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

Area of Science:

• Neuroscience
• Computational Neuroscience
• Statistical Modeling

Background:

• Understanding brain region interactions is vital in neuroscience.
• Existing statistical methods like Gaussian Processes (GP) and Linear Dynamical Systems (LDS) offer different strengths for analyzing neural communication.
• GP models excel at discovering latent variables, frequency bands, and communication directions, while LDS models are computationally efficient but less expressive.

Purpose of the Study:

• To merge the strengths of GP and LDS methodologies for enhanced analysis of multi-region brain communication.
• To develop a novel model, the Multi-Region Markovian Gaussian Process (MRM-GP), that bridges the gap between LDS and multi-output GP.
• To enable efficient and interpretable analysis of neural recordings, revealing communication dynamics across brain regions.

Main Methods:

• Developed a Multi-Region Markovian Gaussian Process (MRM-GP) model.
• The MRM-GP mirrors a multi-output Gaussian Process using a Linear Dynamical System framework.
• Established a theoretical connection between LDS and multi-output GP, explicitly modeling frequencies and phase delays in the latent space.

Main Results:

• The MRM-GP model achieves linear inference cost over time points.
• The model provides an interpretable low-dimensional representation of neural data.
• Successfully revealed communication directions between brain regions and separated oscillatory communications into distinct frequency bands.

Conclusions:

• The MRM-GP model offers a powerful and efficient approach to analyzing complex brain communication patterns.
• This integrated methodology enhances the expressiveness of LDS models while maintaining computational efficiency.
• The model's ability to identify frequency-specific communication pathways provides valuable insights into neural dynamics.