Benchmarking nonparametric Granger causality: Robustness against downsampling and influence of spectral decomposition...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Brain function relies on dynamic, directed interactions within distributed neural networks.
• Characterizing these time-varying interactions is crucial for understanding brain dynamics.
• Nonparametric Granger-Geweke causality (GGC) methods offer potential for analyzing non-stationary neural data.

Purpose of the Study:

• To systematically evaluate and benchmark various nonparametric GGC methods for analyzing time-varying directed connectivity in neural systems.
• To assess the influence of spectral decomposition parameters on wavelet and multitaper approaches.
• To determine the suitability of these methods for analyzing real neural data, specifically rat EEG during whisker stimulation.

Main Methods:

• Utilized nonparametric spectral factorization techniques, including wavelet transforms (Morlet wavelet) and multitapers with sliding time windows.
• Applied these methods to electroencephalography (EEG) data from rat cortex during controlled unilateral whisker stimulation.
• Defined performance criteria based on known somatosensory evoked potentials (SEPs) propagation patterns and latencies.

Main Results:

• Nonparametric GGC methods generally identified the contralateral primary sensory cortex (cS1) as the primary driver of the cortical network.
• Optimized Morlet wavelet approach excelled at detecting cS1's functional targets.
• Sliding-window multitaper provided superior temporal resolution for characterizing whisker-evoked interactions.
• GGC estimates demonstrated robustness against signal downsampling.

Conclusions:

• Nonparametric GGC methods are well-suited for characterizing time-varying directed causal influences in neural systems with good temporal resolution.
• The study provides practical parameter recommendations for wavelet and multitaper spectral decomposition in GGC analysis.
• These findings support the application of advanced signal processing techniques for uncovering complex brain network dynamics.