Description length guided nonlinear unified Granger causality analysis.

Area of Science:

• Neuroscience
• Network Science
• Computational Biology

Background:

• Granger causality analysis (GCA) traditionally uses two-stage schemes based on distinct mathematical theories.
• These methods can be unified as generalized model selection problems.
• Functional brain networks often exhibit nonlinear dynamics, which traditional linear GCA may not fully capture.

Purpose of the Study:

• To develop a unified Granger causality analysis (uGCA) method based on the minimum description length principle.
• To incorporate nonlinear modeling into uGCA to better analyze functional brain networks.
• To compare the performance of nonlinear uGCA against linear GCA and its linear representation.

Main Methods:

• Proposed a unified GCA (uGCA) framework grounded in the minimum description length principle.
• Integrated nonlinear modeling using Taylor expansion approximations into the uGCA method.
• Validated the approach using synthetic data, real fMRI data from mental arithmetic tasks, and patient data from autism spectrum disorder.

Main Results:

• Nonlinear uGCA significantly outperformed linear uGCA and conventional GCA on synthetic data.
• The influence of nonlinear terms diminished with increasing noise levels in both synthetic and real fMRI data.
• Causal network properties derived from uGCA showed better consistency with clinical symptoms in autism spectrum disorder patients compared to conventional GCA.

Conclusions:

• The proposed nonlinear uGCA offers a more robust and accurate method for analyzing functional brain networks.
• Linear causality analysis may suffice for fMRI data with high noise levels.
• uGCA provides more clinically relevant network insights for conditions like autism spectrum disorder.