Higher-order correlations in non-stationary parallel spike trains: statistical modeling and inference

Area of Science:

• Neuroscience
• Computational Neuroscience
• Statistical Analysis

Background:

• Higher-order correlations in neural spiking activity are crucial but difficult to study.
• Existing analysis tools require large datasets, limiting research with multi-neuron recordings.
• Current methods struggle with time-varying firing rates in neural populations.

Purpose of the Study:

• To develop a non-stationary version of the Cumulant-based Inference of Higher-Order Correlations (CuBIC) method.
• To adapt CuBIC to analyze neural data with time-varying firing rates.
• To improve the accuracy and applicability of higher-order correlation analysis in neuroscience.

Main Methods:

• Introduced a non-stationary compound Poisson process (CPP) model.
• Decoupled correlation structure from spiking intensity in the CPP model.
• Validated the adapted CuBIC method using numerical simulations.

Main Results:

• The adapted CuBIC method successfully analyzes neural data with time-varying firing rates.
• The adaptation corrects for false positives caused by rate co-variation.
• Sensitivity to true correlations is minimally affected by temporal firing rate variations.

Conclusions:

• The non-stationary CuBIC method enhances the analysis of higher-order neural correlations.
• This advancement allows for more robust investigation of neural population dynamics.
• It overcomes limitations of previous methods, enabling research with shorter data stretches.