Distinguishing causal interactions in neural populations

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• Understanding causal relationships in neural population activity is crucial for deciphering brain function.
• Distinguishing statistically significant causal interactions from mere correlations in complex neural networks remains a challenge.

Purpose of the Study:

• To introduce a theoretical network analysis for identifying statistically causal neuronal interactions.
• To define and investigate the concept of a "causal core" in neural activity.
• To explore how learning affects causal structures in neural networks.

Main Methods:

• Theoretical network analysis using Granger causality to assess statistical causality.
• Application of the method to Darwin X, a brain-based device with a simulated hippocampal-cortical model.
• Analysis of neuronal unit activity during a spatial navigation task.

Main Results:

• A method was developed to distinguish statistically causal interactions in population neural activity.
• The concept of a "causal core" was introduced, representing causally significant neuronal interactions.
• Analysis of Darwin X showed that large neuronal interaction repertoires contain small causal cores.
• These causal cores were observed to decrease in size during learning.

Conclusions:

• The developed network analysis can identify statistically causal interactions relevant to specific neural outputs.
• The "causal core" concept provides a framework for understanding essential causal pathways.
• The reduction of causal cores during learning in Darwin X suggests a selection of specific causal pathways from diverse neuronal repertoires.