The lab is focused on the algorithms and computation selected by evolution to perform biological decision-making. We address this topic with an interdisciplinary approach mixing statistical physics, Bayesian machine learning, information theory and various experimental biological setups. We are pursuing 4 research axis:
Probabilistic pipelines and Artificial Intelligence to probe single biomolecule random walks (https://goo.gl/4UH7NZ). We designed an probabilistic software that identifies random walks and quantify their properties from single molecule microscopy imaging. We published more than 10 papers on the subject (e.g. Phys. Rev. Lett 2009, Biophys. J. 2012-2014-2016, Nat. Method. 2015, Science 2015, Cell. Rep 2017, Sci. Rep. 2018) and applied our schemes to receptor in synapses, CRISPR in the nucleus, HIV Virion formation etc. Current efforts are focused on TRamWAy, a machine learning software that entirely automates single molecule experiment analysis while being scalable to analyse Terabytes of data in few hours.
Diffusion landscape of Glycine receptor on the surface of a neuron.
Decision-making of biological system (https://goo.gl/28TfNC). We study biological intelligence and computation. We have developed approaches to probe biological decision by joining physical modelling, Bayesian inference and information theory to biological experiments. We probed the memory of bacteria (PNAS, 2012), the search strategies of leucocytes (Current Biol 2013) and decision making of insects in turbulent flows (J. phys. A. 2009, PNAS 2013, J. vis. Exp. 2014, Front. in AI. 2015, Plos. Comp. Biol. 2017, 2 patents). We are now focusing on algorithms implemented in the brain of the drosophila larva (Cell 2017).
Color coded behaviour of a drosophila larva as one neutron is being ontogenetically activated
Machine learning in Virtual Reality: DIVA ( https://goo.gl/dnNueu ). This project is a collaboration between my lab and Maxime Dahan’s lab. The project joins human cognition, probabilistic AI and virtual reality to explore complex unstructured data. Results are included in the software DIVA (J Mol Biol 2019). The DIVA-medical platform is now introduced in 4 hospitals with applications ranging from breast cancer surgery planning to pre-natal diagnosis from MRIs. The DIVA microscopy platform is now installed in more than 10 labs.
Confocal image of mice anterior/dorsal hippocampus neurons in DIVA
Numerical Methods for temporal networks:Time-varying (temporal) networks provide an efficient formalism to study many complex real-world systems. They notably provide a natural modeling framework of structural and functional brain networks, protein interaction networks, social interactions, information and epidemic spreading, and infrastructural and financial networks. Real-world temporal networks and dynamic processes that take place in them show heterogeneous, non-Markovian, and intrinsically correlated dynamics, making their analysis particularly challenging. This project aims at developing versatile, robust, and scalable numerical methods for the analysis of empirical temporal networks and dynamical processes that take place in them. We currently focus on defining a general and consistent framework for numerically generated randomized reference (null) models (RRMs) for temporal networks and an automated procedure for generating and interpreting RRMs for empirical networked systems