Journées annuelles du groupe de travail (3ème édition)

Informations générales

Date : Lundi 13 mars (toute la journée) et mardi 14 mars (matin) 2017

Lieu : Amphitheatre de l'espace colloque du CNRS, 1919 route de Mende à Montpellier.  

Organisateurs : Ovidiu Radulescu, Grégory Batt, Cédric Lhoussaine, Elisabeth Remy et Anne Siegel 

La troisième édition des journées annuelles du GT Bioss va se dérouler juste avant les journées nationales du GDR Informatique-Mathématique. Ainsi, les 13 et 14 mars 2017, les membres du GT auront le plaisir de se rencontrer autour de conférences autour des thèmes suivants :

- la modélisation des systèmes biologiques et ses applications;
- les systèmes dynamiques discrets, hybrides;
- les langages de modélisation et leurs sémantiques (déterministes, non-déterministes, stochastiques);
- la vérification de modèles;
- la réduction de modèles et leur pouvoir prédictif (sous incertitude);
- l'inférence d’interactions et de règles à partir de données biologiques;
- et plus généralement tout problème de modélisation lié à l’intégration de données réelles. 

Inscription

L'inscription, gratuite mais obligatoire, se fait en remplissant le formulaire accessible ici.

Orateurs invités

Anaïs Baudot, Institut de Mathématiques de Marseille.
Jakob Ruess, INRIA Saclay.
Thomas Sturm, CNRS-LORIA, Nancy & Max Planck Institute, Saarbrücken, Germany.

Programme

Lundi 13 Mars

09h00 - 09h25 - Accueil - Café
Chairwoman: Anne Siegel
09h25 - 09h30 - Ovidiu Radulescu - Introduction.
09h30 - 10h15 - Conférencier invité - Jakob Ruess - Control of bio-digital systems in single cells.
10h20 - 10h35 - Ovidiu Radulescu - Time dependent multivariate distributions for piecewise-deterministic models of gene networks. slides
10h40 - 10h55 - Stefano Casagranda - Principal Process Analysis and reduction of biological models with order of magnitude.
slides 11h00 - 11h30 - Pause
Chairman: Cédric Lhoussaine
11h30 - 11h45 - Yves-Stan Le Cornec - Le Projet Kami.
11h50 - 12h05 - Andreea Beica - Synchronous balanced analysis. slidesi
12h10 - 12h25 - François Fages - Complexité algorithmique des calculs analogiques et compilation de fonctions mathématiques en réactions biochimiques élémentaires.

12h30- 13h45 - Pause déjeuner

Chairman: Ovidiu Radulescu
13h45 - 14h30 - Conférencier invité - Thomas Sturm - Symbolic Methods in Bifurcation Analysis. slides
14h35 - 14h50 - Matthieu Pichene - Predicting tumor growth using a statistical layered population abstraction. slides
14h55 - 15h10 - François Boulier - Identifiabilité, équations intégro-différentielles et neurobiologie. slides
15h15 - 15h40 - Pause
Chairman: François Fages
15h40 - 15h55 - Clémence Frioux - Hybrid gap-filling to reconcile qualitative and quantitative abstractions of metabolism.
16h00 - 16h15 - Emilie Allart - Elementary modes refine abstract interpretation of reaction networks with partial kinetic information.
16h20 - 16h35; Marie Beurton-Aimar - How to display patterns inside elementary flux modes.
16h40 - 17h00 - Pause
Chairman: Grégory Batt
17h00 - 17h15 - Florian Bridoux - On The Cost Of Simulating A Parallel Boolean Automata Networks By A Sequential One.
17h20 - 17h35 - Aurélien Naldi - Reversed logical models for the study of basins of attraction.
17h40 - 17h55 - Pierre Siegel - Des logiques non-monotones aux systèmes dynamiques discrets (SDD).
18h00 - 18h15 - Laurent Trilling - Apport de la non monotonie pour la modélisation logique de réseuax de régulation génique.

19h15 - Repas au Restaurant Trinque Fougasse O'Nord 1581 route de Mende (à 350m à pied du lieu de la conférence), formule 30€, vin et concert jazz compris. Il est encore possible de s’inscrire au repas!

Mardi 14 Mars

Chairwoman: Élisabeth Rémy
09h00 - 09h45: Conférencière invitée - Anais Baudot - Mining and modeling biological networks to study rare and common human diseases.
09h50 - 10h10: Pause
10h10 - 10h25: Jean Coquet - Analysis of TGF-β signaling networks to find different families of trajectories.
10h30 - 10h45: Arnaud Poret - Linking Cancer Models with Therapeutic Effects.
10h50 - 11h05: Amos Korman - Confidence sharing: an economic strategy for efficient information flows in animal groups.
11h10 - 11h30: Pause
Chairman: Laurent Trilling
11h30 - 11h45: Ofer Feinerman - Algorithmic challenges in ant cooperative transport.
11h50 - 12h05: Jonathan Behaegel - Réseaux génétiques hybrides: de la logique de Hoare à l'identification de paramètres.
12h10 - 12h25: Celia Biane - Inférence d'action sur les réseaux pour la reprogrammation cellulaire.

Résumés

Jakob Ruess (INRIA Saclay) - Control of bio-digital systems in single cells.
Akin to the developments in industry, also in scientific wet labs more and more manual tasks are being automated and many experiments are nowadays carried out by computer controlled robots. Such experiments are, however, still almost exclusively pre-planned by the scientist as a series of exact instructions that are to be carried out by the robot. Industry, on the other hand, has seen developments such as self-driving cars where the user specifies a goal and the system decides for itself how to best attain that goal. In this talk, I will present our recent efforts to construct autonomous bio-digital systems that are based on real-time computer-to-single cell communication. In particular, I will present an experimental/technological platform in which cells, equipped with an optogenetically inducible promoter, signal to a computer through microscopy. The computer processes the incoming data in real time and sends back light signals to the cells. Experiments on this platform can therefore be specified as algorithms that map the list of received signals from all the cells into a list of optogenetic stimulations. To demonstrate the functionality of this setup, I will show results obtained with two different algorithms implemented on the platform. First, we used a model-predictive controller to drive the expression of a gene coding for a fluorescent reporter protein and managed to ensure that all cells follow the same gene expression target profile despite significant heterogeneity of the cells and even in the presence of environmental perturbations. Subsequently, we used the same controller to make all cells follow different target profiles, demonstrating that the platform can be used to structure populations into pre-definable dynamic gene expression phenotypes. Finally, I will provide results where we coupled an algorithm mimicking a negative feedback loop and a digital cell-to-cell communication mechanism to the gene in order to produce synchronized fluorescence oscillations in the cells.

Ovidiu Radulescu (University of Montpellier, DIMNP lab) - Time dependent multivariate distributions for piecewise-deterministic models of gene networks.
We discuss piecewise-deterministic approximations of gene networks dynamics. These approximations capture in a simple way the stochasticity of gene expression and the propagation of expression noise in networks and circuits. By using partial omega expansions, piecewise deterministic approximations can be formally derived from the more commonly used Markov pure jump processes. We are interested in time dependent multivariate distributions that describe the stochastic dynamics of the gene networks. This problem is difficult even in the simplified framework of piecewise deterministic processes. We consider three methods to compute these distributions: the direct Monte Carlo simulation, the numerical integration of the Liouville-master equation and the push-forward method. We present applications of this approach to stochastic biological switches and logical gates that can be used as memories and computing elements in synthetic biocomputing machines.

Stefano Casagranda (INRIA Sophia-Antipolis, BICOLORE group) - Principal Process Analysis and reduction of biological models with order of magnitude.
We present a simple method that allows to analyze the biological processes of a dynamical model and classify them. Along the system trajectories, we decompose the model into biological meaningful processes and then study their activity or inactivity during the time evolution of the system. The structure of the model is then reduced to the core mechanisms involving only the active processes. The initial conditions are supposed to lie in some rectangle, that could represent one order of magnitude for the variables. Keeping only the active processes, we obtain the principal processes in the rectangle and then in the adjacent rectangles where the trajectories may have a transition. Finally we obtain a partition of the space with a reduced model within each rectangle. We apply these techniques to a classical model of gene expression with protein and messenger RNA.

Yves-Stan Le Cornec (ENS Lyon) Le Projet Kami.
Rule-based modelling enables one to describe complex biological systems by defining and combining together low level mechanisms. However, the details of these mechanisms and the exact conditions for each interaction are not always well known and often spread apart across multiple articles which impedes the process of building a model. KAMI is a software application as well as a graph-based formalism which aims to aggregate biological knowledge in an incremental and flexible way as well as ensuring its consistency. In KAMI, basic pieces of knowledge are represented as graphs called nuggets which are ideally all "typed" by another graph (the action graph). This has the effect of organizing and combining the knowledge contained inside the nuggets as well as providing a graphical overview of it. This action graph is itself typed by a meta-model which provides syntactic restrictions and structure to the model. This in particular facilitates the use of well defined graph transformations when needed, for instance to merge together multiple knowledge bases which can use different formats, to communicate with other existing tools, or to make the meta-model itself evolve if needed.
In practice, we developed a generic graph rewriting library in python (Regraph) and a web interface (RegraphGui) that are used to build and manipulate the biological knowledge base. Rule-based models written in kappa can then be generate and sent to the KaSim simulator.

Andrea Beica (ENS, Paris) - Synchronous balanced analysis.
When modeling Chemical Reaction Networks, a commonly used mathematical formalism is that of Petri Nets, with the usual inter-leaving execution semantics. We aim to substitute to a Chemical Reaction Network, especially a growth one (i.e., for which an exponential stationary phase exists), a piecewise synchronous approximation of the dynamics: a resource-allocation-centered Petri Net with maximal-step execution semantics. In the case of unimolecular chemical reactions, we prove the correctness of our method and show that it can be used either as an approximation of the dynamics, or as a method of constraining the reaction rate constants (an alternative to flux balance analysis, using an emergent formally de ned notion of growth rate as the objective function), or a technique of refuting models.

François Fages (Inria Saclay Ile-de-France, Lifeware group) - Complexité algorithmique des calculs analogiques et compilation de fonctions mathématiques en réactions biochimiques élémentaires.
L’aspect continu de beaucoup d’interactions protéiques nous conduit à considérer des modes de calcul mixtes analogiques-digitaux, pour lesquels des résultats récents de Bournez, Graca et Pouly en théorie de la calculabilité et de la complexité analogiques établissent des liens fondamentaux avec la programmation classique. Nous dérivons de ces résultats un compilateur de spécifications comportementales en systèmes de réactions biochimiques que l’on peut comparer aux circuits naturels résultats de l’évolution. Nous illustrons cette démarche par l’exemple du module de signalisation MAPK qui a une fonction de convertisseur analogique-digital dans la cellule, et par le contrôle du cycle cellulaire.

Thomas Sturm (CNRS-LORIA, Nancy & Max Planck Institute, Saarbrücken, Germany) - Symbolic Methods in Bifurcation Analysis.
We are going to discuss the potential of symbolic computation methods with the analysis of biological networks for bifurcations. Those methods include various real quantifier elimination methods in combination with preprocessing heuristics, real triangularizations, and subtropical approaches. Applications range from the identification of Hopf bifurcations in comparatively small models to work-in-progress towards the identification of saddelnode bifurcations with respect to more than one parameter in a model of MAPK from the BioModels database. After several years of systematic work and collaborations in the area we feel confident that there is a potential for symbolic tools in biological research, in particular with parametric situations, although there are obvious obstacles: Firstly, our approaches are not fully automatic yet, so that users have to understand their theoretical background to some extent. Secondly, for complexity reasons the methods cannot be expected to scale to all relevant problem sizes. It is thus important to identify application areas that are sufficiently large to justify the effort for both developers and potential users. We hope to stimulate critical but open-minded constructive discussions around those topics.

Matthieu Pichene (Inria Rennes, SUMO group) - Predicting tumor growth using a statistical layered population abstraction.
Tumor growth models such as B. Waclaw et al. (2015) can be a useful tool to design protocols for cancer treatment. A challenging problem is that models (usually cell based) can rapidly be time and memory consuming due to tumors reaching hundreds of thousands of cells. These cells can each duplicate, move and die. We designed an abstraction that represent the tumor evolution. Tumor is partitioned into layers. In each layer, few data are kept, e.g. the density of the layer, but not the individual cell data. From statistics drawn from agent based models such as TumourSimulator B. Waclaw et al. (2015) for small tumors, we automatically obtain a statistical model that can reproduce the tumor evolution, and predict the evolution of bigger tumor under treatment. Compared to the original model, we can reproduce very faithfully the evolution of the tumor. Also, we have clear speed up (e.g. 30 times for a tumor of 100.000 cells).

François Boulier (University of Lille, CRIStAL lab) - Identifiabilité, équations intégro-différentielles et neurobiologie.
L'identifiabilité est une étude théorique d'un modèle mathématique destinée à établir si les paramètres du modèle ont une valeur complètement déterminée par les mesures expérimentales. C'est une étude de principe. Dans le cas d'un modèle présenté sous la forme d'un système d'équations différentielles paramétriques non linéaires, cette propriété peut être établie, après une étape de calcul formel, par l'analyse du « polynôme entrée-sortie » du modèle. Le résultat de l'étude peut ensuite être réemployé pour effectuer une estimation de paramètres à partir de données expérimentales réelles. Des travaux récents en calcul formel et en théorie du contrôle ouvrent de nouvelles perspectives théoriques et pratiques visant à généraliser cette approche en faisant apparaître des équations intégro-différentielles. Dans cet exposé, nous exposerons ces idées et les illustrerons sur une problématique issue des neurosciences, où la question posée consiste à clarifier le rôle des astrocytes pathologiques dans le développement de la dépression corticale propagée.
Il s'agit de travaux en cours, menés en collaboration entre une équipe de calcul formel (Univ. Lille/CRIStAL), une équipe de modélisation en mathématiques appliquées (Univ. Le Havre/LMAH) et une équipe de neurobiologie (Rouen/INSERM).

Clémence Frioux (Inria Rennes, Dyliss group) - Hybrid gap-filling to reconcile qualitative and quantitative abstractions of metabolism.
As the amount of genomics data keeps on increasing, so is the need to develop efficient techniques to model the metabolism. An important part of the modelling is the gap-filling step during the reconstruction of metabolic networks. It aims to complete a draft metabolic network based on experimental evidence by selecting adequate reactions from a database to restore the observed metabolic behaviour. Most of the existing techniques to achieve this procedure are quantitative and use linear programming to model flux distribution in the networks. They need accurate data and some linear problems are not solvable without a relaxation of constraints due to the space of solutions. We thus developed lately a topological method called meneco that can perform gap-filling when few data is available. Based on Answer Set Programming (ASP), it takes advantage of efficient solvers to easily sample the space of solutions, but the outputs are often not flux-balanced. Thanks to the progress in the development of ASP, it is now possible to plug a linear programming solver to the ASP one. Here we propose a hybrid method to take the most out of the two modellings and use the efficiency of the topological solving to reach quantitative-compliant solutions for gap-filling. This hybrid solving enables to take into account new constraints and hypotheses about the dynamic system (initial state, steady state, inputs/outputs) and can be relevant to solve new metabolic networks modelling problems.

Emilie Allart (University of Lille, CRIStAL lab) - Elementary modes refine abstract interpretation of reaction networks with partial kinetic information.
Genetic engineering is nowadays widely used to change the genome of cells in order to modify their behaviour, for example to overproduce some metabolite of interest. Gene knockout is one common genetic engineering technique which consists in removing one or more genes from the genome. Given the combinatorial number of possible genetic changes and the consequent impossibility of testing all of them by wet lab experiments, in silico prediction of the most interesting genetic modifications is often desirable. However, the necessary information for the in silico modelling of the organism of interest is usually lacking. This is the case for example for the bacterium B. Subtilis: its genetic engineering for the overproduction of surfactin (a well-known antibiotic [1]) is of high interest in agriculture among other fields, but the lack of information about the biochemical functioning of this organism prevented us from modeling it with the existing methods. In order to overcome this problem, we developed a modeling language to represent reaction networks with partial kinetic information, and a method based on abstract interpretation that allows us to analyse models written in this language even in the absence of quantitative information [2, 3]: from the steady state semantics of the reaction network, we transform arithmetic constraints into abstract constraints over a finite domain where unknowns have been abstracted away. The qualitative behaviour of the system can therefore be evaluated in silico by means of a constraint solver, which gives us the set of all solutions (corresponding to combinations of feeding changes and gene knockouts) that lead to the desired behaviour. However, abstract interpretation usually over-approximates the solution space. As an example, while a finite system of linear equations implies all the equations that can be expressed as a linear combination, this implication no longer holds once the equations have been translated into abstract constraints. My current work consists in optimally generating new arithmetic constraints from the existing in order to minimize the solution space. Here I will discuss in particular our interest in elementary flux modes [4] and how to adapt classical elementary mode analysis to the generation of abstract constraints to improve the qualitative analysis of reaction networks with partial kinetic information.

Marie Beurton-Aimar (University of Bordeaux, Labri lab) - How to display patterns inside elementary flux modes.
Elementary flux modes computing is a powerful tool to identify feasible routes trough a metabolic network(Schuster et al. Nat. Biotechnol., 18 (3), 2000). According to their definitions, the list of all elementary flux modes provides a way to analyze the behaviors of the network both in its current functioning and under perturbations. But the main problem with this tool remains the size of this list. Currently a metabolic network with approximatively 50 reactions and metabolites can generate more than several thousands of elementary flux modes. To be useful, new tools to automatically analyze the results are required. In a first attempt, we have used the computing of Minimal Cut Sets (Gagneur J., Klamt S. BMC Bioinformatics, 2004) of all elementary flux modes to identify patterns, i.e. list of common reactions, in the set of elementary flux modes . To visualize these patterns which can be see as a tree of sub-patterns we have chosen a technique coming from the domain of large data set visualization, the parallel coordinates displaying. This technique presents data as flux through variables put on the x axis. Values of the flux is on the y axis. In our case, reactions are put on the x axis and elementary flux modes are data lines or edges between reactions. The reaction stoichiometry values are set on the y axis: 1 if the reaction is present in forward direction in this elementary flux mode, 2 if it is in backward reaction and 0 if the reaction is absent. The flux size between two reactions informs about the quantity of elementary flux modes which use the same association between two reactions. The figure below shows a result example of patterns sharing by the elementary flux modes in the plant cell metabolism (Beurton-Aimar et al. Plant Metabolic Flux Analysis, 2013). The displaying tool CoPHI is accessible on this website http://www.labri.fr/perso/jsansen/. The display is dynamique, users can change the order of the reactions on x axis and so explore by themselves the patterns that can be found. Information given by MCS computing is used to sort reactions by taking into account set of reactions. Future works will be done to develop a CoPHI version dedicated to metabolic network analysis.

Florian Bridoux (University of Marseille, LIF lab) - On The Cost Of Simulating A Parallel Boolean Automata Networks By A Sequential One.
In this presentation, we study Boolean automata networks (BANs). A given BAN can be associated with several dynamics, depending on the schedule (i.e. the order) we choose to update its automata. In this presentation, we consider all block-sequential update schedules: we group automata into blocks, and we update all automata of a block at once, and iterate the blocks sequentially. For the last 15 years, people have studied the influence of the update schedules on the dynamics of a BAN. Here, we do the opposite. We want to determine the minimum number κ of additional automata that a BAN associated with a given block-sequential update schedule needs to simulate a given BAN with a parallel update schedule. To solve this problem, we introduce a graph that we call confusion graph built from the BAN and the update schedule. We show the relation between κ and the chromatic number of the confusion graph. Thanks to this confusion graph, we bound κ in the worst case between n/2 and 2n/3 + 2 (n being the size of the BAN simulated) and we conjecture that this number equals n/2. We support this conjecture with two results: the clique number of a confusion graph is always less than or equal to n/2 and, for the subclass of bijective BANs, κ is always less than or equal to n/2.

Aurélien Naldi (University of Montpellier, DIMNP lab) - Reversed logical models for the study of basins of attraction.
Boolean, and more generally logical, models are widely used for the study of biological systems, especially differentiation systems. The formal identification of their attractors has been widely studied, however the corresponding basins of attraction received less attention.
In this purpose, we propose a method to construct a "reversed" model regarding the asynchronous dynamical behaviour: the successors of the states of a reversed model correspond to their predecessors in the original one. While the reversal can be generalized only to a particular class of multivalued models, we show how to use model booleanization to study the reversed asynchronous dynamics of any model.
The study of the reversed dynamics then facilitates the identification of the basins of attraction, assuming that we already know the attractors. We can further compute "strong" and "weak" basins (from which we can respectively reach a unique or multiple attractors), as well as their "frontiers", i.e., the sets of states such that the reachable attractors are different from the attractors reachable from neighbouring states (predecessors or successors).
This approached is illustrated on a published model of the cell fate decision in response to death-receptor engagement.

Pierre Siegel (University of Marseille, LIF lab) - Des logiques non-monotones aux systèmes dynamiques discrets (SDD).
Du point de vue logique et représentation des connaissances, un système biologique peut être considéré comme un ensemble de variables qui interagissent. Dans ce cadre la cellule pose des problèmes intéressants à l’Intelligence Artificielle. Il faut d’abord formaliser les interactions mais une formalisation en logique « classique » est difficile et donne des incohérences. Ensuite ce que l’on sait vient en bonne partie d’expériences. On ne connaît donc qu’une petite partie des interactions et cette connaissance peut être révisable, incertaine, contradictoire et même fausse. On veut aussi essayer de compléter in-silico les interactions. Enfin la complexité algorithme est importante. Ces problèmes sont étudiés depuis longtemps en IA en utilisant des logiques non-monotones et des techniques de programmation par contraintes. Dans le cas particulier de la cellule des résultats ont été obtenus en utilisant la logique des défauts [1].
D’un autre côté, les systèmes biologiques peuvent être regardés dans le contexte des réseaux d’automates et des SDD. Des théorèmes importants portent sur les cycles d’interactions dont l’étude est fondamentale. Mais une représentation des SDD par la logique des défauts (et par la plupart des logiques non-monotones) n’est pas adaptée ; par exemple l’équivalent d’un circuit négatif n’a pas d’extension (de solution, de point fixes…) Cette absence d’extension pour les logiques des défauts a été étudiée [2,3] et cette étude a donné la Logique des Hypothèses. Pour cette logique on a toujours des extensions mais certaines d’entre elles, les extensions fantômes, bien caractérisées sont spéciales (et leur utilité n’était pas claire).
Nous étudions une représentation des SDD par la Logique des Hypothèses. Un but est de permettre de discriminer les états stables, les cycles stables et les cycles instables. Les extensions fantômes semblent permettre de le faire. Un autre but est de donner des algorithmes efficaces pour calculer les états et cycles. Quelques premiers résultats seront présentés.

Laurent Trilling (University of Grenoble, TIMC-IMAG lab) - Apport de la non monotonie pour la modélisation logique de réseuax de régulation génique.
Nous nous intéresserons à la modélisation déclarative des réseaux logiques de régulation génique proposés par René Thomas, à l'aide de la technologie de programmation logique ASP (Answer Set Programming). Cette technologie est basée sur une logique non monotone n'autorisant que certains types de modèles logiques, dits stables: intuitivement, ne sont vrais que les atomes qui sont prouvables grâce aux axiomes. Un modèle stable est minimal en ce sens que la suppression d'un atome du modèle ne peut résulter en un modèle. La technologie ASP connait un certain engouement à l'heure actuelle, assurément dû au pouvoir d'expression des langages proposés et aussi aux performances des solveurs associés. Mais l'intérêt de cette propriété de non monotonie et son impact sur la modélisation d'applications en termes de méthodologie et d'optimisation ne sont pas toujours soulignés précisément. Nous tâcherons de le faire dans le cadre de la modélisation des réseaux de Thomas.
Par approche déclarative, nous entendons : représentation de toutes les données biologiques disponibles (sur la structure ou la dynamique), sous forme d'axiomes logiques(contraintes) et obtention sous forme intensionelle (implicite) des réseaux de Tomas cohérents avec ces données en cas de satisfaisabilité. Dans le cas contraire, une procédure automatique de réparation justifiée doit être applicable. L'approche fournit, entre autres, une faculté d'apprentissage appréciable : celle de tous les théorèmes, restreints à une formulation fixée (par exemple les clauses d'une taille limitée sur un ensemble déterminé d'atomes), déductibles de ces modèles.
Deux avantages généraux d'ASP sont reconnus : un traitement des données incomplètes radical (les atomes non prouvables sont réputés faux) et un pouvoir de déduction augmenté (les modèles stables forment un sous-ensemble des modèles). Nous aborderons plutôt trois aspects précis, concernant la modélisation étudiée : 1) l'utilisation de défauts pour spécifier la réparation d'inconsistance en soulignant l'intérêt d'exprimer la minimisation sous-jacente en termes logiques (et non pas algorithmiques), 2) l'emploi (et la méthodologie de construction) d'une conjonction de défauts originale pour exprimer la notion de compositions d'interactions géniques seulement généralement vraies (sauf à être prouvée effectivement fausses), 3) la mise en œuvre de formules CTL générales, en particulier du type AF (indispensables pour exprimer des propriétés exhaustives sur les comportements), en prenant appui sur la minimalité des modèles stables.

Anaïs Baudot (Institut de Mathématiques de Marseille) - Mining and modeling biological networks to study rare and common human diseases.
Networks are scaling-up the analysis of gene and protein functions, hence offering new avenues to study the diseases in which these genes and proteins are involved. In this context, I will talk about the exploration of interactome networks containing thousands of physical and functional interactions between proteins. We develop partitioning algorithms to recover communities – or functional modules – from these large-scale networks, and use them to study the cellular functions of proteins of interest. Recently, we have extended the community detection to multiplex networks, i.e., networks containing different layers of different interaction categories, such as protein-protein interaction or gene co-expression. I will show how these approaches can be used to study the relationships between different common diseases, in particular their inverse comorbidities. The last part of my talk will be dedicated to our recent work on random walks with restart on multiplexes, in order to study rare monogenic diseases. I will finally briefly mention logical modeling to decipher drug synergies in cancer, and how mining large-scale static networks and modeling smaller-scale dynamical network could unite around expression data integration.

Jean Coquet (Inria Rennes, Dyliss group) - Analysis of TGF-β signaling networks to find different families of trajectories.
The Transforming Growth Factor TGF-β is a multifunctional cytokine that regulates mammalian development, differentiation, and homeostasis (Ikushima and Miyazono, 2011). As a growth inhibitor of epithelial, endothelial, and hematopoietic cells, TGF-β is a potent anticancer agent in normal tissue. At the opposite TGF-β acts as a promoter of tumor by inducing the hallmarks of the cancer. The context-dependent pleiotropic nature of TGF-β is associated with complex signaling pathways. Our group recently developed a model for TGF-β- dependent signalling based on the integration of the 137 signaling pathways from Pathway Interaction Database (Andrieux et al, 2014). Based on guarded transition formalism, we identified 15,934 chains of reactions (trajectories) linking TGF-β to at least one of 159 target genes. The combined size and complexity of this model place it beyond current understanding. Its analysis require automated tools for identifying general patterns. In this study, we focused on designing a reasoning method to characterize the composition of these trajectories. We used an exhaustive and without prior assumptions approach to explore them. First, we grouped the trajectories into several families using the Relevant Set Correlation model (Houle, 2008). Then, we extracted the over-represented molecules for each families. Finally, we identified pairs of over-represented molecules included in the same trajectories but not described in pathway databases. The main results of the study were the identification of two trajectory families representing different TGF-β-dependent signaling pathways, and the prediction of hundreds of molecule pairs occurring in trajectories and never reported in signaling databases.

Arnaud Poret (Ecole Centrale de Nantes, LS2N lab) - Linking Cancer Models with Therapeutic Effects.
The goal is to model the signaling pathways involved in the drug response of cancerous cells to find how to counteract resistance. We have the gene expression profile (GEP) of several breast cancer and multiple myeloma cells treated with various drugs. Along with the GEPs, we also have the phenotypic drug response of the cells in term of survival percentage. Building models of the signaling pathways involved in drug response/ resistance could allow us to predict molecular interventions able to counteract resistance, and then decrease the survival of resistant cells. For reconstructing these signaling pathways from the GEPs, we will use existing tools such as PID2SIF2Graph and MCWalk, operating on public databases such as KEGG, Reactome, Wikipathways and NCI-PID. For modeling the reconstructed networks, we envision two approaches: logical programming, namely answer set programming, and Boolean networks. The obtained models will be used to assess several molecular interventions in their ability at decreasing the predicted survival percentages.

Amos Korman (University Paris Diderot, IRIF lab) - Confidence sharing: an economic strategy for efficient information flows in animal groups.
Social animals may share information to obtain a more complete and accurate picture of their surroundings. However, physical constraints on communication limit the flow of information between interacting individuals in a way that can cause an accumulation of errors and deteriorated collective behaviors. In this talk, I will theoretically discuss a general model of information sharing within animal groups, and take an algorithmic perspective to identify efficient communication schemes that are, nevertheless, economic in terms of communication, memory and individual internal computation. I will present a simple algorithm in which each agent compresses all information it has gathered into a single parameter that represents its confidence in its behavior. Confidence is communicated between agents by means of active signaling. It turns out that this algorithm competes extremely well with the best possible algorithm that operates without any computational constraints.
The proofs rely on the Cramer-Rao bound and on our definition of a Fisher Channel Capacity. These concepts are used to quantify information flows within the group and to obtain lower bounds on collective performance. The results suggest confidence sharing as a central notion in the context of animal communication.
The talk is based on a paper published in PLoS Computational Biology.

Oder Feinerman (Weizmann institute of science) - Algorithmic challenges in ant cooperative transport.
Ants that encounter a large food item may join forces to cooperatively carry it towards the nest. To do this, the ants must coordinate their forces while simultaneously navigating towards the nest, often through complex environments.This talk will discuss the behavioral algorithm that allows that ants to achieve this non-trivial feat. I will show how the ants' carrying algorithm allows the group to efficiently extract useful information that is made available by the navigationally competent individuals. Furthermore, I will show how the same simple algorithmic laws lead to the emergence of alternative group-level solutions when individual ant capabilities do not suffice.

Jonathan Behaegel (University of Nice, I3S lab) - Réseaux génétiques hybrides: de la logique de Hoare à l'identification de paramètres.
La modélisation de la dynamique des réseaux de gènes repose sur des paramètres qui reproduisent les comportements du système. Lorsque le temps considéré évolue de manière continu (temps chronométrique), l'identi cation des paramètres devient extrêmement di cile. Il devient alors nécessaire de prendre en compte de nouvelles contraintes provenant des données biologiques qui représentent, dans la plupart des cas, des durées entre des événements observés. Pour prendre en compte de telles durées, nous étendons le cadre de mod- élisation de René Thomas en considérant que chaque état qualitatif devient un sous-domaine de l'espace des concentrations dans lequel les trajectoires évoluent de manière continue. Ainsi, chaque trajectoire met un temps non nul à traverser chaque sous-domaine. Il s'agit d'une classe particulière d'automates hybrides linéaires dont les paramètres sont appelés "célérités".
Les traces biologiques observées (succession d'états qualitatifs combinés avec des durées) peuvent alors être interprétées comme une exécution de programme impératif, et la logique de Hoare munie du calcul de la plus faible précondition nous permet de construire des contraintes sur les paramètres du modèle. Nous illustrons ce calcul de la plus faible précondition sur un modèle simpli é du cycle circadien.

Celia Biane (University of Evry, Ibisc lab) - Inférence d'action sur les réseaux pour la reprogrammation cellulaire.
La reprogrammation consiste à modifier un système dynamique initial afin d’atteindre ou d’éviter certains comportements. En biologie, la reprogrammation cellulaire est étudiée notamment dans le cas du Cancer et de l’étude des cellules souches. Par exemple, au cours de la tumorigenèse, des évènements génétiques et épi-génétiques sont responsables d’une transition d’une cellule saine à une cellule cancéreuse et les médicaments reprogramment la cellule cancéreuse vers la mort cellulaire. Un défi majeur est l’identification, sur l'interactome (réseau biologique décrivant les interactions entre molécules), des perturbations causales de la maladie, et inversement, des cibles d’action des médicaments. Dans ce cas, la reprogrammation se caractérise par des actions structurelles de délétions ou de destruction de sous-ensembles de nœuds ou d’arcs responsables de transitions de comportement cellulaire. L'inférence d’actions causales est combinatoire et constitue un problème inverse à une simulation de perturbations structurelles responsables de modifications de la dynamique du réseau moléculaire. Dans cette présentation, nous décrirons la formalisation de ce problème par le gain et la perte de propriétés à l’équilibre dans un réseau booléen, nous présenterons des méthodes algorithmiques d’inférence d’actions sur les réseaux et montrerons un exemple d’application de ces méthodes pour la prédiction de «drivers» (évènements génétiques initiant un processus cancéreux) et de cibles thérapeutiques dans le cas du Cancer du Sein. Reprogrammation cellulaire, Problème Inverse, Inférence d’actions structurelles, Réseaux Booléens, Dynamique moléculaire, Cancer.


Participants