%0 Conference Paper
%A van der Vlag, Michiel
%A Diaz, Sandra
%T Vast TVB parameter space exploration: A Modular Framework for Accelerating the Multi-Scale Simulation of Human Brain Dynamics
%M FZJ-2023-05175
%D 2022
%X Neural dynamics arise from the intricate multi-scale structures of the brain, where neurons communicate through synapses, forming transient assemblies that contribute to global brain dynamics. Local network activity is regulated by a complex interplay of intercellular communication, intracellular signaling cascades, and the extracellular molecular environment. Recent multi-scale models of brain function have successfully linked the emergence of global brain dynamics in both conscious and unconscious states to microscopic changes influencing local networks.Specifically, mean-field models, such as the Adaptive Exponential (AdEx) models representing statistical properties of local neuron populations, have been connected using human tractography data to simulate multi-scale neural phenomena within The Virtual Brain (TVB). While mean-field models can be run on personal computers for short simulations or on high-performance computing (HPC) architectures for longer simulations, the computational demands remain high, leaving extensive areas of the parameter space unexplored. In this work, we introduce our TVB-HPC framework, a modular set of methods designed to implement the TVB-AdEx model for GPU, enhancing simulation speed and significantly reducing computational resource requirements. This framework maintains the stability and robustness of the TVB-AdEx model, enabling more detailed exploration of vast parameter spaces and longer simulations that were previously challenging. Comparisons between our TVB-HPC framework and TVB-AdEx demonstrate the similarity in generating patterns of functional connectivity between brain regions. By varying global coupling and spike-frequency adaptation, we reproduce their interdependence in inducing transitions between dynamics associated with conscious and unconscious brain states. Exploring theparameter space further, we unveil a nonlinear interplay between spike-frequency adaptation and subthreshold adaptation, along with previously unnoticed interactions between global coupling, adaptation, and the propagation velocity of action potentials along the human connectome. As our simulation and analysis toolkits are openly accessible as open-source packages, our TVB-HPC framework serves as a versatile template for scripting other models. This approach facilitates the use of personalized datasets in the study of inter-individual variability in parameters related to functional brain dynamics. Consequently, our results present potentially influential, publicly-available methods for simulating and analyzing various human brain states.
%B JSC's End-of-Year Colloquium 2023
%C 5 Dec 2023 - 5 Dec 2023, Jülich (Germany)
Y2 5 Dec 2023 - 5 Dec 2023
M2 Jülich, Germany
%F PUB:(DE-HGF)24
%9 Poster
%U https://juser.fz-juelich.de/record/1019122