Franco-Japanese collaboration on bio-informatics at Centrale Nantes

Mathematics and computer science at the service of living things

The numerous applications for health in Morgan Magnin's research work are centered around support for disease diagnosis and design of new therapies. They will thus contribute to the implementation of personalized medicine adapted to the needs of each individual.

As a professor at Centrale Nantes within the MeForBio research group at the LS2N Laboratory (UMR 6004 CNRS), Morgan Magnin works closely with Professor Katsumi Inoue, of the laboratory of the same name, at NII Tokyo. In the course of the last seven years, their collaboration has produced the following results:
 

• 4 Centrale Nantes/LS2N PhD students have undertaken internships (at least 3 months each) at Katsumi Inoue’s laboratory through the NII Call for Internships; 
• M. Magnin, who has held a visiting professor position in the Katsumi Inoue's laboratory since 2015, has spent a total of 2 years to date at NII, on Japanese budgets (notably a JSPS Invitation Fellowship);
• A former PhD student under Katsumi Inoue, Tony Ribeiro, became a postdoctoral fellow for 3 years at Centrale Nantes / LS2N;
• Morgan Magnin and Tony Ribeiro are members of the Katsumi Inoue’s KAKENHI Kiban-A Project "Learning Relational Dynamics from State Transition” (2017-2020)
• Joint publications: 5 international journals, 8 international peer-reviewed conferences with proceedings, and 3 international workshops or domestic conferences. 

Systems Biology

Their work focuses on systems biology and considers living organisms as interacting network systems. The behaviour of these systems is observed in order to construct a model or logical programme, which can explain or predict a situation. For example: Can we rely on a person's genetic makeup to predict whether or not they will catch and/or pass on the flu virus?

Morgan Magnin is also taking the lead on the Franco-Japanese PHC Sakura project involving Centrale Nantes and the University of Kobe, which launched in early 2019.
 

Learn more: Summary of academic exchanges

Katsumi Inoue’s laboratory at the National Institute of Informatics (NII) in Tokyo and the MeForBio team at Centrale Nantes/LS2N have been working together since 2011.
 
They are working on the respective merits - and complementarity - of logical modeling (in terms of learning and verification) to tackle complex dynamic systems, especially in Systems Biology.
 
While analyzing large-scale dynamic systems, there are some key properties that are common to all systems, whatever their nature (real-time, biological, logistics). The MeForBio team at Centrale Nantes / LS2N and Katsumi Inoue’s laboratory at NII have been historically interested in such properties like attractors, reachability etc. The collaboration started through the common interest of both teams in formal approaches to perform dynamical analysis of large-scale complex systems: the French team developed efficient static analysis algorithms to address dynamic properties of automata networks; the Japanese team has a strong background in Boolean networks and their logical modeling. The combination of these approaches led to successful methods to tackle models with incomplete knowledge about the cooperation between components.
 
Since then, collaboration has focused on learning dynamic systems from the observation of their changes in terms of state-transitions. Based on seminal works from Katsumi Inoue, they designed various extensions about the LFIT (Learning from Interpretation Transition) framework to learn dynamical systems from state-transitions observations. They applied this approach to real-life data coming, on one hand, from biologists, and on the other, from open worldwide machine learning challenges (DREAM Challenges). Together, in 2016, they took part in the DREAM11 Challenge on the predictability of developing influenza symptoms with regard to gene expression. Taking part in this challenge gave them further ideas for their research. When faced with real-life data, the main drawback of a pure logical-based approach is the need to discretize raw data. To overcome this limit, they recently designed a method that learns the dynamics of the system directly from continuous time-series data as well as a semantics-free LFIT extension.