Testing non-linearity and directedness of interactions between neural groups in the macaque inferotemporal cortex

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Visual cortex information processing relies on complex, context-dependent neuronal interactions.
• Existing methods for functional connectivity analysis often assume linearity or instantaneous interactions, which may not accurately reflect neural dynamics.

Purpose of the Study:

• To present a general framework for linear and non-linear modeling of neurophysiological time series data using Local Linear Non-linear Autoregressive (LLNAR) models.
• To introduce novel statistical tests for assessing non-linearity in time series and directed neural interactions.
• To generalize the concept of Granger causality for both linear and non-linear systems.

Main Methods:

• Development of the Local Linear Non-linear Autoregressive (LLNAR) modeling framework.
• Application of bootstrap techniques to compare the goodness-of-fit for linear versus non-linear models.
• Illustration using artificial, reference, and macaque local field potential (LFP) data.

Main Results:

• LLNAR models effectively capture both linear and non-linear aspects of neurophysiological time series.
• A new test for non-linearity in directed neural interactions was successfully developed and applied.
• Analysis of macaque LFP data showed that the linear variant of LLNAR adequately describes the data.
• Models incorporating lagged values (25-60 ms) identified both uni- and bi-directional influences between recording sites.

Conclusions:

• The LLNAR framework provides a robust method for analyzing complex neural interactions, overcoming limitations of previous linear or instantaneous models.
• The developed tests offer a statistically sound approach to quantify non-linearity and directedness in neural communication.
• The findings demonstrate the presence of both linear and non-linear influences in visual cortex functional connectivity.