Identifying the pulsed neuron networks' structures by a nonlinear Granger causality method.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• Mapping functional connectivity in biological neural networks (BNNs) is crucial for understanding brain structure-function relationships.
• Existing methods often rely on linear models, which may not capture the complex dynamics of neural systems.

Purpose of the Study:

• To extend linear Granger causality to a nonlinear framework for identifying BNN structures.
• To develop a method for inferring information flow and functional connections in BNNs.

Main Methods:

• Utilized Radial Basis Functions to model nonlinear dynamics of neuronal pulse firing.
• Introduced presynaptic neuron contributions to predict postsynaptic neuron activity.
• Applied the Nonlinear Granger Causality Identification Method (NGCIM) to Spiking Neural Networks (SNNs).

Main Results:

• NGCIM achieved high identification accuracy across small (2-6 nodes), medium (20 nodes), and large (100 nodes) scale SNNs.
• Accuracies ranged from 100% for small networks to 80.56% for large networks.
• Performance significantly surpassed traditional Linear Granger Causality methods.

Conclusions:

• NGCIM is a promising method for network modeling and inferring functional connective maps of BNNs.
• The method's effectiveness is demonstrated on SNNs, a biologically inspired model.
• NGCIM can leverage accumulated data from methods like EEG, fMRI, and Multi-Electrode Array.