Synthetic neuronal datasets for benchmarking directed functional connectivity metrics.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biophysics

Background:

• Synthetic neural data are crucial for validating directed functional connectivity metrics.
• Debates on fMRI connectivity metrics highlight the need for realistic BOLD signal emulations.
• Benchmarking datasets require efficient generative models balancing realism and computational cost.

Purpose of the Study:

• To explore simple generative models for producing synthetic neural data for benchmarking.
• To assess the ability of these models to reflect simulated effective connectivity.
• To generate synthetic EEG and fMRI BOLD signals from these models.

Main Methods:

• Utilized autoregressive (AR) processes, neural mass models (linear/nonlinear SDEs), and spiking neuron populations.
• Employed a three-shell head model for synthetic EEG generation.
• Modeled fMRI BOLD signals using the Balloon-Windkessel model or HRF convolution.

Main Results:

• Detected modeled effective connectivity in synthetic data across varying connection strengths and delays.
• Granger causality analysis confirmed the presence of causal effects in all generated datasets.
• Less biophysically realistic models provided greater control over modeled causal relations.

Conclusions:

• Simple generative models can produce synthetic neural data with detectable causal effects for benchmarking.
• The relationship between modeled and detected causality varies across models.
• Efficient simulations are preferred for generating realistic neural dynamics for connectivity analysis.