Multivariate autoregressive modeling and granger causality analysis of multiple spike trains

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Time-Series Analysis

Background:

• Simultaneous recording of neural activity from large neuronal populations is now possible using microelectrode arrays and optical methods.
• Multivariate time-series analysis methods are powerful for continuous neural signals (e.g., EEG) but not well-adapted for point processes like neural spike trains.

Purpose of the Study:

• To adapt existing multivariate time-series analysis methods for the analysis of multiple neural spike train data.
• To develop a novel framework for fitting hidden multivariate autoregressive models to neural spike train data.
• To apply Granger causality analysis to extract directed information flow patterns in simulated neural networks.

Main Methods:

• Utilized recent findings on correlation distortions in multivariate Linear-Nonlinear-Poisson (LNP) spiking neuron models.
• Derived generalized Yule-Walker-type equations for fitting hidden multivariate autoregressive (MAR) models.
• Applied the new framework to perform Granger causality analysis on simulated spiking neuron data.

Main Results:

• Successfully derived generalized Yule-Walker-type equations for fitting hidden MAR models to point process data.
• Demonstrated the application of this framework for Granger causality analysis in simulated neural networks.
• Provided insights into the directed information flow patterns within these networks.

Conclusions:

• The developed method offers a novel approach to analyze multivariate neural spike train data by adapting time-series analysis techniques.
• This framework extends the applicability of Granger causality analysis to point process data, enabling the study of directed information flow in neural networks.
• The study discusses the advantages and limitations of the new method for analyzing complex neural data.