Multi-Region Markovian Gaussian Process: An Efficient Method to Discover Directional Communications Across Multiple...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Statistical Modeling

Background:

• Understanding neural communication across brain regions is vital in neuroscience.
• Existing methods like Gaussian Processes (GP) and Linear Dynamical Systems (LDS) have limitations in capturing complex brain interactions.
• GPs excel at identifying latent variables and frequency bands, while LDS offer computational efficiency but limited expressiveness.

Purpose of the Study:

• To develop a novel statistical framework that integrates the strengths of both GP and LDS models.
• To create a model capable of explicitly representing frequencies and phase delays in neural data.
• To enable efficient and interpretable analysis of multi-region brain communication.

Main Methods:

• We propose the Multi-Region Markovian Gaussian Process (MRM-GP), a novel model structured as a Linear Dynamical System that mirrors a multi-output Gaussian Process.
• This approach establishes a direct link between LDS and multi-output GP formalisms.
• The MRM-GP is designed to operate with a linear inference cost over time points.

Main Results:

• The MRM-GP successfully models frequencies and phase delays within the latent space of neural recordings.
• The model provides an interpretable, low-dimensional representation of neural activity.
• We demonstrate the model's ability to reveal communication directions between brain regions and separate oscillatory communications into distinct frequency bands.

Conclusions:

• The MRM-GP offers a powerful and computationally efficient method for analyzing complex neural communication patterns.
• This integrated approach enhances the interpretability of latent representations in multi-region brain recordings.
• Our findings advance the statistical toolkit for investigating brain connectivity and dynamics across different frequency spectra.