Classification-Based Prediction of Effective Connectivity Between Timeseries With a Realistic Cortical Network Model

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Effective connectivity analysis aims to infer causal interactions between brain regions.
• Traditional methods include non-parametric and parametric approaches.
• Biophysically plausible neural network models offer mechanistic insights but are challenging to invert for effective connectivity estimation.

Purpose of the Study:

• To develop a classification-based method to approximate the inversion of complex biophysical neural network models for effective connectivity estimation.
• To improve the accuracy of inferring causal interactions from multivariate time series data.

Main Methods:

• A classification-based approach was proposed to predict causal interaction patterns from multivariate time series.
• The classifier was trained using simulated data from a biophysically plausible neural network model.
• The method was evaluated in simulated experiments against current best-practice techniques.

Main Results:

• The proposed classification-based method demonstrated significantly higher accuracy in detecting causal structures compared to existing methods.
• The study validated the generative neural network model and the classifier's adaptability to different data-generating models.

Conclusions:

• The classification-based method provides a more accurate and potentially more interpretable approach to estimating effective connectivity from neural data.
• This method offers a promising solution for overcoming the challenges of inverting complex biophysical models for neural systems analysis.