Latent Dynamic Factor Analysis of High-Dimensional Neural Recordings.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• High-dimensional neural recordings offer precise spatial and temporal insights into brain function.
• Establishing functional connectivity between brain regions is crucial for understanding information flow.

Purpose of the Study:

• To introduce Latent Dynamic Factor Analysis of High-dimensional time series (LDFA-H), a novel method for analyzing complex neural data.
• To improve the estimation of cross-correlations among latent variables in high-dimensional time series, reflecting inter-regional information transfer.

Main Methods:

• Developed a new approach for estimating covariance structure in high-dimensional time series.
• Extended probabilistic canonical correlation analysis (CCA) to dynamic time series for latent variable analysis.
• Applied LDFA-H to local field potential (LFP) recordings from Prefrontal Cortex (PFC) and visual area V4.

Main Results:

• LDFA-H successfully identifies underlying factors even when noise obscures correlations.
• Simulations demonstrated superior performance compared to existing methods in capturing target factors.
• The method revealed time-varying lead-lag dependencies and spatial distributions between PFC and V4 signals during a memory-guided saccade task.

Conclusions:

• LDFA-H provides a robust framework for uncovering dynamic functional connectivity from high-dimensional neural recordings.
• The findings highlight the utility of LDFA-H in characterizing inter-regional information flow in the brain.
• This method advances the analysis of neural dynamics and brain region interactions.