PhD in BioInformatics, Evolutionary Computation and Reinforcement Learning

 Autre · Thèse  · 36 mois    Bac+5 / Master   Laboratoire I3S · Sophia Antipolis (France)

 Date de prise de poste : 1 octobre 2021


Modelling of gene networks AI Reinforcement Learning bio-inspired methods.


The group « Bioinfo formelle » of I3S laboratory has proposed different formal frameworks to study the behaviour of gene regulatory networks. A regulatory network, modelled as a graph, defines the static interactions of a biological system : in the gene regulatory networks modelling setting, each interaction abstracts the individual influence of a gene x on the expression of another gene y. The dynamic of the network is governed by numerous unknown parameters that we want to identify.

René Thomas has proposed a discrete modelling framework that allows approximations of sigmoid functions (that usually represent the evolution of target gene expression level according to the level of the regulatoryentity) by step functions. Using this modelling framework, the exploration of the qualitative dynamics of the system can be described as path exploration in a finite state space and formal methods from computer science are useful.

In order to take into account temporal information, which plays crucial role in a wide range of biological systems, we also have developed, for several years, a hybrid modelling framework, which measures the time to go from a state to another one. This framework consists in a particular class of hybrid automata, but the crucial problem remains in the determination of accurate values for all numerous parameters. In order to address this question, we developed a "weakest precondition calculus" inspired by Hoare’s Logic (initially dedicated to imperative programs) which leads to constraints on parameters which has to befulfilled in order to make the model’s dynamics to be consistent with the observations. However, the exploitation of the constraints generated by the Hoare Logics is not so easy : classical constraints solvers are not able to extract solutions.

In this PhD, we envisage to use Artificial Intelligence (AI) techniques to overcome this problem whose difficulty is reinforced by the enormous number of parameters piloting the currently studied gene networks. Several techniques will be considered: bio-inspired meta-heuristics and einforcement learning.

Detailed PhD proposal is available at:


Procédure : Prendre d'abord contact auprès de et en envoyant un CV.

Date limite : 9 mai 2021


Jean-Paul Comet et Denis Pallez

Offre publiée le 6 avril 2021, affichage jusqu'au 20 mai 2021