6533b831fe1ef96bd12994d9

RESEARCH PRODUCT

Computational analysis of a multi-varied biomimetic neural network

subject

description

In recent years, astrocytes have emerged as such elements involved in inter-neuronal communication that they are increasingly considered in the synaptic coupling connecting neurons and synapses, thus constituting tripartite synapses. Many studies have focused on this astrocytic influence. With the description of the basic mechanisms of synapses, as well as the influence of gliotransmission, we emphasize that gliotransmission to extra-synaptic areas has so far been mainly studied in vivo and in vitro. We come closer to the paradigms used in AI, by proposing an analysis of the astrocyte influence on the variations of the parameters governing synaptic and extra-synaptic plasticity.We sought to apply a method that could allow us to make the link between neuronal functioning and cognition, while satisfying the constraints of computational modeling. We have thus modeled the astrocytic influence on extra-synaptic areas, by applying plasticity rules linked to the evolution of postsynaptic calcium. In this thesis work, we also propose an approach in the analysis of the relationship between synaptic activity and astrocyte activity.First, we identified the area of interest for the parameters that describe the conductance of synaptic and postsynaptic ionotropic receptors. Indeed, gliotransmission seems to target NMDA receptors, of which conductance is lower than that of AMPA receptors. In addition, by modeling the modulation of AMPAR conductance as a function of postsynaptic calcium, we demonstrate that the influence of gliotransmission can lead to increasing or attenuating the propagation of synaptic signals.We have set up a method to study the relationship between neuronal and astrocyte activity over time, which is based on the encoding of a signaling pattern that will direct the activity of neurons. The population of astrocytes that unites the synapses concerned will then come into activity. In the same way that we can analyze the activity of a population of neurons by averaging their activity frequencies, we analyzed the proportion of astrocytes in an active state. More specifically, we binarized the activity according to the astrocytic calcium threshold which conditions gliotransmission. Using this method, we can assess the relationship between astrocyte activity, neuronal activity, and the signaling pattern. By applying this process to the architecture of layers of neurons, we observe the relationship between neuronal activity and astrocyte activity increases with the depth of the population. This analysis highlights a subtle relationship between neurons and astrocytes, which can be compared to the problems encountered in artificial intelligence, more particularly to catastrophic forgetting.We conclude that one of the roles of astrocytes could be related to the methods used to fight against catastrophic forgetfulness and therefore justify the permanence of complex cognitive functions. We also highlight the difficulty of this analysis, based on the dynamics studied and on the computational resource necessary for the demonstration of the astrocytic influence.