological Process (GO-BP), AG aimed to eliminate genes that were weakly coexpressed with the largest clique of the GO-BP and to add genes that "improved" the topology and functionality of the module. We tested TopoFunc on the 193 murine GO-BPs comprising 50-100 genes and showed that TopoFunc aggregated a number of novel genes to the initial GO-BP while improving module topology and functional similarity. These studies can be conducted on several species (humans, mice, rats, and possibly chicken and zebrafish) to identify functional modules preserved during evolution.