A goal-driven modular neural network predicts parietofrontal neural dynamics during grasping.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Primate Motor Control

Background:

• The primate hand-grasping circuit, involving the anterior intraparietal area, ventral premotor cortex, and primary motor cortex, is crucial for transforming visual information into motor actions.
• A comprehensive computational model linking visual processing to grasping movements is currently lacking.

Purpose of the Study:

• To develop and validate a recurrent neural network model that computationally replicates the primate grasping circuit.
• To gain insights into the neural computations underlying visually guided grasping.

Main Methods:

• A modular recurrent neural network was designed to mimic the anatomical structure of the primate grasping circuit.
• The network was trained using visual object features to generate primate-like muscle dynamics for grasping.
• Internal network activity and lesion effects were compared to neurophysiological data from primate grasping tasks.

Main Results:

• The trained network's internal activity closely resembled neural recordings from the primate grasping circuit during object manipulation.
• Network dynamics explained different phases of the task, including maintaining movement plans and generating muscle kinematics.
• The modular models outperformed alternative computational models in explaining neural data.

Conclusions:

• The developed recurrent neural network model serves as a viable computational tool for understanding the primate grasping circuit.
• The model successfully recapitulates key aspects of neural processing and lesion effects within the grasping circuit.
• This approach offers a framework for investigating how brain regions coordinate for visually guided grasping.