Je suis directeur de recherche en neurosciences théoriques et computationnelles à l'Institut de la Vision où je dirige l'équipe de Neurosciences computationnelles des systèmes sensoriels (CV). Je m'intéresse à la modélisation neuronale de la perception, à l'initiation des potentiels d'action, et également à l'épistémologie.
I am a research director in computational and theoretical neuroscience in the Vision Institute, where I lead the Computational neuroscience of sensory systems group (CV). My main interests are neural modeling of perception and spike initiation. I also have an interest in epistemology.
Please don't hesitate to contact me if you are interested in working in my group.
1. Goethals S, Brette R (2020). Theoretical relation between axon initial segment geometry and excitability. Spikes are initiated in the axon, close to the soma. To understand this situation theoretically, I previously introduced resistive coupling theory, which applies to the case when a hot spot of excitability (the axon initial segment or AIS) sits on a thin process (the axon) connected resistively to a large somatodendritic compartment that acts as a current sink (Brette 2013; resistive coupling theory was called “compartmentalization” in that paper). Here we develop the theory further to understand quantitatively the relation between the geometry of the AIS (position and length) and excitability. We find in particular that the neuron gets more excitable when the AIS moves away from the soma, but only slightly so, and that this relation may be reversed if the AIS is hyperpolarized relative to the soma. The analysis suggests that the main effect of AIS movements might be not so much on excitability per se as on other electrical aspects like the backpropagating current. Other aspects of the theory were developed in Hamada et al. (2016), Teleńczuk et al. (2017, 2018), and reviewed in Kole & Brette (2018). (See also code)
2. Brette R (2019). Is coding a relevant metaphor for the brain? I argue that the mainstream view that neural activity “encodes” properties of things in the world is misguided. It is more than a figure of speech: it is a model of the brain, a dualistic model in which some neurons are in the business of producing representations, and some others are in the business of manipulating them – as if spikes were variables (something that can be manipulated) and not activity (something that happens). The coding model of the brain is wrong on two accounts: neural codes do not have the quality of representations, both on empirical and theoretical grounds, and neural codes have no causal powers, and therefore cannot be the basis of a model of the brain. (see the Open call for commentary, my response to commentaries, and a podcast interview on Brain Inspired).
3. Le Mouel C and Brette R (2017). Mobility as the purpose of postural control. We analyze the organization of postural movements in a variety of contexts, for example when standing and when anticipating movements. We show that the adjustments of posture are organized not to stabilize the body, as usually claimed, but to contribute to the action itself. This is most obvious in sports, where for example sprinters shift their center of mass forwards before the race starts to help initiate the run, but it is also seen in ordinary situations such as standing.
4. Brette R (2015). What Is the Most Realistic Single-Compartment Model of Spike Initiation? I review experimental evidence demonstrating that, as far as spike initiation is concerned, the integrate-and-fire model is more realistic than the single-compartment Hodgkin-Huxley model – which described the space-clamped squid giant axon.
5. Brette R (2013). Subjective physics. This is an epistemological paper where I describe the structure and properties of “subjective physics”, the laws that govern the interface of an organism with the world (sensory signals and actions). I propose that this is the “computational level” that is relevant to computational and systems neuroscience, rather than objective descriptions of the world meant for an external observer.
Being firstly a theoretical neuroscientist, I study the nervous system by means of designing, analyzing and simulating mathematical models. We use models and theories to connect two levels of understanding, for example molecular properties with neuron function, or neuron properties with behavior. The term computational in computational neuroscience refers to “how the brain computes”, i.e. the articulation between neuron properties and behavior, but the computer metaphor should not be taken too seriously (see also my work on the epistemology of neuroscience). I also have an experimental activity, mainly patch-clamp experiments to test theories of spike initiation.
I am currently interested in three broad topics: 1) the initiation of spikes (theoretical neuroscience); 2) sensory systems (computational neuroscience); 3) simulation technology (neuroinformatics). I am also working on the design of intracellular recording techniques and on epistemology of neuroscience.
Vertebrate neurons interact mainly by stereotypical electrical impulses called action potentials or “spikes” (see my paper on firing rate vs. spike timing). Thus a critical question is how neurons transform input signals into spike trains. At a general level, this is of course well known: sodium channels open when the membrane potential exceeds a threshold. But there are many subtleties. For example, the spike threshold depends on previous activity, on multiple timescales, with changes mediated by ionic channel properties (on a short timescale) and by structural changes (on a long timescale). Space also plays a critical role: I have recently shown theoretically that the axonal initiation of spikes makes sodium channels open as a discrete function of somatic voltage, effectively making the integrate-and-fire model much more realistic than previously thought (see this graphical explanation). There are also many unanswered questions, for example: how are the various ionic channels coordinated (in properties and in spatial distribution) so that spike initiation is functional and efficient? how is spike initiation modulated by activity on the long term? what is the function of the various types of channels in the axonal initial segment? For theoretical neuroscience, this is largely unexplored territory.
I am interested in how sensory systems work in ecological environments, and in particular in the perception of space, which is shared across almost all sensory modalities (including pain). First, unlike lab environments, ecological environments are never simple. I have developed models of how neural circuits might deal with this complexity, from the view that perception relies on the identification and manipulation of models of the world, understood as relations between sensory signals. I call these perceiver-oriented models “Subjective physics”. This view connects with major theories in psychology (Gestalt psychology, Gibson's ecological approach, O'Regan's sensorimotor theory), philosophy of mind (Poincaré, Merleau-Ponty) and linguistics (Lakoff). I have proposed that sensory relations are identified as temporal invariants in the sensory flow (ie relations that are satisfied over a contiguous period of time), which is most directly connected with Gibson's notion of “invariant structure”. I have developed this idea mostly in the context of spatial hearing and pitch perception, by proposing the concept of “synchrony receptive field”. Second, I have come to realize that perception, especially spatial perception, can only be understood with integrative sensorimotor models. Thus, I have started working on the theory of sensorimotor systems. Recently, I started to develop a completely new line of research on the integrative modeling of a spiking unicellular organism, Paramecium, which is a simple multimodal sensorimotor system.
In 2008, I started the Brian simulator with Dan Goodman (postdoc at the time and now lecturer in Imperial College, UK). It is a simulator for spiking neural networks written in Python. The focus is on flexibility and ease of use, which has made it a highly popular tool in neuroscience. All models are directly specified by users with their equations – there are no predefined models, which has many benefits. It is also possible to simulate multicompartemental models. The new version (2.0), which is now mainly developed by Marcel Stimberg, relies on code generation, which makes Brian much faster. We are currently working on running it on multiple types of hardware (collaborations are welcome).