Brain-Computer Interface (BCI), by translating brain activity into commands for communication or control, is a promising tool for patients who suffer from neuromuscular pathologies or lesions. Nevertheless, it fails to detect intents in 15-30 % of the BCI users, due notably to a poor understanding of the mechanisms underlying the BCI performance. Here, we aim at using a biophysically interpretable and analytical model to identify biophysical changes occurring while controlling a BCI. We hypothesized that excitatory and inhibitory neuronal populations model parameters will differ when comparing the performed tasks in a BCI setting. We used source-reconstructed magnetoencephalography signals in a BCI framework where 19 subjects were instructed to modulate their brain activity to control the position of a cursor displayed on a screen by either performing a motor imagery task or remaining at rest [1]. We divided the cohort into two subgroups, namely G1 and G2, with subjects who performed better or worse than the average respectively. We employed a linearized neural mass model to infer four biophysically realistic parameters from the estimation of the power spectra: two neural gains capturing overall synaptic strength between excitatory and inhibitory neuronal populations (g_ei) and among inhibitory neuronal populations (g_ii), time constant of the excitatory neuronal population (tau_e), and time constant of the inhibitory neuronal population (tau_i) [2]. We inferred the optimal model parameters to match the shape of the modeled power spectra with the empirical power spectra for each subject during both rest and MI. We then compared the model parameters between rest and MI. To check that the spectral power in the alpha frequency band carried relevant information, we performed statistical tests between the Rest and the MI conditions on data from G1 and G2. Whereas no significant difference between Rest and MI in G2, in G1 significant condition effects were observed in associative and sensorimotor regions (Fig 1A). Then, we studied to which extent the excitatory/inhibitory neuronal population parameters could differ depending on the performed task. The neural gain g_ei shows a significant condition effect in regions involved in visual motion processing in G1 and in regions involved in the default mode network in G2. The neural gain g_ii significantly differs between Rest and MI in regions involved in decision-making processes in G1 and in areas involved in attention processes in G2. The time constant tau_e shows no significant condition effect in G1 whereas in G2, such an effect was observed in areas involved in visual recognition. Lastly, the time constant tau_i shows a significant condition effect in regions involved in motor imagery performance and in decision making processes in G1 (Fig 1B) and in areas involved in attention processes in G2. These results indicate changes in the excitatory and the inhibitory between the conditions with an alteration of the inhibitory neuronal population activity over sensorimotor areas in the most responsive subjects only. These can be potentially used as biophysically realistic markers of BCI performance.
[1] Corsi, M-C, et al. Functional disconnection of associative cortical areas predicts performance during BCI training.
Transient or partial synchronization can be used to do computations, although a fully synchronized network is frequently related to epileptic seizures. Here, we propose a homeostatic mechanism that is capable of maintaining a neuronal network at the edge of a synchronization transition, thereby avoiding the harmful consequences of a fully synchronized network. We model neurons by maps since they are dynamically richer than integrate-and-fire models and more computationally efficient than conductance-based approaches. We first describe the synchronization phase transition of a dense network of neurons with different tonic spiking frequencies coupled by gap junctions. Then, we introduce a local homeostatic dynamics in the synaptic coupling and show that it produces a robust tuning towards the edge of this phase transition. We discuss the potential biological consequences of this self-organization process, such as its relation to the Brain Criticality hypothesis, its input processing capacity, and how its malfunction could lead to pathological synchronization
Berger first recorded human EEG on 6 July 1924, the first noninvasive measurement of human brain activity; he noted the ~10 Hz alpha rhythm to be the most prominent activity [1]. Alpha is concentrated over visual cortex at the back of the head, sometimes displays a double peak, and is suppressed by visual inputs [2]; the beta rhythm occurs at its harmonic. Later, the ~10 Hz mu rhythm was discovered, concentrated over sensorimotor cortex near the crown of the head, suppressed by motor activity, and sometimes associated with ~20 Hz activity [2]. The ~10 Hz tau rhythm is concentrated over auditory cortex near the ears and is suppressed by sound. Early theories argued that separate groups of neurons fire at ~10 Hz or ~20 Hz at the relevant locations, but these were ad hoc and lacked explanatory power [3]. More recently, the alpha rhythm was argued to be a natural mode of activity in the cortex [3] or of the corticothalamic (CT) system [4,5], and was analyzed using neural field theory (NFT), which averages over the activity of large numbers of neurons in order to calculate the dynamics of activity fields. Here, we show that just 4 corticothalamic eigenmodes of activity can explain the key features of spontaneous alpha, mu, and tau rhythms, including their frequency structure and topography [5]. Splitting is due to eigenmodes having different frequencies, whereas CT loops account for the basic 10 Hz frequency and correlations between alpha and beta, and between mu and its harmonic. Observed split-alpha, split-beta, and split-mu rhythms are explained, and it is predicted that split-tau and split second-harmonic mu and tau rhythms can occur. Spatial concentrations of activity are found to be due to constructive interference of modes in the relevant sensory regions, supported by enhanced CT gains , and are suppressed when those gains are reduced by attention [5. Fits of theory to data will enable brain states to be probed in real time, as is already the case for spectra including the basic alpha rhythm [7]. Links to evoked responses and other phenomena can also be made via NFT using eigenmode analysis, thereby unifying many classes of observations and phenomena and providing a systematic means of calculation.
A major challenge to understand how neurons process synaptic input is to build detailed biophysical models of neuronal function that integrate morphology, electrophysiology, and biochemical reactions. The Purkinje cell is the principal neuron of the cerebellar cortex. Modeling of Purkinje cell electrophysiology spans more than 60 years of efforts [1]. This neuron contains voltage activated calcium conductances as well as calcium activated potassium conductances. These conductances are distributed over the soma, axonal initial segment, and a complex dendritic tree. As such, there is a need to accurately and efficiently model intracellular calcium diffusion. Furthermore, calcium is a second messenger essential for the activation of biochemical reactions involved in the expression of long-term synaptic plasticity [2]. In all models of Purkinje cells, calcium diffusion is assumed to be exclusively a radial process. However, synaptic plasticity in the granule cell to Purkinje cell synapse requires calcium influx through membrane conductances in conjunction with calcium released from intracellular stores. The synapses are on passive dendritic spines. Since voltage activated calcium conductances are only on the dendrite, but not on the spine, there is a need to model axial diffusion between these compartments to study synaptic plasticity in full Purkinje cell models. In this project we will describe our efforts to integrate the reaction-diffusion (RXD) NEURON [3] module into a highly detailed model of a Purkinje cell. The first study aims to reproduce the normal excitability of the cell under current clamp conditions. The second implements axial diffusion in the spiny dendrites and 3D diffusion in the smooth dendrites and soma compartments for computational efficiency. The third study looks at axial diffusion between the dendrite and a passive spine head after activation of the climbing fiber input. Finally, we implement a reduced model where a dendritic segment has one spine. The spine contains the biochemical reactions involved in the expression of long-term depression (LTD) [4]. We will describe the technical challenges, advantages and disadvantages of each implementation compared to traditional methods. Our study will be a platform for all those interested in using realistic reaction-diffusion models with morphologically complex neuronal models.
References 1. Bower, J.M., The 40-year history of modeling active dendrites in cerebellar Purkinje cells: emergence of the first single cell "community model". Front Comput Neurosci, 2015. 9 p. 129. 2. Zamora Chimal, C.G. and E. De Schutter, Ca(2+) Requirements for Long-Term Depression Are Frequency Sensitive in Purkinje Cells. Front Mol Neurosci, 2018. 11 p. 438. 3. Carnevale, N.T.a.H., M.L. The NEURON Book. and U.C.U.P. Cambridge, 2006. 4. Kuroda, S., N. Schweighofer, and M. Kawato, Exploration of signal transduction pathways in cerebellar long-term depression by kinetic simulation. J Neurosci, 2001. 21(15): p. 5693-702
A major challenge to understand how neurons process synaptic input is to build detailed biophysical models of neuronal function that integrate morphology, electrophysiology, and biochemical reactions. The Purkinje cell is the principal neuron of the cerebellar cortex. Modeling of Purkinje cell electrophysiology spans more than 60 years of efforts [1]. This neuron contains voltage activated calcium conductances as well as calcium activated potassium conductances. These conductances are distributed over the soma, axonal initial segment, and a complex dendritic tree. As such, there is a need to accurately and efficiently model intracellular calcium diffusion. Furthermore, calcium is a second messenger essential for the activation of biochemical reactions involved in the expression of long-term synaptic plasticity [2]. In all models of Purkinje cells, calcium diffusion is assumed to be exclusively a radial process. However, synaptic plasticity in the granule cell to Purkinje cell synapse requires calcium influx through membrane conductances in conjunction with calcium released from intracellular stores. The synapses are on passive dendritic spines. Since voltage activated calcium conductances are only on the dendrite, but not on the spine, there is a need to model axial diffusion between these compartments to study synaptic plasticity in full Purkinje cell models. In this project we will describe our efforts to integrate the reaction-diffusion (RXD) NEURON [3] module into a highly detailed model of a Purkinje cell. The first study aims to reproduce the normal excitability of the cell under current clamp conditions. The second implements axial diffusion in the spiny dendrites and 3D diffusion in the smooth dendrites and soma compartments for computational efficiency. The third study looks at axial diffusion between the dendrite and a passive spine head after activation of the climbing fiber input. Finally, we implement a reduced model where a dendritic segment has one spine. The spine contains the biochemical reactions involved in the expression of long-term depression (LTD) [4]. We will describe the technical challenges, advantages and disadvantages of each implementation compared to traditional methods. Our study will be a platform for all those interested in using realistic reaction-diffusion models with morphologically complex neuronal models.
References 1. Bower, J.M., The 40-year history of modeling active dendrites in cerebellar Purkinje cells: emergence of the first single cell "community model". Front Comput Neurosci, 2015. 9 p. 129. 2. Zamora Chimal, C.G. and E. De Schutter, Ca(2+) Requirements for Long-Term Depression Are Frequency Sensitive in Purkinje Cells. Front Mol Neurosci, 2018. 11 p. 438. 3. Carnevale, N.T.a.H., M.L. The NEURON Book. and U.C.U.P. Cambridge, 2006. 4. Kuroda, S., N. Schweighofer, and M. Kawato, Exploration of signal transduction pathways in cerebellar long-term depression by kinetic simulation. J Neurosci, 2001. 21(15): p. 5693-702
A major challenge to understand how neurons process synaptic input is to build detailed biophysical models of neuronal function that integrate morphology, electrophysiology, and biochemical reactions. The Purkinje cell is the principal neuron of the cerebellar cortex. Modeling of Purkinje cell electrophysiology spans more than 60 years of efforts [1]. This neuron contains voltage activated calcium conductances as well as calcium activated potassium conductances. These conductances are distributed over the soma, axonal initial segment, and a complex dendritic tree. As such, there is a need to accurately and efficiently model intracellular calcium diffusion. Furthermore, calcium is a second messenger essential for the activation of biochemical reactions involved in the expression of long-term synaptic plasticity [2]. In all models of Purkinje cells, calcium diffusion is assumed to be exclusively a radial process. However, synaptic plasticity in the granule cell to Purkinje cell synapse requires calcium influx through membrane conductances in conjunction with calcium released from intracellular stores. The synapses are on passive dendritic spines. Since voltage activated calcium conductances are only on the dendrite, but not on the spine, there is a need to model axial diffusion between these compartments to study synaptic plasticity in full Purkinje cell models. In this project we will describe our efforts to integrate the reaction-diffusion (RXD) NEURON [3] module into a highly detailed model of a Purkinje cell. The first study aims to reproduce the normal excitability of the cell under current clamp conditions. The second implements axial diffusion in the spiny dendrites and 3D diffusion in the smooth dendrites and soma compartments for computational efficiency. The third study looks at axial diffusion between the dendrite and a passive spine head after activation of the climbing fiber input. Finally, we implement a reduced model where a dendritic segment has one spine. The spine contains the biochemical reactions involved in the expression of long-term depression (LTD) [4]. We will describe the technical challenges, advantages and disadvantages of each implementation compared to traditional methods. Our study will be a platform for all those interested in using realistic reaction-diffusion models with morphologically complex neuronal models.
References 1. Bower, J.M., The 40-year history of modeling active dendrites in cerebellar Purkinje cells: emergence of the first single cell "community model". Front Comput Neurosci, 2015. 9 p. 129. 2. Zamora Chimal, C.G. and E. De Schutter, Ca(2+) Requirements for Long-Term Depression Are Frequency Sensitive in Purkinje Cells. Front Mol Neurosci, 2018. 11 p. 438. 3. Carnevale, N.T.a.H., M.L. The NEURON Book. and U.C.U.P. Cambridge, 2006. 4. Kuroda, S., N. Schweighofer, and M. Kawato, Exploration of signal transduction pathways in cerebellar long-term depression by kinetic simulation. J Neurosci, 2001. 21(15): p. 5693-702
In this workshop, we will explore how animal brains solve problems, and how AI can take inspiration from biological systems that have evolved specifically to solve problems flexibly and rapidly, and to adapt over the lifetime of an individual, whilst being computation- and energy-efficient. Topics will include computational models to explore how brain circuits can solve problems and how this creates new hypotheses to explore experimentally. We will also examine how this can contribute to solving problems in autonomous robotics and provide inspiration for other AI applications.
Schedule: Daniel Yasumasa Takahashi, Federal University of Rio Grande do Norte Thomas Nowotny, University of Sussex Renan Moioli, Federal University of Rio Grande do Norte Rachael Stentiford, University of Sussex Marcelo Bussotti Reyes, Universidade Federal do ABC (UFABC)
Professor of Informatics, University of Sussex, UK
I do research in computational neuroscience and bio-inspired AI. More details are on my home page http://users.sussex.ac.uk/~tn41/ and institutional homepage (link above).
Associate Professor, Centre for Complex Systems, The University of Sydney
My research focusses on studying the dynamics of information processing in biological and bio-inspired complex systems and networks, using tools from information theory such as transfer entropy to reveal when and where in a complex system information is being stored, transferred and... Read More →
My research is mainly driven by the aim of enhancing the capability of information theory in studying complex systems. Currently, I'm focusing on introducing novel approaches to recently established areas of information theory such as partial information decomposition (PID). My work... Read More →
