Abstract:Understanding the complex connectivity of brain networks is essential for deciphering neural function and its disruptions in neurological disorders. Traditional graph theoretical approaches have provided valuable insights into the topological properties of neural networks, yet they often overlook the temporal dynamics intrinsic to neuronal activity. In this work, we introduce Cerebrum, a novel framework that seamlessly integrates biologically plausible Hodgkin-Huxley (HH) neuron models with Graph Neural Networks (GNNs) to simulate and infer synaptic connectivity in large-scale brain networks. Cerebrum leverages canonical network topologies: Erdös-Rényi, Small-World, and Scale-Free, to generate ground-truth connectivity matrices and employs advanced GNN architectures to accurately predict these connections from simulated neuronal activity patterns. By incorporating disease-specific parameter modifications, such as those mimicking Parkinson's disease and epilepsy, our framework enables the exploration of how pathological states influence network dynamics and connectivity inference. Additionally, we incorporate empirical synaptic data from C. elegans to enhance the biological fidelity of our simulations. Cerebrum is accompanied by an open-source computational toolkit, providing tools for neural dynamics simulation, connectivity inference, and interactive network perturbations. Our work represents a significant advancement in computational neuroscience, offering a biologically grounded machine learning framework for modeling and analyzing intricate brain network dynamics.

Biologically Plausible Graph Neural Networks for Simulating Brain Dynamics and Inferring Connectivity