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

Weihan Li,Chengrui Li,Yule Wang,Anqi Wu

2024-05-30

Abstract:Studying the complex interactions between different brain regions is crucial in neuroscience. Various statistical methods have explored the latent communication across multiple brain regions. Two main categories are the Gaussian Process (GP) and Linear Dynamical System (LDS), each with unique strengths. The GP-based approach effectively discovers latent variables with frequency bands and communication directions. Conversely, the LDS-based approach is computationally efficient but lacks powerful expressiveness in latent representation. In this study, we merge both methodologies by creating an LDS mirroring a multi-output GP, termed Multi-Region Markovian Gaussian Process (MRM-GP). Our work establishes a connection between an LDS and a multi-output GP that explicitly models frequencies and phase delays within the latent space of neural recordings. Consequently, the model achieves a linear inference cost over time points and provides an interpretable low-dimensional representation, revealing communication directions across brain regions and separating oscillatory communications into different frequency bands.

Neurons and Cognition,Machine Learning

What problem does this paper attempt to address?

This paper discusses the Multi-region Markov Gaussian Process (MRM-GP), an efficient method for discovering directional communication between multiple brain regions. The background of this research is the crucial study of complex interactions between different brain regions in neuroscience. Existing statistical methods such as Gaussian Process (GP) and Linear Dynamic System (LDS) each have their advantages, with GP being skilled at discovering key information of latent variables such as frequency and communication direction, while LDS has higher computational efficiency at time points but weaker expressive power in latent representation.

The paper proposes to combine LDS with multi-output GP to create an LDS that reflects the MRM-GP. Through this approach, the model is able to achieve linear time point inference cost and provide interpretable low-dimensional representations, revealing the direction of communication between brain regions and separating oscillatory communication.

MRM-GP uses a complex multi-region kernel function specifically designed to learn latent interactions across brain regions that cover frequency and phase delays. The paper describes how to transform this multi-output GP into an LDS and demonstrates through experiments the ability of MRM-GP to handle multi-region spike training and local field potential recordings, generating interpretable low-dimensional representations and effectively separating oscillatory communication of different frequencies.

In summary, this paper aims to address how to analyze and interpret complex communication patterns between multiple brain regions more efficiently. By combining the advantages of GP and LDS, a new model called MRM-GP is proposed to reveal the communication direction and frequency characteristics.