Cell functions are steered by complex interactions of gene products, like forming a temporary or stable complex, altering gene expression or catalyzing a reaction. Mapping these interactions is the key in understanding biological processes and therefore is the focus of numerous experiments and studies. Small-scale experiments deliver high quality data but lack coverage whereas high-throughput techniques cover thousands of interactions but can be error-prone. Unfortunately all of these approaches can only focus on one type of interaction at the time. This makes experimental mapping of the genome-wide network a cost and time intensive procedure. However, to overcome these problems, different computational approaches have been suggested that integrate multiple data sets and/or different evidence types. This widens the stringent definition of an interaction and introduces a more general term - functional association.  FunCoup is a database for genome-wide functional association networks of Homo sapiens and 16 model organisms. FunCoup distinguishes between five different functional associations: co-membership in a protein complex, physical interaction, participation in the same signaling cascade, participation in the same metabolic process and for prokaryotic species, co-occurrence in the same operon. For each class, FunCoup applies naive Bayesian integration of ten different evidence types of data, to predict novel interactions. It further uses orthologs to transfer interaction evidence between species. This considerably increases coverage, and allows inference of comprehensive networks even for not well studied organisms.  BinoX is a novel method for pathway analysis and determining the relation between gene sets, using functional association networks. Traditionally, pathway annotation has been done using gene overlap only, but these methods only get a small part of the whole picture. Placing the gene sets in context of a network provides additional evidence for pathway analysis, revealing a global picture based on the whole genome. PathwAX is a web server based on the BinoX algorithm. A user can input a gene set and get online network crosstalk based pathway annotation. PathwAX uses the FunCoup networks and 280 pre-defined pathways. Most runs take just a few seconds and the results are summarized in an interactive chart the user can manipulate to gain further insights of the gene set's pathway associations. / <p>At the time of the doctoral defense, the following paper was unpublished and had a status as follows: Paper 2: Manuscript.</p>

Dans cette thèse, nous cherchons à caractériser les facteurs de transcription de la plante Arabidopsis thaliana, gènes importants pour la régulation de l'expression du génome. À l'aide de données d'expression, notre objectif biologique est de classer ces facteurs de transcription en groupes de gènes co-régulateurs et en groupes de gènes co-régulés. Nous procédons en deux phases pour y parvenir. La première phase consiste à construire un réseau de régulation entre les facteurs de transcription. La seconde phase consiste en la classification des facteurs de transcription selon les liens de régulation établis par ce réseau. D'un point de vue statistique, les facteurs de transcription sont les variables et les données d'expression sont les observations. Nous représentons le réseau à inférer par un graphe orienté dont les nœuds sont les variables. L'estimation de ses arêtes est vue comme un problème de sélection de variables en grande dimension avec un faible nombre d'unités statistiques. Nous traitons ce problème à l'aide de régressions linéaires pénalisées de type LASSO. Une approche préliminaire qui consiste à sélectionner un ensemble de variables du chemin de régularisation par le biais de critères de vraisemblance pénalisée s'avère être instable et fournit trop de variables explicatives. Pour contrecarrer cela, nous proposons et mettons en compétition deux procédures de sélection, adaptées au problème de la haute dimension et mêlant régression linéaire pénalisée et rééchantillonnage. L'estimation des différents paramètres de ces procédures a été effectuée dans le but d'obtenir des ensembles de variables stables. Nous évaluons la stabilité des résultats à l'aide de jeux de données simulés selon notre modèle graphique. Nous faisons appel ensuite à une méthode de classification non supervisée sur chacun des graphes orientés obtenus pour former des groupes de nœuds vus comme contrôleurs et des groupes de nœuds vus comme contrôlés. Pour évaluer la proximité entre les classifications doubles des nœuds obtenus sur différents graphes, nous avons développé un indice de comparaison de couples de partition dont nous éprouvons et promouvons la pertinence. D'un point de vue pratique, nous proposons une méthode de simulation en cascade, exigée par la complexité de notre modèle et inspirée du bootstrap paramétrique, pour simuler des jeux de données en accord avec notre modèle. Nous avons validé notre modèle en évaluant la proximité des classifications obtenues par application de la procédure statistique sur les données réelles et sur ces données simulées. / This thesis deals with the characterisation of key genes in gene expression regulation, called transcription factors, in the plant Arabidopsis thaliana. Using expression data, our biological goal is to cluster transcription factors in groups of co-regulator transcription factors, and in groups of co-regulated transcription factors. To do so, we propose a two-step procedure. First, we infer the network of regulation between transcription factors. Second, we cluster transcription factors based on their connexion patterns to other transcriptions factors.From a statistical point of view, the transcription factors are the variables and the samples are the observations. The regulatory network between the transcription factors is modelled using a directed graph, where variables are nodes. The estimation of the nodes can be interpreted as a problem of variables selection. To infer the network, we perform LASSO type penalised linear regression. A preliminary approach selects a set of variable along the regularisation path using penalised likelihood criterion. However, this approach is unstable and leads to select too many variables. To overcome this difficulty, we propose to put in competition two selection procedures, designed to deal with high dimension data and mixing linear penalised regression and subsampling. Parameters estimation of the two procedures are designed to lead to select stable set of variables. Stability of results is evaluated on simulated data under a graphical model. Subsequently, we use an unsupervised clustering method on each inferred oriented graph to detect groups of co-regulators and groups of co-regulated. To evaluate the proximity between the two classifications, we have developed an index of comparaison of pairs of partitions whose relevance is tested and promoted. From a practical point of view, we propose a cascade simulation method required to respect the model complexity and inspired from parametric bootstrap, to simulate data under our model. We have validated our model by inspecting the proximity between the two classifications on simulated and real data.

Mapping of the human genome has been instrumental in understanding diseasescaused by changes in single genes. However, disease mechanisms involvingmultiple genes have proven to be much more elusive. Their complexityemerges from interactions of intracellular molecules and makes them immuneto the traditional reductionist approach. Only by modelling this complexinteraction pattern using networks is it possible to understand the emergentproperties that give rise to diseases.The overarching term used to describe both physical and indirect interactionsinvolved in the same functions is functional association. FunCoup is oneof the most comprehensive networks of functional association. It uses a naïveBayesian approach to integrate high-throughput experimental evidence of intracellularinteractions in humans and multiple model organisms. In the firstupdate, both the coverage and the quality of the interactions, were increasedand a feature for comparing interactions across species was added. The latestupdate involved a complete overhaul of all data sources, including a refinementof the training data and addition of new class and sources of interactionsas well as six new species.Disease-specific changes in genes can be identified using high-throughputgenome-wide studies of patients and healthy individuals. To understand theunderlying mechanisms that produce these changes, they can be mapped tocollections of genes with known functions, such as pathways. BinoX wasdeveloped to map altered genes to pathways using the topology of FunCoup.This approach combined with a new random model for comparison enables BinoXto outperform traditional gene-overlap-based methods and other networkbasedtechniques.Results from high-throughput experiments are challenged by noise and biases,resulting in many false positives. Statistical attempts to correct for thesechallenges have led to a reduction in coverage. Both limitations can be remediedusing prioritisation tools such as MaxLink, which ranks genes using guiltby association in the context of a functional association network. MaxLink’salgorithm was generalised to work with any disease phenotype and its statisticalfoundation was strengthened. MaxLink’s predictions were validatedexperimentally using FRET.The availability of prioritisation tools without an appropriate way to comparethem makes it difficult to select the correct tool for a problem domain.A benchmark to assess performance of prioritisation tools in terms of theirability to generalise to new data was developed. FunCoup was used for prioritisationwhile testing was done using cross-validation of terms derived fromGene Ontology. This resulted in a robust and unbiased benchmark for evaluationof current and future prioritisation tools. Surprisingly, previously superiortools based on global network structure were shown to be inferior to a localnetwork-based tool when performance was analysed on the most relevant partof the output, i.e. the top ranked genes.This thesis demonstrates how a network that models the intricate biologyof the cell can contribute with valuable insights for researchers that study diseaseswith complex genetic origins. The developed tools will help the researchcommunity to understand the underlying causes of such diseases and discovernew treatment targets. The robust way to benchmark such tools will help researchersto select the proper tool for their problem domain. / <p>At the time of the doctoral defense, the following papers were unpublished and had a status as follows: Paper 5: Manuscript. Paper 6: Manuscript.</p>