Vous êtes ici

Hybrid formalisms for learning and simulating multi-layer systems in biology

Equipe et encadrants
Département / Equipe: 
Site Web Equipe: 
Directeur de thèse
Anne Siegel
Co-directeur(s), co-encadrant(s)
NomAdresse e-mailTéléphone
Anne Siegel
Sujet de thèse

One of the characteristic of biological systems which make them much more complex that other experimental systems is that they are inherently built from several types of interactions which act at different time-scales and are modeled with different formalisms.  Metabolism operates compound transformations. They are controlled by specific proteins named enzymes whose production is them-self controlled by transcriptional regulatory events. The activation of this transcriptional layer is itself controlled by external inputs, based on signaling events which consist of chains of activations of proteins. 

However, a biological system cannot be viewed as a purely hierarchical controlled system because feed-forward mechanisms inform every layer on the state of the other ones.  In the last decade, a bunch of methods have been developed to identify the best models explaining the observed response of a biological system. The main drawback of all these methods is that they each rely on a simplified assumption of the system functioning, allowing to focus on a single layer of the biological system, adjoined with several dynamical assumptions (steady-state, synchronous dynamics). In this case, interactions between layers are rather neglected. Recent progresses in the measurements of biological process have shown that these simplifying assumptions are not more relevant to depict the complexity reported by the observed responses. For instance, Buescher et al. [2], by integrating metabolic and regulatory network analyses, simulated different regulation strategies for controlling nutritional shifts and compared their evolutionary benefit. They thus succeeded to identify the key regulatory events involved in the metabolic adaptive response to nutritional transitions in B. subtilis [5]. To handle this issue on a modelling framework, we need to reconsider the existing hierarchical approaches between the regulatory, signaling and metabolic scales and to set up a hybrid formalism for the simulation and the learning of multi-scale biological system.

The goal of the thesis will be to build scalable hybrid formalisms for learning multi-layer systems. In practice, the approach will consist in designing design a relevant framework and the corresponding methods to efficiently infer Boolean rules which may govern the response of a continuous system to different perturbations. The cornerstone is to encompass the biological complexity within a hybrid framework gathering a dynamic optimization canvas using differential-algebraic equations. We already now how to learn Boolean models from times-series datasets representing the direct output of the Boolean models [7,8] by solving combinatorial problems with a paradigm of constraint programming (Answer Set Programming) [1,4]. Learning multi-layer systems requires to learn Boolean models from the output of the lower scale, that is, the results of Mixed Integer Linear Programming problems. To that goal, we will rely on a recent breakthrough in Answer Set Programming showing that is becomes possible to solve hybrid problems mixing combinatorial and linear constraints [3,6].  The methods developed during the thesis will therefore allow us both to have a refined vision of a biological system dynamics and to refine the learning methods for regulatory processes. 


[1] C. Baral. Knowledge Representation, Reasoning and Declarative Problem Solving. Cambridge University Press, 2010. 

[2] J. M. Buescher, W. Liebermeister, M. Jules, et al . Global network reorganization during dynamic adaptations of Bacillus subtilis metabolism. Science, 335(6072) :1099–1103, Mar 2012.

[3] C. Frioux, T. Schaub, S. Schellorn, A. Siegel, P. Wanko, Hybrid Metabolic Network Completion, in : 14th International Conference on Logic Programming and Nonmonotonic Reasoning - LPNMR 2017, M. Balduccini, T. Janhunen (editors), Logic Programming and Nonmonotonic Reasoning, 10377, Springer, p. 308–321, Espoo, Finland, July 2017.

[4] Martin Gebser, Roland Kaminski, Benjamin Kaufmann, and Torsten Schaub. Answer Set Solving in Practice. Synthesis Lectures on Artificial Intelligence and Machine Learning. Morgan and Claypool Publishers, 2012.

[5] Radhakrishnan Mahadevan, Jeremy S Edwards, and Francis J Doyle. Dynamic flux balance analysis of diauxic growth in escherichia coli. Biophysical journal, 83(3) :1331–1340, 2002.

[6] S. Prigent, C. Frioux, S. Dittami et al ., Meneco, a Topology-Based Gap-Filling Tool Applicable to Degraded Genome- Wide Metabolic Networks, PLoS Computational Biology 13, 1, January 2017, p. 32.

[7] S. Videla, C. Guziolowski, F . Eduarti et al, Learning Boolean logic models of signaling networks with ASP, Journal of Theoretical Computer Science (TCS), 599, September 2015, p. 79–101.

[8] S. Videla, J. Saez-Rodriguez, C. Guziolowski, A. Siegel, caspo : a toolbox for automated reasoning on the response of logical signaling networks families, Bioinformatics, 2017.

Début des travaux: 
Mots clés: 
bioinformatique, représentation des connaissances, systèmes dynamiques, apprentissage
IRISA - Campus universitaire de Beaulieu, Rennes