Similarity search is a task fundamental to many machine learning and data analytics applications, where distance metric learning plays an important role. However, since modern online applications continuously produce objects with new characteristics which tend to change over time, state-of-the-art similarity search using distance metric learning methods tends to fail when deployed in such applications without taking the change into consideration. In this work, we propose a Distance Metric Learning-based Continuous Similarity Search approach (CSS for short) to account for the dynamic nature of such data. CSS system adopts an online metric learning model to achieve distance metric evolving to adapt the dynamic nature of continuous data without large latency. To improve the accuracy of online metric learning model, a compact labeled dataset which is representative of the updated data is dynamically updated. Also, to accelerate similarity search, CSS includes an online maintained Locality Sensitive Hashing index to accelerate the similarity search. One, our labeled data update strategy progressively enriches the labeled data to assure continued representativeness, yet without excessively growing its size to ensure that the computation costs of metric learning remain bounded. Two, our continuous distance metric learning strategy ensures that each update only requires one linear time k-NN search in contrast to the cubic time complexity of relearning the distance metric from scratch. Three, our LSH update mechanism leverages our theoretical insight that the LSH built based on the original distance metric is equally effective in supporting similarity search using the new distance metric as long as the transform matrix learned for the new distance metric is reversible. This important observation empowers CSS to avoid the modification of LSH in most cases. Our experimental study using real-world public datasets and large synthetic datasets confirms the effectiveness of CSS in improving the accuracy of classification and information retrieval tasks. Also, CSS achieves 3 orders of magnitude speedup of our incremental distance metric learning strategy (and its three underlying components) over the state-of-art methods.

This thesis is concerned with improving the effectiveness of nearest neighbour search. Nearest neighbour search is the problem of finding the most similar data-points to a query in a database, and is a fundamental operation that has found wide applicability in many fields. In this thesis the focus is placed on hashing-based approximate nearest neighbour search methods that generate similar binary hashcodes for similar data-points. These hashcodes can be used as the indices into the buckets of hashtables for fast search. This work explores how the quality of search can be improved by learning task specific binary hashcodes. The generation of a binary hashcode comprises two main steps carried out sequentially: projection of the image feature vector onto the normal vectors of a set of hyperplanes partitioning the input feature space followed by a quantisation operation that uses a single threshold to binarise the resulting projections to obtain the hashcodes. The degree to which these operations preserve the relative distances between the datapoints in the input feature space has a direct influence on the effectiveness of using the resulting hashcodes for nearest neighbour search. In this thesis I argue that the retrieval effectiveness of existing hashing-based nearest neighbour search methods can be increased by learning the thresholds and hyperplanes based on the distribution of the input data. The first contribution is a model for learning multiple quantisation thresholds. I demonstrate that the best threshold positioning is projection specific and introduce a novel clustering algorithm for threshold optimisation. The second contribution extends this algorithm by learning the optimal allocation of quantisation thresholds per hyperplane. In doing so I argue that some hyperplanes are naturally more effective than others at capturing the distribution of the data and should therefore attract a greater allocation of quantisation thresholds. The third contribution focuses on the complementary problem of learning the hashing hyperplanes. I introduce a multi-step iterative model that, in the first step, regularises the hashcodes over a data-point adjacency graph, which encourages similar data-points to be assigned similar hashcodes. In the second step, binary classifiers are learnt to separate opposing bits with maximum margin. This algorithm is extended to learn hyperplanes that can generate similar hashcodes for similar data-points in two different feature spaces (e.g. text and images). Individually the performance of these algorithms is often superior to competitive baselines. I unify my contributions by demonstrating that learning hyperplanes and thresholds as part of the same model can yield an additive increase in retrieval effectiveness.

DNA damage from both normal metabolic activities and environmental factors such as UV and radiation can cause as many as 1 million individual lesions to the DNA per cell per day (Lodish et al 2004). Cells respond to this continuous damage by employing many, highly efficient DNA repair mechanisms and undergo apoptosis when normal DNA repair fails. Of the many types of DNA damage that can occur, double strand breaks (DSBs) are the most toxic (Featherstone & Jackson 1999). A single unrepaired DSB is enough to induce cellular apoptosis and several mechanisms have developed to repair DSBs. The recognition, signalling and repair of DSBs involve large multi-­‐subunit complexes that bind to both the DNA and modified histone tails, which require modification of the chromatin in order to access their bind sites and function effectively (Allard et al 2004). Consequently several chromatin-­‐remodelling proteins have been implicated in DSB repair (van Attikum et al 2004, Chai et al 2005). LSH (Lymphoid specific helicase) is a putative chromatin-­‐remodelling enzyme that interacts with DNA methyltransferases and has been connected to DNA methylation (Myant & Stancheva, 2008). Knockouts of LSH or its homologues in A. thaliana and M. musculus show a reduction in DNA methylation of 60-­‐70% (Jeddeloh et al 1999, Dennis et al 2001). However in addition to this phenotype, knockout A. thaliana also have an increased sensitivity to DNA damage (Shaked et al 2006). A homologue of LSH has also been identified in S. cerevisiae, which interacts with known repair proteins (Collins et al 2007) and may be involved in DSB repair. Although the majority of Lsh-­‐/-­‐ mice die shortly after birth, 40% of the line produced by Sun et al survive and show unexplained premature aging (Sun et al 2004). As premature aging is a hallmark of increased acquisition of DNA damage there is the possibility of a conserved role for LSH in mammalian DNA damage repair. Here I show that LSH depleted mammalian cells have an increased sensitivity specifically to DSB inducing agents and show increased levels of apoptosis. Further analysis shows that cells lacking LSH repair DSBs slower, indicating a novel role for LSH in mammalian repair of DSB. I performed an in depth analysis of the DSB defects in LSH depleted cells in an attempt to elucidate the function of LSH in DSB repair. I found that LSH depleted cells can correctly recognise DSBs but recruit downstream signalling and repair factors, such as γH2AX, less efficiently. I show that reduced recruitment of downstream DSB repair factors is not accompanied by extended cell cycle checkpoint signalling. This suggests that LSH depleted cells continue through the mitosis with unrepaired DSBs, which most likely leads to apoptosis and the increased sensitivity to DSB inducing agents. These experiments also showed that recruitment of DSB signalling and repair factors is not impaired equally at all breaks, and I present a model system created to quantitatively compare individually breaks between WT and LSH depleted cells to identify DSB that require LSH for efficient repair. I also preformed an analysis of Lsh-­/-­ MEFs containing WT or catalytic null mutant LSH rescue constructs and I show that WT but not catalytic null LSH can restore efficient DSB repair. These studies identify a novel role for LSH in mammalian DSB repair and demonstrate the importance of its catalytic activity.

Avec la révolution numérique de cette dernière décennie, la quantité de photos numériques mise à disposition de chacun augmente plus rapidement que la capacité de traitement des ordinateurs. Les outils de recherche actuels ont été conçus pour traiter de faibles volumes de données. Leur complexité ne permet généralement pas d'effectuer des recherches dans des corpus de grande taille avec des temps de calculs acceptables pour les utilisateurs. Dans cette thèse, nous proposons des solutions pour passer à l'échelle les moteurs de recherche d'images par le contenu. Dans un premier temps, nous avons considéré les moteurs de recherche automatique traitant des images indexées sous la forme d'histogrammes globaux. Le passage à l'échelle de ces systèmes est obtenu avec l'introduction d'une nouvelle structure d'index adaptée à ce contexte qui nous permet d'effectuer des recherches de plus proches voisins approximées mais plus efficaces. Dans un second temps, nous nous sommes intéressés à des moteurs plus sophistiqués permettant d'améliorer la qualité de recherche en travaillant avec des index locaux tels que les points d'intérêt. Dans un dernier temps, nous avons proposé une stratégie pour réduire la complexité de calcul des moteurs de recherche interactifs. Ces moteurs permettent d'améliorer les résultats en utilisant des annotations que les utilisateurs fournissent au système lors des sessions de recherche. Notre stratégie permet de sélectionner rapidement les images les plus pertinentes à annoter en optimisant une méthode d'apprentissage actif. / In this last decade, would the digital revolution and its ancillary consequence of a massive increases in digital picture quantities. The database size grow much faster than the processing capacity of computers. The current search engine which conceived for small data volumes do not any more allow to make searches in these new corpus with acceptable response times for users.In this thesis, we propose scalable content-based image retrieval engines.At first, we considered automatic search engines where images are indexed with global histograms. Secondly, we were interested in more sophisticated engines allowing to improve the search quality by working with bag of feature. In a last time, we proposed a strategy to reduce the complexity of interactive search engines. These engines allow to improve the results by using labels which the users supply to the system during the search sessions.

DNA methylation is a key epigenetic mark which undergoes global reprogramming during early mammalian embryonic development, resulting in almost complete erasure of the mark after fertilisation of the zygote. Genome-wide patterns of DNA methylation are subsequently re-established in the implanting blastocyst by de novo DNA methyltransferases Dnmt3a and Dnmt3b along with their catalytically inactive co-factor Dnmt3l, while these DNA methylation patterns are maintained through cell divisions by maintenance methyltransferase Dnmt1. The exact mechanisms by which these DNA methyltransferase enzymes are targeted to specific genomic regions remain unclear, but may involve interaction with modified histones and/or the participation of co-factors. Lsh (lymphoid specific helicase), a putative chromatin remodelling helicase, has been implicated in facilitating de novo methylation, as Lsh knockout embryos and derived somatic cell lines display substantial but specific DNA methylation losses at repetitive elements and single copy genes. This study aims to define the requirement for Lsh in establishing de novo DNA methylation and gene expression patterns during the early stages of mouse embryonic development. The '2i' culture system using two small molecule kinase inhibitors was harnessed to convert lsh-/- mouse embryonic stem cells (mESCs) to a hypomethylated 'ground state' of pluripotency. Culture conditions were then altered to transition these ground state mESCs to cells representing later, more methylated stages of development ('serum' mESCs, epiblast stem cells and embryoid bodies). Implementation of this model system suggests that Lsh does not contribute to DNA methylation establishment in a pluripotent context, but rather is important for facilitating de novo DNA methylation during differentiation to culture models representing later developmental stages. These investigations also reveal that Lsh differentially regulates DNA methylation at major and minor satellite repeats depending on cellular context, and that this regulation may involve a role for Lsh in maintenance of DNA methylation.

DNA methylation is essential for mammalian development and transcriptional repression of genes and retrotransposons during embryo development and in somatic cells. The patterns of DNA methylation are established by de novo DNA methyltransferases, which are regulated by developmental signalling and require access to chromatin. Besides DNA methyltransferases, other proteins have recently been implicated in DNA methylation, such as the ATP-dependent chromatin remodeler LSH. The absence of LSH in mouse embryos leads to defects in DNA methylation and development. In relation to this, mutations in LSH have been found to cause Immunodeficiency-Centromeric instability-Facial anomalies (ICF) syndrome. This syndrome is characterized by centromeric instability and CpG hypomethylation of centromeric satellite repeats, and is most often caused by mutations in the catalytic domain of the DNA methyltransferase DNMT3B. LSH is essential for developmentally programmed de novo DNA methylation of large chromosomal domains including promoters of protein coding genes and repetitive sequences. Importantly, fibroblasts derived from chromatin remodeling ATPase LSH-null mouse embryos, which lack DNA methylation at transposons and specific gene promoters, are capable of re-establishing normal patterns of DNA methylation and transcriptional silencing of misregulated genes upon re-expression of LSH. The ATP hydrolysis by LSH is essential for its function in gene silencing and de novo DNA methylation. However, the molecular mechanisms of LSH-dependent gene silencing and de novo DNA methylation are yet unclear. Here we use an inducible system that enables controlled expression of LSH in Lsh-null mouse embryonic fibroblasts (MEFs) to follow chromatin dynamics, transcriptional silencing and establishment of de novo DNA methylation. This conditionally reversible Lsh knockout cellular system allowed us to study the order of events occurring immediately after LSH restoration in MEF cell lines in order to elucidate the molecular mechanism of LSH-dependent gene silencing. We have demonstrated that LSH upon its restoration localises to the promoters of LSH-dependent loci leading to a mild decrease in the occupancy of H3, which reinforces the previously shown role of LSH as a chromatin remodeler. Simultaneously, there is removal of acetyl groups from H3 tails when LSH is bound to these target regions, which might be facilitated by the interaction of HDACs with LSH. The removal of H3Ac marks is followed by deposition of H3K9me2 by G9a/GLP histone methylases at the same time point when misregulated genes are silenced. This suggests that LSH creates a suitable substrate for G9a/GLP promoting gene silencing. Surprisingly, transcriptional repression occurs without acquisition of DNA methylation at the promoters of these loci. This order of events implies that LSH plays a role as a chromatin remodeler leading to changes in chromatin structure and modifications that facilitate epigenetic gene silencing without DNA methylation in the initial period when LSH is restored in MEF cell lines. Furthermore, deposition of H3K9me2 by the G9a/GLP complex is critical for silencing of specific genes, but not for repetitive elements such as IAPs. The histone modification H3K27me3 seems to play a transitory role in the silencing of IAP retrotransposons in the absence of G9a/GLP activity. In conclusion, this work has demonstrated that changes in chromatin modifications leading to a transcriptionally repressive chromatin state can be established in somatic cells by the chromatin remodeler LSH without acquisition of DNA methylation. This suggests that the primary role of LSH is to promote changes in chromatin structure and modifications that lead to gene silencing and not DNA methylation, which most likely occurs as a consequence of transcriptional silencing.

Till grund för denna uppsats ligger datamaterial från LSH-studien -Livsvillkor, Stress och Hälsa, som har genomförts av forskare vid Hälsouniversitetet vid Linköpings universitet. Med studien undersöker forskarna hur människors livsvillkor kan ge upphov till stress och hur stress i sin tur kan påverka hälsan. Datamaterialet är insamlat år 2003, det omfattar tio vårdcentraler runt om i Östergötland och består av drygt 1000 individer i åldern 45-69 år jämnt fördelade över ålder och kön mellan vårdcentralerna. Syftet med denna uppsats är att identifiera och försöka justera variationer i resultaten mellan de tio medverkande vårdcentralerna i LSH-studien. Den statistiska analysen är genomförd i tre steg. I steg ett har ett stort antal variabler valts ut från ursprungsmaterialet. Därefter har skillnader identifierats med Chi2-tester och variansanalys för några av variablerna rörande självskattad hälsa, stress, psykosociala faktorer samt traditionella riskfaktorer för hjärt- och kärlsjukdomar. I steg två har vi med bakgrundsfaktorer försökt justera dessa skillnader. Av de 19 variabler som har analyserats med linjär eller logistisk regression är det fem där bakgrundsfaktorerna förklarar skillnaderna mellan vårdcentralerna. Frapperande är att samtliga fem variabler har med den psykiska hälsan att göra, ingen av variablerna angående fysisk hälsa eller riskfaktorer har justerats. Efter justering för bakgrundsfaktorerna kvarstår skillnader för fyra variabler om psykisk hälsa, dessa är Tillit, Upplevelse av stress, Upplevelse av sömnkvalitet, Psykiskt välbefinnande. Även för samtliga tio riskfaktorer och mätvärden kvarstår skillnader, dessa är Puls, P-LDL-Kolesterol, Systoliskt blodtryck logaritmerad, Systoliskt blodtryck utan uteliggare, Diagnostiserad diabetes, Medicin för blodtryck de senaste två veckorna, Alkohol riskgrupper, Motion totalt, B-Hemoglobin, P-Glukos logaritmerad. Orsaker till dessa skillnader kan vi inte fastställa, men för mätdata kan mätfel vara en tänkbar anledning. I sista steget, steg tre, har vi tagit upp analysen ett plan till vårdcentralsnivå. Genom ekologisk korrelation har vi undersökt samband mellan inkomst och betaskattningar för vårdcentralerna när justeringar har gjorts för bakgrundsfaktorer. För variabeln Tillit har vi funnit ett positivt samband, där högre inkomst i ett område innebär generellt mer tillit. Dock avviker vårdcentralen i Ödeshög där man istället har en hög grad av tillit trots relativt låg inkomst.

<p>Till grund för denna uppsats ligger datamaterial från LSH-studien -Livsvillkor, Stress och Hälsa, som har genomförts av forskare vid Hälsouniversitetet vid Linköpings universitet. Med studien undersöker forskarna hur människors livsvillkor kan ge upphov till stress och hur stress i sin tur kan påverka hälsan. Datamaterialet är insamlat år 2003, det omfattar tio vårdcentraler runt om i Östergötland och består av drygt 1000 individer i åldern 45-69 år jämnt fördelade över ålder och kön mellan vårdcentralerna.</p><p>Syftet med denna uppsats är att identifiera och försöka justera variationer i resultaten mellan de tio medverkande vårdcentralerna i LSH-studien. Den statistiska analysen är genomförd i tre steg. I steg ett har ett stort antal variabler valts ut från ursprungsmaterialet. Därefter har skillnader identifierats med Chi2-tester och variansanalys för några av variablerna rörande självskattad hälsa, stress, psykosociala faktorer samt traditionella riskfaktorer för hjärt- och kärlsjukdomar. I steg två har vi med bakgrundsfaktorer försökt justera dessa skillnader. Av de 19 variabler som har analyserats med linjär eller logistisk regression är det fem där bakgrundsfaktorerna förklarar skillnaderna mellan vårdcentralerna. Frapperande är att samtliga fem variabler har med den psykiska hälsan att göra, ingen av variablerna angående fysisk hälsa eller riskfaktorer har justerats.</p><p>Efter justering för bakgrundsfaktorerna kvarstår skillnader för fyra variabler om psykisk hälsa, dessa är Tillit, Upplevelse av stress, Upplevelse av sömnkvalitet, Psykiskt välbefinnande. Även för samtliga tio riskfaktorer och mätvärden kvarstår skillnader, dessa är Puls, P-LDL-Kolesterol, Systoliskt blodtryck logaritmerad, Systoliskt blodtryck utan uteliggare, Diagnostiserad diabetes, Medicin för blodtryck de senaste två veckorna, Alkohol riskgrupper, Motion totalt, B-Hemoglobin, P-Glukos logaritmerad. Orsaker till dessa skillnader kan vi inte fastställa, men för mätdata kan mätfel vara en tänkbar anledning.</p><p>I sista steget, steg tre, har vi tagit upp analysen ett plan till vårdcentralsnivå. Genom ekologisk korrelation har vi undersökt samband mellan inkomst och betaskattningar för vårdcentralerna när justeringar har gjorts för bakgrundsfaktorer. För variabeln Tillit har vi funnit ett positivt samband, där högre inkomst i ett område innebär generellt mer tillit. Dock avviker vårdcentralen i Ödeshög där man istället har en hög grad av tillit trots relativt låg inkomst.</p>

Les travaux décrits portent sur des matériaux hybrides organiques-inorganiques présentant des propriétés magnétiques, de formule générale Mx(OH)2x-ny(Xn-)y (avec M(II) = Co, Cu, Zn, X= phtalocyanines métallées tétrasulfonate, MPc). Deux objectifs distincts ont guidé cette thèse : coupler la conductivité de la MPc au magnétisme des plans, d'une part, et déterminer expérimentalement, d'autre part, les couplages entre plans de Co conduisant à la mise en ordre autour de 6 K. Des étapes préliminaires de synthèse originale ont été nécessaires concernant la LiPc tétrasulfonate et des HSL de Zn. Ni la synthèse de la LiPc tétrasulfonate, ni les essais de dopage des MPc n'ont été suffisamment concluant pour pouvoir bénéficier des conductivités des MPc dans l'hybride associé. Les mesures magnétiques réalisées par SQUID ont montré que les HSL à base de feuillets de Cu sont antiferromagnétique sans mise en ordre. Ceux dont les feuillets sont constitués de Co s'ordonnent ferrimagnétiquement autour de 6 K quelle que soit la nature de Pc insérée, prouvant qu'il n'y a pas d'échange magnétique entre les feuillets,et que seule l’ interaction dipolaire est responsable de la mise en ordre. La CuPc insérée a joué le rôle de sonde paramagnétique et a permis, grâce à des mesures par RPE en bande X et en bande Q, d'estimer la valeur du champ dipolaire à 30 mT. Ces études valident le modèle théorique proposé par M. Drillon et al. Des études ont été réalisées sur CoPc insérées. qui sont majoritairement muettes en RPE du fait de leur agrégation et ne constituent pas de bonnes sondes RPE. Les HSL MCoPc montrent un pouvoir de réduction de l'O2. / This PhD work focus on hybrid layered organic-inorganic materials, with general formulation Mx(OH)2x-ny(Xn-)y (M(II)=Co, Cu; X- = metal phthalocyanine sulfate MPcTs). Two leading ideas founded this PhD work: i) the investigation of the possible interplay between the magnetic properties of the LSH hosts and possible charge transport through phthalocyanine guests ii) the experimental evidence of the underlying mechanism for magnetic ordering in LSH. Synthesis of new molecules was required, as LiPcTs and Zn LSH. The yield of the synthesis of LiPcTs was not high to allow for subsequent insertion. The next route under investigation considered the anions and cations issued from the reduction and oxidation of CuPc and CoPc. Those revealed much too instable to be inserted into lamellar hosts. The investigations of the magnetic properties by SQUID magnetometry showed that all copper hydroxide-based compounds behave similarly, presenting an overall antiferromagnetic behaviour without magnetic ordering. The cobalt HSL compound presents a ferrimagnetic ordering around 6K, whatever the nature of the metal phthalocyanine. This points to the solely dipolar interactions between the inorganic layers. Using the inserted CuPc as a fine paramagnetic probe, X-band and Q-band EPR spectroscopy allowed to assess the amplitude of the dipolar field Bdipolar≈30mT. These measurements validate the theoretical model of M. Drillon and al on the role of dipolar interactions in layered ferromagnets. The formation of EPR silent aggregates of inserted CoPc discards it as a suitable probe for similar experiments. Preliminary results show that these hybrids catalyze the reduction of O2.

This thesis deals with the clustering of tunnels in data obtained from the protein molecular dynamics simulation. This process is very computationaly intensive and it has been a challenge for scientific communities. The goal is to find such an algorithm with optimal time and space complexity ratio. The research of clustering algorithms, work with huge highdimensional datasets, visualisation and cluster-comparing methods are discussed. The thesis provides a proposal of the solution of this problem using the Twister Tries algorithm. The implementation details are analysed and the testing results of the solution quality and space complexity are provided. The goal of the thesis was to prove that we could achieve the same results with a stochastic algorithm - Twister Tries , as with an exact algorithm ( average-linkage ). This assumption was not confirmed confidently. Another finding of the hashing functions analysis shows that we could obtain the same results of hashing with a low dimensional hashing function but in much better computational time.