BAND: Behavior-Aligned Neural Dynamics is all you need to capture motor corrections.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Motor Control

Background:

• Motor cortical activity is characterized by latent neural population dynamics, with preparatory activity explaining movement variability.
• Preparatory activity alone cannot account for real-time movement corrections guided by sensory feedback.

Purpose of the Study:

• To investigate how sensory feedback influences motor cortex activity during unplanned movement adjustments.
• To develop a novel computational model for analyzing neural dynamics during both planned and unplanned movements.

Main Methods:

• Introduction of the Behavior-Aligned Neural Dynamics (BAND) model, utilizing semi-supervised learning.
• Analysis of neural population trajectories and their deviations during movement execution.
• Comparison of BAND's capabilities against unsupervised inference methods.

Main Results:

• Sensory-guided movement corrections manifest as transient deviations from autonomous neural trajectories.
• These corrective dynamics are often subtle, encoded within small neural variability.
• A sparse sub-population of primary motor cortex (M1) neurons is involved in encoding motor corrections.
• The BAND model successfully captures both planned movement trajectories and unplanned corrections.

Conclusions:

• Combining latent dynamical modeling with behavioral supervision is crucial for a comprehensive understanding of motor control.
• The BAND model offers a powerful tool for dissecting neural mechanisms underlying adaptive motor behavior.
• Neural variability and sparse coding play significant roles in real-time motor adjustments.