Projet de thèse en Neurosciences
Sous la direction de Thierry Mora et de Olivier Marre.
Thèses en préparation à Paris Sciences et Lettres , dans le cadre de Cerveau, cognition, comportement , en partenariat avec LABORATOIRE DE PHYSIQUE STATISTIQUE DE L'E.N.S. (laboratoire) et de Ecole normale supérieure (établissement de préparation de la thèse) depuis le 01-09-2016 .
C.f. résumé en anglais.
Online Characterization of the Retinal Network
# Context We are able to record from large ensemble of ganglion cells, the retinal output, while it is stimulated with complex, natural scenes [Marre et al., 2012]. However, how the retinal network processes these scenes is still poorly understood. This is mostly due to two issues, which pertain most sensory systems. First, the large dimension of the input and output space make a full characterization impossible. This is the famous curse of dimensionality issue. Second, sensory processing is context dependent. Most neurons will not respond the same way to a stimulus when it is presented alone or embedded in a complex visual scene. Therefore it is still hard to draw conclusions about the way natural scenes are processed by sensory networks from their responses to artificial stimuli. # Purpose To circumvent these problems, this interdisciplinary project will develop a novel approach to characterize sensory neurons. The key novelty of this project is a perturbative approach, where we will look for the perturbations of a natural scene that are the most efficient at driving the neuronal response of individual neurons. By finding the perturbations in a complex stimulus that can trigger the largest change in the neuronal response, or conversely the ones that leave it invariant, we will characterize the selectivity of each neuron. To characterize the selectivity of individual cells this way, we need to perform closed-loop experiments: starting from a given stimulus, we will actively look for the perturbations triggering the largest change. To do so, we need to be able to extract the activity of well-isolated cells from the large-scale extracellular recordings during the time course of the experiment. This task of online spike sorting has not been achieved for such large-scale recordings yet, and is an important technical challenge. Recently we have developed a spike sorting software that can extract activity of individual cells from these recordings almost automatically [Yger et al., 2018]. A key advantage of our algorithm is to be entirely parallel, and to use optimally the computing resources (both CPUs and GPUs). As a result, it can process an hour of recording in an hour of time using 4-5 desktop computers working in parallel. # Workplan A first step of this project will be devoted to developing an online version of our spike sorting software, to enable closed-loop experiments. Starting from the current software, we will update the method so that recordings are processed on the fly. Our purpose is to have a software that extracts isolated spikes ~100 ms after the recording, a delay short enough for most applications. For this part the student will benefit from the help of Pierre Yger (Institut de la Vision), who has been leading the development of our novel spike sorting algorithm. In a second step, we will use this tool to characterize the selectivity of single ganglion cells responding to natural stimuli. For this we will stimulate cells with natural movies, add on top of the movies discs of different sizes and polarity, and search actively for the size and polarity that will trigger the largest changes in the neuronal response for different individual cells. We will compare the obtained selectivity to the one classically measured with artificial stimuli, and determine if this selectivity changes with the type of natural movie used. Our purpose is to test if the selectivity measured here remains invariant to the visual context. We will also apply a similar approach to ensembles of cells of the same type. We will assume that the most important information coming from a cell type is the one that remains invariant to the context in which it is presented. As a consequence, we will look for perturbation that can be decoded reliably from the activity of the neuronal population despite being presented in different visual contexts. Finally, having determined what information is carried by individual cells and populations, we can also measure the specific contribution of a given cell: we will look for perturbations that maximize the information carried by one cell that was not present in the rest of the population. # Previous work This project builds on a recent work of the advisors where we performed closed-loop experiments where the stimulus was modified in real time to find the smallest perturbations that were effective at changing the population response. However, we could only analyze the population response since we did not have access to the response of individual cells online. This project will remove this roadblock and design novel methods and concepts to understand sensory processing and impact the entire field of sensory systems. # Impact Together, these experiments will design novel methods to characterize how sensory cells process visual stimuli. The methodology developed here to understand retinal processing could potentially be applied in any sensory structure to better understand sensory processing. The development of an online spike sorting method will produce a valuable tool that can be useful for many applications in neuroscience, where analyzing the neural activity during the experiment is needed. A software to conduct online analysis will considerably ease the monitoring of long-lasting experiments by providing a high quality feedback instead of raw recordings. It will make accessible a totally new class of closed-loop experiments. We are currently in contact with several companies to sell our current version of the spike software. The online spike sorting software that will be designed during this PhD project could also be commercialized, and the student will be associated to this.