Discovering Low-Dimensional Descriptions of Multineuronal Dependencies

Area of Science:

• Neuroscience
• Computational Neuroscience
• Data Analysis

Background:

• Coordinated neural activity is vital for brain information processing.
• Understanding complex neural codes requires analyzing multivariate dependencies beyond simple correlations.
• Existing methods often fail to capture intricate interaction patterns in neural populations.

Purpose of the Study:

• To develop and validate a new approach for dissecting complex neural dependencies.
• To identify latent features and low-dimensional structures within neural population activity.
• To apply the method to neural recordings from mouse and macaque brains.

Main Methods:

• Utilized a C-Vine copula approach combined with normalizing flows for dependency modeling.
• Employed weighted non-negative matrix factorization to extract shared latent features from copula densities.
• Validated the methodology using simulated data and real neural recordings.

Main Results:

• Neural dependencies were found to reside in low-dimensional subspaces.
• Distinct neural modules interact synergistically, creating diverse patterns.
• The approach successfully leveraged latent features to understand population-level interactions.

Conclusions:

• The developed method effectively dissects complex neural dependencies.
• Neural population activity is organized within low-dimensional structures shaped by synergistic modular interactions.
• This work provides new insights into the neural codes underlying brain function.