Combining Multiple Functional Connectivity Methods to Improve Causal Inferences.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Network Science

Background:

• Brain function relies on network interactions, necessitating causal analysis.
• Current functional connectivity (FC) methods often use bivariate associations, leading to false positives due to unaddressed confounders.
• Simplicity of methods like Pearson correlation and coherence contributes to their widespread use despite limitations.

Purpose of the Study:

• To identify a simple functional connectivity (FC) method that enhances causal inferences compared to existing approaches.
• To address limitations of bivariate and partial correlation methods in accurately inferring causal interactions in neural networks.

Main Methods:

• Evaluated partial correlation against bivariate correlation using neural network simulations.
• Developed a novel combined functional connectivity (combinedFC) method integrating bivariate and partial correlation measures.
• Released a computational toolbox for implementing the combinedFC method.

Main Results:

• Partial correlation improved causal inferences over bivariate correlation but produced false positives with colliders.
• The proposed combinedFC method demonstrated more valid causal inferences than either bivariate or partial correlation alone.
• CombinedFC is applicable to both resting-state and task-based functional magnetic resonance imaging (fMRI) and electrophysiology data.

Conclusions:

• CombinedFC offers a more robust and accurate approach to inferring causal interactions in brain networks.
• The developed toolbox facilitates the adoption of improved FC methods for studying brain function.
• This work advances the field of network neuroscience by providing a more reliable tool for causal inference.