Random forest Granger causality for detection of effective brain connectivity using high-dimensional data

Area of Science:

• Neuroscience
• Computational Biology
• Network Science

Background:

• Brain functions rely on complex interconnected networks, not isolated regions.
• Granger causality (GC) with ordinary least squares (OLS) is standard for network analysis.
• Existing OLS limitations include accuracy, precision, and false discovery rate (FDR) issues.

Purpose of the Study:

• To propose Random Forest as a novel regularization technique for Granger causality.
• To improve the accuracy, precision, and FDR of brain network analysis.
• To apply the enhanced GC method to map networks involved in deductive reasoning.

Main Methods:

• Implemented Random Forest as a regularization method for Granger causality computation.
• Compared the proposed Random Forest GC with Least Absolute Shrinkage and Selection Operator (LASSO) and Elastic-Net regularized GC.
• Validated the methodology using simulated datasets and the real StarPlus dataset for deductive reasoning.

Main Results:

• The Random Forest approach demonstrated improved accuracy and precision over LASSO and Elastic-Net.
• The proposed method showed a reduced false discovery rate in network analysis.
• Successfully mapped brain networks associated with deductive reasoning using the new technique.

Conclusions:

• Random Forest offers a superior regularization strategy for Granger causality analysis.
• This advancement enhances the reliability of brain network mapping.
• The method provides a more precise tool for understanding cognitive processes like deductive reasoning.