Identifying Distinct Neural Features between the Initial and Corrective Phases of Precise Reaching Using AutoLFADS

Area of Science:

• Neuroscience
• Motor Control
• Computational Neuroscience

Background:

• Many movements require corrections, but the neural basis for these adjustments in the motor cortex is not fully understood.
• Existing models often assume global neural encoding for movement, which may not apply to corrective submovements.

Purpose of the Study:

• To investigate how the motor cortex signals initial and corrective movements during a precision reaching task.
• To compare the neural encoding of initial versus corrective movements and assess decoding generalization.

Main Methods:

• Recorded neural activity from a large population of neurons in two macaques during reaching.
• Utilized AutoLFADS, a deep-learning model, to analyze neural firing rates during corrective submovements.
• Developed and tested a state-dependent decoder to improve corrective movement decoding.

Main Results:

• Decoding of reach velocity from initial movements did not generalize well to corrective submovements.
• Traditional linear decoding methods struggled to predict corrective movement velocity in a global neural space.
• A state-dependent decoder significantly improved the prediction of corrective movements by incorporating initial population firing rates.

Conclusions:

• Neural activity differs significantly between initial and corrective movements, encoding specific velocity and position combinations.
• Findings challenge the assumption of global and independent neural encoding for kinematic features during online corrections.
• State-dependent decoding approaches are crucial for accurately capturing the diverse neural processes underlying corrective movements.