Latent state-space models for neural decoding.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Motor cortex ensembles exhibit low-dimensional collective dynamics.
• Traditional neural decoding uses the full neuronal population, which can be computationally intensive.

Purpose of the Study:

• To investigate neural decoding using estimated low-dimensional dynamics instead of the full neuronal population.
• To assess the feasibility of simplifying neural decoding through dimensionality reduction.

Main Methods:

• Employed a latent state-space model (SSM) to estimate low-dimensional neural dynamics from spiking activity.
• Utilized a second state-space model with a Kalman filter to decode kinematics from the estimated dynamics.
• Applied the approach to neuronal recordings from a monkey performing 3-D reach and grasp movements in the primary motor cortex.

Main Results:

• Decoding performance for 3-D reaches using estimated low-dimensional dynamics was comparable to decoding using the full neuronal population.
• The latent SSM-based decoding approach effectively captured essential neural information for movement decoding.

Conclusions:

• Decoding neural activity from estimated low-dimensional dynamics is a viable and efficient alternative to using the full population.
• This method offers a promising approach for simplifying neural decoding in motor cortex research.