Les fonctions biologiques sont le résultat de l'interaction de beaucoup de processus, avec différents objectifs, complexités, niveaux de hiérarchie, et changements de conditions que modifient le comportement de systèmes. Nous utilisons des équations différentielles ou dynamiques plus générales, et systèmes stochastiques de transition pour décrire la dynamique de changements des modèles. La composition, réconciliation et réutilisation des modèles nous permettent d'obtenir des descriptions de systèmes biologiques complètes et compatibles et leur combiner. Notre spécification de systèmes hybrides avec BioRica assure l'intégrité de modèles, et implémente notre approche. Nous appliquons notre approche pour décrire in-silico deux systèmes: la dynamique de la fermentation du vin, et des décisions cellulaires associées à la formation de tissu d'os. / The main aim of this thesis is to develop an approach that allows us to describe biological systems with theoretical sustenance and good results in practice. Biological functions are the result of the interaction of many processes, that connect different hierarchy levels going from macroscopic to microscopic level. Each process works in different way, with its own goal, complexity and hierarchy level. In addition, it is common to observe that changes in the conditions, such as nutrients or environment, modify the behavior of the systems. So, to describe the behavior of a biological system over time, it is convenient to combine different types of models: continuous models for gradual changes, discrete models for instantaneous changes, deterministic models for completely predictable behaviors, and stochastic or non- deterministic models to describe behaviors with imprecise or incomplete information. In this thesis we use the theory of Composition and Hybrid Systems as basis, and the BioRica framework as tool to model biological systems and analyze their emergent properties in silico.With respect to Hybrid Systems, we considered continuous models given by sets of differential equations or more general dynamics. We used Stochastic Transition Systems to describe the dynamics of model changes, allowing cofficient switches that control the parameters of the continuous model, and strong switches that choose different models. Composition, reconciliation and reusing of models allow us to build complete and consistent descriptions of complex biological systems by combining them. Compositions of hybrid systems are hybrid systems, and the refinement of a model forming part of a composed system results in a refinement of the composed system. To implement our approach ideas we complemented the theory of our approach with the improving of the BioRica framework. We contributed to do that giving a BioRica specification of Hybrid Systems that assures integrity of models, allowing composition, reconciliation, and reuse of models with SBML specification.We applied our approach to describe two systems: wine fermentation kinetics, and cell fate decisions leading to bone and fat formation. In the case of wine fermentation, we reused known models that describe the responses of yeasts cells to different temperatures, quantities of resources and toxins, and we reconciled these models choosing the model with best adjustment to experimental data depending on the initial conditions and fermentation variable. The resulting model can be applied to avoid process problems as stuck and sluggish fermentations. With respect to cell fate decisions the idea is very ambitious. By using accurate models to predict the bone and fat formation in response to activation of pathways such as the Wnt pathway, and changes of conditions affecting these functions such as increments in Homocysteine, one can analyze the responses to treatments for osteoporosis and other bone mass disorders. We think that here we are giving a first step to obtain in silico evaluations of medical treatments before testing them in vitro or in vivo.

L’apprentissage par renforcement profond (RL) a connu d’énormes progrès ces dernières années, mais il est encore difficile d’appliquer le RL aux problèmes de prise de décision du monde réel. Cette thèse identifie trois défis clés avec la façon dont nous faisons la recherche RL elle-même qui entravent les progrès de la recherche RL. — Évaluation et comparaison peu fiables des algorithmes RL ; les méthodes d’évaluation actuelles conduisent souvent à des résultats peu fiables. — Manque d’informations préalables dans la recherche RL ; Les algorithmes RL sont souvent formés à partir de zéro, ce qui peut nécessiter de grandes quantités de données ou de ressources informatiques. — Manque de compréhension de la façon dont les réseaux de neurones profonds interagissent avec RL, ce qui rend difficile le développement de méthodes évolutives de RL. Pour relever ces défis susmentionnés, cette thèse apporte les contributions suivantes : — Une méthodologie plus rigoureuse pour évaluer les algorithmes RL. — Un flux de travail de recherche alternatif qui se concentre sur la réutilisation des progrès existants sur une tâche. — Identification d’un phénomène de perte de capacité implicite avec un entraînement RL hors ligne prolongé. Dans l’ensemble, cette thèse remet en question le statu quo dans le RL profond et montre comment cela peut conduire à des algorithmes de RL plus efficaces, fiables et mieux applicables dans le monde réel. / Deep reinforcement learning (RL) has seen tremendous progress in recent years, but it is still difficult to apply RL to real-world decision-making problems. This thesis identifies three key challenges with how we do RL research itself that hinder the progress of RL research. — Unreliable evaluation and comparison of RL algorithms; current evaluation methods often lead to unreliable results. — Lack of prior information in RL research; RL algorithms are often trained from scratch, which can require large amounts of data or computational resources. — Lack of understanding of how deep neural networks interact with RL, making it hard to develop scalable RL methods. To tackle these aforementioned challenges, this thesis makes the following contributions: — A more rigorous methodology for evaluating RL algorithms. — An alternative research workflow that focuses on reusing existing progress on a task. — Identifying an implicit capacity loss phenomenon with prolonged offline RL training. Overall, this thesis challenges the status quo in deep reinforcement learning and shows that doing so can make RL more efficient, reliable and improve its real-world applicability