Le domaine scientifique de la Biologie des Systèmes étudie les interactions entre les composantes d'un système biologique afin d'en comprendre son fonctionnement global. Au cours de cette these, nous avons d’abord utilisé des graphes simples. Cette approche a permis d' appréhender la manière dont un réseau biologique peut interagir avec son environnement, lui-même modélisé par un autre réseau. Nous avons ensuite défini le formalisme MIB (Model of Interactions in Biology) qui permet de définir, rechercher et étudier les motifs hétérogènes. Enfin pour approfondir l'étude de la structure et de la dynamique, nous avons proposé le formalisme MIN. MIN possède la structure bipartie de MIB, mais permet d'avoir des annotations beaucoup plus riches des noeuds et des arcs du réseau qui peuvent être utilisées pour la traduction des données automatiquement en d'autres formalismes couramment utilisés en modélisation biologique, tels que les équations différentielles ou la modélisation logique. / The scientific domain of the Systems Biology studies the interactions between the components of a biological system in order to understand its functioning as a whole. In this thesis, we first used searched to apprehend how a biological network, modelled as a simple graph, interact with its environment, modelled by another graph. Next, we have defined the MIB formalism (for Model of Interactions in Biology) that enables to model, to search and to study the heterogeneous motifs in biological networks. Finally, for deepening the study of structure and dynamics of biological networks, we have proposed the MIN formalism (for Modular Interaction Network). MIN inherited the bipartite structure of MIB, but also includes the richer annotations for nodes, arcs and possible states of the network, thus enabling the automatic translation of data contained in MIN into other formalisms commonly used in biology for dynamics modelling, such as logical networks, differential equations or Petri nets.

Several biomolecular pathways governing the control of cellular processes have been discovered over the last several years. Additionally, advances resulting from combining these pathways into networks have produced new insights into the complex behaviors observed in cell function assays. Unfortunately, identification of important subnetworks, or “motifs”, in these networks has been slower in development. This study focused on identifying important network motifs and their rate of occurrence in two different biomolecular networks. The two networks evaluated for this study represented both ends of the spectrum of interaction knowledge by comparing a well defined network (apoptosis) with and poorly studied network that was early in development (autism). This study identified several motifs that could be important in governing and controlling cellular processes in healthy and diseased cells. Additionally, this study revealed an inverse relationship when comparing the occurrence rate of these motifs in apoptosis and autism.

One of the primary goals of systems biology is to understand the complex underlying network of biochemical interactions which allow an organism to respond to environmental stimuli. Models of these biological interactions serve as a tool to both codify current understanding of these interactions as well as a starting point for scientific discovery. Due to the massive amount of information which is required for this modeling process, systems biology studies must often attempt to construct models which reflect the whole of the system while having access to only partial information. In some cases, the missing information will not have a confounding effect on the accuracy of the model. In other cases, there is the danger that this missing information will make the model useless. The focus of this thesis is to study the effect which missing information has on systems level studies within several different contexts. Specifically, we study two contexts : when the missing information takes the role of incomplete molecular interaction network knowledge and when it takes the role of unknown kinetic rate laws. These studies yield interesting results. We show that when metabolism is isolated from gene expression, the effects are not limited to those reactions under strong control by gene expression. Thus, incomplete understanding of molecular interaction networks may have unexpected effects on the resulting analysis. We also reveal that under the conditions of the current study, mass action was shown to be the superior substitute when the true rate equations for a biological system are unknown. In addition to studying the effect of missing information in the aforementioned contexts, we propose a method for limiting the parameter search space of biochemical systems. Even in ideal scenarios where both the molecular interaction network and the relevant kinetic rate equations are known, obtaining appropriate estimates for the unknown system parameters can be challenging. By employing a method which limits the parameter search space, we are able to acquire estimates for parameter values which are much closer to the true values than those which could be obtained otherwise. / Ph. D.

In the area of information visualization social or biological networks are visualized ina way so that they can be explored easily and one can get more information about thestructure of the network out of it. The use of network centralities in the field of network analysis plays an importantrole when it comes to the rating of the relative importance of vertices within the networkstructure based on the neighborhood of them. Such a single network can be renderedeasily by the use of standard graph drawing algorithms. But it is not only the explorationof one centrality which is important. Furthermore, the comparison of two or more of themis important to get some further meaning out of it. When visualizing the comparisonof two or more network centralities we are facing new problems of how to visualizethem in a way to get out the most meaning of it. We want to be able to track all thechanges in the networks between two centralities as well as visualize the single networksas best as possible. In the life sciences centrality measures help scientists to understand theunderlying biological processes and have been successfully applied to different biologicalnetworks. The aim of the thesis is it to overcome those problems and to come up with a new solutionof how to visualize networks and its centralities. This thesis introduces a new way ofrendering networks including their centrality values along a circular view. Researches canthen be focused on the exploration of the centrality values including the network structure,without dealing with visual clutter or occlusions of nodes. Furthermore, filtering based instatistical data concerning the datasets and centrality values support this.

Στην εποχή της Συστημικής Ιατρικής, οι τεχνολογίες μαζικής καταγραφής της γονιδιακής και miRNA έκφρασης (π.χ. μικροσυστοιχίες, RNA-seq) αλλά και οι τεχνολογίες ανίχνευσης πρωτεϊνικών αλληλεπιδράσεων (π.χ. yeast two-hybrid, co-immunoprecipitation) απελευθέρωσαν τεράστια ποσά δεδομένων για την αποσαφήνιση των μηχανισμών των πολύπλοκων ασθενειών. Η παρούσα διδακτορική διατριβή συμβάλλει προσφέροντας νέες υπολογιστικές μεθοδολογίες και εργαλεία και παραθέτοντας νέες αξιόπιστες βιολογικές υποθέσεις για την επίλυση σύνθετων ασθενειών του ανθρώπου. Καταρχήν, αποκτήθηκε γνώση του θεωρητικού υπόβαθρου διάφορων μέγαλης κλίμακας μοριακών τεχνικών, τεχνικών εξόρυξης δεδομένων όπως η ομαδοποίηση καθώς και γραφοθεωρητικών προσεγγίσεων. Έπειτα, σχεδιάστηκε μια μεθοδολογία για συνδυασμό πρωτεωμικών και μεταγραφωμικών δεδομένων και αναπτύχθηκε ένα αλγόριθμος ομαδοποίησης γράφων, που ονομάζεται Detect Modules (DetMod), ο οποίος ανιχνεύει κοινοτήτες/υπο-δομές (modules) πρωτεϊνών με διακριτή βιολογική λειτουργία και έντονη δυναμική συσχέτιση σε επίπεδο έκφρασης. Η απόδοση και αξιοπιστία της μεθόδου εξετάστηκε και πιστοποιήθηκε στον απλό οργανισμό-μοντέλο Saccharomyces cerevisiae προτού εφαρμοστεί στην επίλυση προβλημάτων της φαρμακογονιδιωματικής όπως η απόκριση του μεταγραφήματος στην θεραπεία με ταμοξιφένη στην περίπτωση του θετικού στην απόκριση σε οιστρογόνα καρκίνου του μαστού. Αποτέλεσμα της μεθόδου είναι δυναμικοί βιοδείκτες της απόκρισης στην ταμοξιφένη με μορφή υπο-δομών αντί μεμονωμένων πρωτεϊνών. Παράλληλα, στα πλαίσια της σύγχρονης βιβλιογραφίας όπου οι εμπλεκόμενοι μηχανισμοί του καρκίνου αλληλοεπικαλύπτονται με αυτούς της γήρανσης, μια προσαρμοσμένη μεθοδολογία ανάλογη με την προαναφερόμενη εφαρμόστηκε στη μελέτη του φαινομένου της γήρανσης. Τα αποτελέσματα της μεθόδου σε πολλαπλούς ιστούς του ποντικού, και σε δεύτερο στάδιο μεμονωμένα στον καρδιακό ιστό, ανέδειξαν ποια μοριακά μονοπάτια εμπλέκονται στη γήρανση όλων των ιστών και ποια εξειδικεύονται σε ένα μόνο ιστό. Στην περίπτωση του καρδιακού ιστού βιοδείκτες σε μορφή υπο-δομών αποτυπώνουν τα εμπλεκόμενα μονοπάτια αλλά και τη συνεργατική δράση και υπαιτιότητα των miRNA. Σε επόμενο στάδιο μελετήθηκαν οι μηχανισμοί απόκρισης στη γρίπη Α (Η1Ν1) μέσω της ανακατασκευής Γονιδιακών Ρυθμιστικών Δικτύων (ΓΡΔ) που αναπαριστούν τις χρονικά μεταβαλλόμενες αιτιατές σχέσεις μεταξύ μοριακών μονοπατιών από χρονοσειρές γονιδιακής έκφρασης. Το χρονικά μεταβαλλόμενο ΓΡΔ προέκυψε μέσα από μια μέθοδο συνδυασμού πολλαπλών αλγορίθμων ανακατασκευής από διαφορετικές κλάσεις του μαθηματικού φορμαλισμού. Η μέθοδος προσέφερε νέα γνώση για τη συνδεσιμότητα των μοριακών μονοπατιών μέχρι και την 60η ημέρα μετά την εισβολή του ιού στον πνευμονικό ιστό του ποντικού από το στάδιο της φυσικής ανοσίας, στη χυμική ανοσία και τέλος στη διαδικασία αποκατάστασης. Τέλος, παρουσιάζεται ο OLYMPUS, ένας νέος υβριδικός μη επιβλεπόμενος αλγόριθμος ομαδοποίησης που εφαρμόστηκε σε χρονοσειρές γονιδιακής έκφρασης σε απόκριση στη γρίπη Α (Η1Ν1). Ο OLYMPUS χρησιμοποιεί τον Διαφορεξελικτικό αλγόριθμο ως στρατηγική βελτιστοποίησης ενός ασαφούς αλγορίθμου ομαδοποίησης και παράλληλα ενσωματώνει το κριτήριο Bayesian Information με σκοπό την αυτόματη εύρεση του βέλτιστου αριθμού ομάδων. Η ανάλυση των εξαγόμενων ομάδων προσέφερε νέες υποθέσεις σχετικά με τη δυναμική πολλών μοριακών μονοπατιών που εμπλέκονται στην ανοσολογική απόκριση και για πρώτη φορά αναδείχθηκε ο ρόλος των κατασταλμένων διεργασιών στο κινητικό μοντέλο της γρίπης Α. / In the Systems Medicine era, the large scale gene and miRNA expression techniques (e.g. microarrays, RNA-seq) as well as techniques for the detection of protein interactions (e.g. yeast two-hybrid, co-immunoprecipitation) have released mass amounts of data for deciphering the underlying mechanisms of complex diseases. The present PhD thesis contributes by providing new computational methodologies and tools, and by offering novel biological hypotheses for solving complex human diseases. Initially, a good grasp of the current high-throughput molecular techniques was acquired along with familiarization with data mining tecniques such as clustering and with graph-theoretic approaches. Then, a methodology for integrating proteomic and transcriptomic data was designed and a graph clustering algorithm was developed, called Detect Modules (DetMod), which detects, on the composite transcriptome-proteome network, communities/modules with distinct biological function and enhanced association at the dynamic expression level. The performance and reliability of the method was tested and validated in the simple model organism Saccharomyces cerevisiae before solving pharmacogenomics problems such as the transcriptome response mechanisms during tamoxifen response in estogen-response-positive breast cancer cases. The output of the method was dynamic biomarkers of tamoxifen response in the form of modules instead of individual proteins. In parallel, the recent literature associates the mechanisms involved in cancer with those involved in aging. In this context, an adapted methodology similar to the aforementioned was applied in the study of aging. The findings of the method in multiple mouse tissues and on second level in cardiac tissue highlighted the cross-tissue aging molecular pathways as well as the tissue-specific. In the case of cardiac tissue modular biomarkers captured the underlying tissue-specific pathways as well as the synergism of miRNAs. On next level, the response mechanisms to Influenza A (H1N1) were explored through the reconstruction of Gene Regulatory Networks (GRNs), which in turn represent the time-varying causal pathway interactions based on time series expression data. The final time-varying GRN was derived from an ensemble of reconstruction algorithms from different classes of mathematical formalism. The method offered new knowledge for the pathway interactivity until the day 60 after the viral invasion in the mouse lung tissue, from the innate response to the humoral and the late repair phase. Finally, OLYMPUS is presented, a novel unsupervised hybrid clustering algorithm which was applied at time series expression data in response to Influenza A (H1N1). OLYMPUS uses the Differential Evolutionary algorithm as optimization strategy of a fuzzy clustering algorithm and in parallel integrates the Bayesian Information Criterion in order to detect automatically the optimal cluster number. The cluster analysis offered new hypotheses regarding the dynamics of several molecular pathways and for the first time, the role of suppressed biological processes was highlighted in the Influenza A kinetic model.

Biological network data is plentiful with practically every experimental methodology giving 'network views' into cellular function and behavior. Bioinformatic screens that yield network data include, for example, genome-wide deletion screens, protein-protein interaction assays, RNA interference experiments, and methods to probe metabolic pathways. Efficient and comprehensive computational approaches are required to model these screens and gain insight into the nature of biological networks. This thesis presents three new algorithms to model and mine network datasets. First, we present an algorithm that models genome-wide perturbation screens by deriving relations between phenotypes and subsequently using these relations in a local manner to derive genephenotype relationships. We show how this algorithm outperforms all previously described algorithms for gene-phenotype modeling. We also present theoretical insight into the convergence and accuracy properties of this approach. Second, we define a new data mining problem–constrained minimal separator mining—and propose algorithms as well as applications to modeling gene perturbation screens by viewing the perturbed genes as a graph separator. Both of these data mining applications are evaluated on network datasets from S. cerevisiae and C. elegans. Finally, we present an approach to model the relationship between metabolic pathways and operon structure in prokaryotic genomes. In this approach, we present a new pattern class—biclusters over domains with supplied partial orders—and present algorithms for systematically detecting such biclusters. Together, our data mining algorithms provide a comprehensive arsenal of techniques for modeling gene perturbation screens and metabolic pathways. / Ph. D.

In the last few years, large-scale experiments have generated genome-wide protein interaction networks for many organisms including Saccharomyces cerevisiae (baker's yeast), Caenorhabditis elegans (worm) and Drosophila melanogaster (fruit fly). In this thesis, we examine the vertex and edge betweenness centrality measures of these graphs. These measures capture how "central" a vertex or an edge is in the graph by considering the fraction of shortest paths that pass through that vertex or edge. Our primary observation is that the distribution of the vertex betweenness centrality follows a power law, but the distribution of the edge betweenness centrality has a Poisson-like distribution with a very sharp spike. To investigate this phenomenon, we generated random networks with degree distribution identical to those of the protein interaction networks. To our surprise, we found out that the random networks and the protein interaction networks had almost identical distribution of edge betweenness. We conjecture that the "Poisson-like" distribution of the edge betweenness centrality is the property of any graph whose degree distribution satisfies power law. / Master of Science

No description available.

京都大学 / 0048 / 新制・課程博士 / 博士(薬科学) / 甲第19673号 / 薬科博第61号 / 新制||薬科||7(附属図書館) / 32709 / 京都大学大学院薬学研究科医薬創成情報科学専攻 / (主査)教授 馬見塚 拓, 教授 掛谷 秀昭, 教授 緒方 博之 / 学位規則第4条第1項該当 / Doctor of Pharmaceutical Sciences / Kyoto University / DFAM

Dans de nombreuses applications telles que la médecine, l'environnement ou les biotechnologies par exemple, la découverte de nouveau processus de régulations de gènes permet une meilleure compréhension des réponses phénotypiques des cellules à des stimuli externes. Pour cela, il est alors d'usage de générer et d'analyser les données transcriptomiques issues d'expériences de types puces à ADN ou plus récemment de RNAseq. Ainsi, pour chaque gène d'un organisme d'étude placé dans différentes conditions expérimentales, un ensemble de niveau d'expression est obtenu. A partir de ces données, les mécanismes de régulation des gènes peuvent être obtenus à travers un ensemble de liens dans des graphes. Dans ces réseaux, les nœuds correspondent aux gènes. A lien entre deux nœuds est identifié si une relation de régulation existent entre les deux gènes correspondant. De tels réseaux sont appelés Réseaux de Régulation de Gènes (RRGs). Malgré la profusion de méthodes d'inférence disponible, leur construction et leur analyse restent encore à ce jour un défi.Dans cette thèse, nous proposons de répondre au problème d'inférence de réseaux par des techniques d'optimisation dans des graphes. A partir d'information de régulation sur l'ensemble des couples de gènes, nous proposons de déterminer la présence d'arêtes dans le RRG final en adoptant une formulation de fonction objectif intégrant des contraintes. Des a priori à la fois biologiques (sur les interactions entre les gènes) et structuraux (sur la connectivité des nœuds) ont été considérés pour restreindre l'espace des solutions possibles. Les différents a priori donnent des fonctions objectifs ayant des propriétés différentes, pour lesquelles des stratégies d'optimisation adaptées (continue et/ou discrète) peuvent être appliquées. Les post-traitement que nous avons développé ont mené à un ensemble de méthodes nommés BRANE, pour "Biologically-Related A priori for Network Enhancement". Pour chacune des méthodes développées (BRANE Cut, BRANE Relax et BRANE Clust), nos contributions sont triples : formulation de la fonction objectif à l'aide d'a priori, développement de la stratégie d'optimisation et validation (numérique et biologique) sur des données de parangonnage issues des challenges DREAM4 et DREAM5, montrant ainsi des améliorations pouvant atteindre 20%.En complément de l'inférence de réseaux, notre travail s'est étendu à des traitements de données sur graphe plus génériques, tels que les problèmes inverses. Nous avons notamment étudié HOGMep, une approche Bayésienne utilisant des stratégies d'approximation Bayésienne variationnelle. Cette méthode a été développée pour résoudre de façon conjointe, des problèmes de restauration et de classification sur des données multi-composantes (signaux et images). Les performances d'HOGMep dans un contexte de déconvolution d'image couleur montrent de bonnes qualités de reconstruction et de segmentation. Une étude préliminaire dans un contexte de classification de données médicales liant génotype et phénotype a également montré des résultats prometteurs pour des adaptions à venir en bioinformatiques. / The discovery of novel gene regulatory processes improves the understanding of cell phenotypicresponses to external stimuli for many biological applications, such as medicine, environmentor biotechnologies. To this purpose, transcriptomic data are generated and analyzed from mi-croarrays or more recently RNAseq experiments. For each gene of a studied organism placed indifferent living conditions, they consist in a sequence of genetic expression levels. From thesedata, gene regulation mechanisms can be recovered by revealing topological links encoded ingeometric graphs. In regulatory graphs, nodes correspond to genes. A link between two nodesis identified if a regulation relationship exists between the two corresponding genes. Such net-works are called Gene Regulatory Networks (GRNs). Their construction as well as their analysisremain challenging despite the large number of available inference methods.In this thesis, we propose to address this network inference problem with recently developedtechniques pertaining to graph optimization. Given all the pairwise gene regulation informa-tion available, we propose to determine the presence of edges in the final GRN by adoptingan energy optimization formulation integrating additional constraints. Either biological (infor-mation about gene interactions) or structural (information about node connectivity) a priorihave been considered to reduce the space of possible solutions. Different priors lead to differentproperties of the global cost function, for which various optimization strategies can be applied.The post-processing network refinements we proposed led to a software suite named BRANE for“Biologically-Related A priori for Network Enhancement”. For each of the proposed methodsBRANE Cut, BRANE Relax and BRANE Clust, our contributions are threefold: a priori-based for-mulation, design of the optimization strategy and validation (numerical and/or biological) onbenchmark datasets.In a ramification of this thesis, we slide from graph inference to more generic data processingsuch as inverse problems. We notably invest in HOGMep, a Bayesian-based approach using aVariation Bayesian Approximation framework for its resolution. This approach allows to jointlyperform reconstruction and clustering/segmentation tasks on multi-component data (for instancesignals or images). Its performance in a color image deconvolution context demonstrates bothquality of reconstruction and segmentation. A preliminary study in a medical data classificationcontext linking genotype and phenotype yields promising results for forthcoming bioinformaticsadaptations.