Task-driven neural network models predict neural dynamics of proprioception.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Proprioception, the sense of body position and movement, relies on sensory neurons but its neural processing principles remain unclear.
• The cuneate nucleus (CN) and somatosensory cortex area 2 (S1) are critical brain regions involved in processing proprioceptive information.

Purpose of the Study:

• To investigate the neural code of proprioceptive processing in the CN and S1 using a task-driven modeling approach.
• To identify computational goals that best explain neural activity patterns related to proprioception.

Main Methods:

• Simulated muscle spindle signals using musculoskeletal modeling to create a comprehensive movement repertoire.
• Trained neural networks based on 16 distinct hypotheses representing different computational objectives.
• Validated the model's ability to predict neural dynamics in primate CN and S1 using synthetic data.

Main Results:

• Task-optimized internal representations derived from synthetic data successfully predicted neural dynamics in both CN and S1.
• Computational tasks focused on predicting limb position and velocity demonstrated the highest accuracy in explaining neural activity.
• Neural representations showed better prediction of activity during active movements compared to passive movements.

Conclusions:

• The brain's processing of proprioception is optimized for predicting limb state during active, goal-directed movements.
• Neural activity in the CN and S1 is likely modulated by top-down signals during voluntary movements.
• This study provides insights into the computational principles underlying the neural code for proprioception.