Pierre-Éric Lutz & Vincent Vigon
Biography - Pierre-Éric Lutz
Pierre-Éric Lutz studied medicine before pursuing research training in Strasbourg and Montreal, Canada. He is a researcher at the French National Centre for Scientific Research (CNRS), and works within the Pain and Psychopathology team at the Institute of Cellular and Integrative Neuroscience (INCI) in Strasbourg.
After completing a bachelor’s degree in biological and medical sciences and a master’s degree in neuroscience in parallel with his medical studies, he completed a PhD in Strasbourg and a post-doctorate in Montreal, with a focus on epigenetics applied to the understanding of psychiatric disorders. He was recruited as a CNRS researcher in 2019.
During his studies on depression and substance use disorders (or addiction), Pierre-Éric Lutz gradually became more interested in the peculiar epigenetic properties of the main brain cell types, whether neuronal or glial. Over the last decade, he has been characterizing the epigenetic plasticity of these cells as a function of life experiences, and how such plasticity may help to better understand the life-long course of psychiatric disorders. His group addresses this question by analyzing human brain tissues and by developing behavioral and genetic models in mice.
During his recent work on DNA methylation, a major epigenetic mechanism, he developed an interest in recent sequencing methods that enable the analysis of increasingly long DNA fragments. This important technological breakthrough is expected to deepen the understanding of the informational nature of epigenetic encoding. It requires the use of machine learning approaches, and motivated an interdisciplinary collaboration with Vincent Vigon, from the Institute for Advanced Mathematical Research (IRMA) in Strasbourg.
Fellowship 2025
Dates - 01/10/2025-30/09/2027
Biography - Vincent Vigon
Vincent Vigon is assistant professor at the Institute for Advanced Mathematical Research (IRMA) of the University of Strasbourg, where he works within the “Probability” team. His research focuses on the applications of deep learning and the development of neural network architectures.
He studied applied mathematics at the University of Rouen Normandy (France) and carried out his PhD at the University of Manchester (United Kingdom). His research initially focused on the theory of Lévy processes and the study of their infinitesimal and asymptotic fluctuations, and then on Markov chains, revealing a link between Wiener-Hopf factorization and LU decomposition. In parallel with his research, he taught fundamental mathematics before specializing in teaching applied mathematics (signal processing, data processing, scientific computing), gradually integrating deep learning into his courses. He also contributed to the dissemination of science by creating and organizing the "parcours mathématique", an enigma-solving competition.
His research subsequently became more applied, specializing in the applications of deep learning, particularly by developing neural network architectures adapted to the structure and symmetries of the data, and then by following the evolution of PINNs (physics-informed neural networks). This new direction allowed him to collaborate with researchers from various disciplines, exploring applications in animal behaviour (sea turtle monitoring), medical imaging (brain vessel segmentation), astrophysics (modelling the solar corona and stars), and the resolution of partial differential equations (equilibrium states of organs and improvement of solvers for fluids, gases and plasmas).
Recently, Vincent Vigon initiated a collaboration in genetics with Pierre-Éric Lutz, exploring the complexity of the epigenome through the analysis of sequential and heterogeneous data. His objective is to develop neural network architectures adapted to this data, notably from long-read sequencing. He places particular importance on the valorisation of existing data, elegant and efficient programming (in particular with JAX), and the implementation of energy-efficient training strategies for neural networks.
Project summary
LEVERAGING LONG-READ SEQUENCING AND MACHINE LEARNING TO DEFINE THE NEURONAL METHYLOME TURNOVER AT SINGLE-CELL RESOLUTION
Epigenetic mechanisms are physical and chemical processes that determine the emergence of diverse cellular identities from a DNA sequence, which is identical across every cell of a given organism. They notably include non-coding RNAs, histone modifications, changes in the three-dimensional structure and conformation of chromatin, as well as DNA methylation, the substrate at the centre of the present project.
The dynamics that modify the genomic distribution of DNA methylation have been well-characterized at multiple temporal scales. Over evolutionary time, variations across cell-types have been associated with the individualization of species and organs, as shown in relation to the brain in both vertebrate and primate lineages. In the shorter lifespan of mammalian organisms, successive waves of writing and erasure of DNA methylation are necessary for the formation of germ cells, upon fertilization and during embryological development.
At the molecular level, the machinery that drives these kinetics is now well-characterized, and relies on specific enzymes for methylating and demethylating the DNA. It has long been assumed that somatic cells, such as post-mitotic neurons in the brain, show terminal and fixed DNA methylation profiles once they have reached a fully differentiated state. However, this paradigm has been challenged over the last decade. A growing number of studies indicate that neurons exhibit bidirectional DNA methylation changes as a function of aging, environmental exposure, or disease, among other factors. While these data suggest that DNA methylation enzymes remain active and may contribute to functional plasticity in neurons, the kinetics of such processes remain to be characterized.
This project was designed to address this key question in the understanding of brain function across physiological and pathological contexts. Through this collaboration between a neurobiologist, Pierre-Éric Lutz, and a mathematician, Vincent Vigon, the aim is to combine Nanopore long-read sequencing, machine learning and mouse genetic tools, in order to identify the specific properties and the turnover rate of the neuronal methylome.
