Non exhaustive list of research subjects for Ph.D. theses, Post docs, or internships at different levels. If you are a motivated talented student with your own subject on topics relate to LIFEWARE, or for any question on the subjects below, please contact by email François Fages.


Ph.D. theses and PostDoc positions in LIFEWARE


> Biochemical Computing Ph.D. Thesis (3 years) or PostDoc (2 years)

funded by ANR-MOST project BIOPSY

> Machine learning models from temporal data Ph.D. Thesis (3 years)


Research Internships for Masters and Engineering Schools (with possible continuation for a Ph.D. Thesis)


> Graph Matching by SAT solving research internship (3-6 months)

funded by ANR project HYCLOCK

> Biochemical Computing research internship (3-6 months)

funded by ANR-MOST project BIOPSY

> Machine learning models from temporal data research internship (3-6 months)

funded by ANR project HYCLOCK

> Model reduction by tropical algebra constraint solving research internship (3-6 months)

> Biochemical graphs, loops and multistationarity (3-6 months)

Contact : Sylvain Soliman

Abstract : The main objective of this internship is to evaluate and improve on the results of [1] about stronger necessary (circuit) conditions for multistationarity than those of Thomas [3] / Soulé [2]. The test models will come from the literature, from the BioModels database, and from our biologist partners studying the Circadian Rhythms (Nobel prize 2017 theme). The first task will be to find which models did fill the Thomas / Soulé necessary conditions, then which do fit the stronger conditions. For the conditions relying on the composition of graph transformations, heuristic search will probably be necessary. Once the models identified, going back to the original articles and determining what was known about the stationary states of the model will allow us to evaluate the real output of the method.

Graphes biochimiques, boucles et multistationnarité

Résumé : L’objectif principal du stage est d’évaluer et d'améliorer les résultats de [1] sur des conditions (de circuit) nécessaires pour la multistationnarité plus strictes que celles de Thomas [3] / Soulé [2]. Les modèles de tests pourront provenir de la littérature, de la base de modèles BioModels ou de nos partenaires biologistes qui étudient les rythmes Circadiens (thème du prix Nobel 2017). Il s’agira dans un premier temps de vérifier quels modèles remplissent les conditions initiales de Thomas/Soulé, puis de vérifier avec les conditions plus strictes. Pour les conditions les plus avancées nécessitant des compositions de transformations de graphe, il sera sans doute nécessaire de procéder de manière heuristique. Une fois des modèles identifiés, il s’agira de comparer avec les articles d’origine afin de déterminer si la multistationnarité était déjà connue/supposée.

Références :

[1] Sylvain Soliman. A stronger necessary condition for the multistationarity of chemical reaction networks. to appear in the Bulletin of Mathematical Biology, September 2013.

[2] Christophe Soulé. Graphic requirements for multistationarity. ComplexUs, 1 :123–133, 2003.

[3] René Thomas. On the relation between the logical structure of systems and their ability to generate multiple steady states or sustained oscillations. Springer Ser. Synergetics, 9 :180–193, 1981.


Research Initiation Internships for L3, M1 and Engineering school students:


> Static analyses for biochemical reaction network visualization

Contact: François Fages

In the domain of computational systems biology, biochemical reaction diagrams are a central medium of communication between biologists and modellers. Unfortunately in this context, automatic graph visualization tools such as GraphViz [1] do not produce well-organized diagrams, and manual edition tools such as Cell Designers are preferred.

The purpose of this internship is to explore the use of static analyzers of reaction networks to extract information relevant to the "logical" drawing of biochemical reaction networks. More specifically, we propose to explore the use of the static analyzers of our modeling software, the Biochemical Abstract Machine (Biocham) [2], to design placement constraints for tools like GraphViz. For instance, the computation of linear conservation laws (i.e. Petri net P-invariants [3]) in Biocham provide information on groups of molecules that are transformed but conserved under their different forms and that could be preferably drawn on the same line. The use of dynamic analyzers, e.g. relying on numerical simulation, will be considered also if necessary [3].

References:

[1] GraphViz http://www.graphviz.org

[2] Biocham v4 http://lifeware.inria.fr/biocham/

[3] François Fages. AI in Biological Modeling. To appear in A Guided Tour of Artificial Intelligence Research. Springer-Verlag, 2017. preprint

> Benchmarking modern ODE integration methods for biochemical reaction networks

Contact: Sylvain Soliman

Currently the Biocham v4 [2] modellng software uses the GNU Scientific Library as go-to method for integrating ODE systems corresponding to the time-evolution of biochemical networks. Biocham v3 used a (Prolog) hand-coded version of some Rosenbrock method.

As is now common knowledge in the ODE-integration community [1] GSL is quite bad at this task and some more recent methods might do much better either for stiff or non-stiff systems.

Using the above blog post as reference, a benchmark of those recent methods, mostly those available through the Sundials library, notably the recent ARKODE, on real dynamical systems coming from Systems Biology publications, would be a huge improvement for the Biocham suite. It is currently not known how often the considered systems are stiff or semi-stiff, how often they use events, etc.

The results would be directly added to the Biocham suite [2] and might result in a tool-article.

Benchmarking des méthodes modernes d'intégration d'EDOs pour les réseaux biochimiques

Contact: Sylvain Soliman

À l'heure actuelle, l'outil Biocham v4 [2] utilise GNU Scientific Library comme source générique de méthodes d'intégration de systèmes d'équations différentielles correspondant à l'évolution de réseaux biochimiques. Biocham v3 utilisait une méthode de Rosenbrock implémentée en interne (en Prolog).

Il est désormais bien connu de la communauté d'intégration d'EDOs [1] que GSL est un assez mauvais choix et que des méthodes plus récentes seraient bien plus efficaces dans le cas de systèmes stiffs comme non-stiffs.

En partant de l'article de blog ci-dessus comme référence, le stage consistera à évaluer les méthodes récentes, notamment celles disponibles dans la bibliothèque Sundials, et en particulier ARKODE, sur des systèmes dynamiques réels provenant de publications en Biologie Systémique. Cela permettra une amélioration sensible de la suite Biocham [2] dans laquelle les résultats seraient immédiatement intégrés. Par exemple, on ne sait pas aujourd'hui quelle proportion de ces systèmes sont stiffs ou semi-stiffs, lesquels utilisent des évènements, etc.

Une publication sous forme d'article-outil est envisagée.

References:

[1] http://www.stochasticlifestyle.com/comparison-differential-equation-solver-suites-matlab-r-julia-python-c-fortran/

[2] Biocham http://lifeware.inria.fr/biocham/