A Bayesian supervised dual-dimensionality reduction model for simultaneous decoding of LFP and spike train signals

Area of Science:

• Computational Neuroscience
• Neuroimaging Analysis
• Machine Learning in Neuroscience

Background:

• Neuroscience research increasingly utilizes multimodal data (e.g., EEG, fMRI, LFP, spike trains).
• Different data types offer complementary insights into neural phenomena.
• Joint modeling of local field potentials (LFP) and spike trains is crucial for comprehensive understanding.

Purpose of the Study:

• To present a novel Bayesian method for neural decoding using joint LFP and spike train data.
• To infer behavioral and experimental conditions from multimodal neural recordings.
• To develop a model for supervised dual-dimensionality reduction of neural data.

Main Methods:

• Implemented a hierarchical Bayesian model integrating LFP and spike train data.
• Utilized exponential PCA for spike train dimensionality reduction and wavelet PCA for LFP dimensionality reduction.
• Incorporated a Bayesian binary regression module for outcome prediction.

Main Results:

• The proposed hierarchical model demonstrated superior predictive performance compared to models using single data modalities.
• The model effectively performs prediction, parametric inference, and identifies influential predictors.
• Two methods for modeling the loading matrix showed comparable performance.

Conclusions:

• Joint modeling of LFP and spike train data offers enhanced predictive power in neural decoding.
• The developed Bayesian method provides a robust framework for analyzing multimodal neural data.
• Model parameters and posterior distributions offer valuable scientific insights into neural processes.