In a recent study published in SaleMarian Vargas and Bessie et al.1 present an innovative method to unravel the computational principles underlying proprioceptive processing in nonhuman primates. Their findings demonstrate the utility of task-based modeling in advancing neuroscience and offer translational potential by providing fundamental insights into the goals and mechanisms by which the brain encodes body position and movements. .
Proprioception allows us to sense the position and movement of our body parts and is crucial for motor control and coordination, such as when reaching for a light switch in the dark. Proprioceptive signals originate from specific mechanoreceptors in muscles, tendons, and joints, and pass through the dorsal column-medial limbus pathway. Within this pathway, the cuneate nucleus (CN) plays an important role in the processing of sensory information from the upper limbs and trunk. It then directs this information via the thalamus to both the primary (S1) and secondary somatosensory cortices. In these cortical areas, proprioceptive signals are integrated with other sensory information, usually unconsciously forming our perception of body position and movement. Despite this understanding, the exact mechanisms involved in proprioception remain unclear. Specifically, what are the computational goals of the proprioceptive pathway, and how does it encode proprioceptive signals to support these goals?
Marin Vargas and Bisi et al. Address these questions through advanced computational modeling. Artificial neural networks have become powerful tools for studying neural processing in both sensory and motor pathways.2,3 These models not only achieve high predictive accuracy but also offer deep insight into the computational principles underlying neural responses. By training these networks on different tasks and comparing the learned representations to actual neural activity, researchers can explore the specific functions that these neural responses may represent, potentially leading to new understanding of neural processing mechanisms. can be opened.4
Building on this concept, Marin Vargas and Bessi et al. developed a qualitative framework to uncover the computational principles underlying the proprioceptive pathway. Using a multi-pronged strategy, they integrated several techniques: (i) simulation of proprioceptive inputs through advanced muscle modeling, (ii) hypotheses representing distinct targets of proprioceptive processing. based on improving neural network models, and (iii) predicting neural activity in CN and S1. of monkeys performing active and passive arm movements (Fig. 1).
Standard framework. Standard framework for revealing the computational principles of the proprioceptive pathway. Neural network models were trained on a large dataset of synthetic proprioceptive data, each optimized for specific hypothetical tasks, and the cuneate nucleus of active and passive performing monkeys. (CN) and was tested by predicting experimental data recorded from primary somatosensory cortex (S1). Arm movements Portions of this figure were created with BioRender.com.
Training appropriate models of proprioception requires a diverse and extensive repertoire of movements and their associated muscle spindle signals, which are difficult to obtain due to the anatomical location of the proprioceptive pathway. To address this challenge, Marin Vargas and Bessi et al. generated a large-scale dataset of simulated muscle spindle signals using a sophisticated three-dimensional muscle arm model. This dataset provided the necessary foundation for training thousands of neural networks based on different architectures and learning algorithms, each designed to efficiently model time series data.
Based on this primary dataset, the authors used these neural networks to implicitly test candidate hypotheses. They reviewed 16 separate hypotheses that have been proposed over decades of research, covering areas such as kinematic state estimation, action recognition, sensorimotor control, and efficient coding. Each hypothesis was formulated as a computational objective for which a set of neural networks was specifically trained.
To test which task optimization most accurately reflects proprioceptive processing in the brain, the authors evaluated their models to predict extracellular electrophysiological recordings in the CN and S1. Perform both active and passive arm movements in an off-center reaching paradigm. . Critically, task-optimized models were able to generalize from synthetic to experimental proprioceptive data, validating the effectiveness of this approach. Furthermore, these models outperformed several control models, including classical linear encoding models and data-driven neural networks trained directly on experimental proprioceptive data, in predicting neural signal dynamics.
This task-based modeling approach yielded a number of insights into the computational mechanisms underlying proprioception, highlighting several important findings. First, models fitting kinematic state estimation of limb position and velocity were most effective in predicting neural activity in CN and S1, illustrating the importance of these coding signals in the proprioceptive pathway. Second, a positive correlation was observed between the models' effectiveness in solving computational tasks and their predictive accuracy with real neural data, including model architecture and task optimization in developing brain-like representations. The role of Third, the task-optimized models significantly outperformed the randomly triggered untrained models during active movements, but not during passive movements, suggesting top-down potentials of CN and S1 during voluntary movements. Suggest modulation. Fourth, despite their hierarchical anatomical organization, both CN and S1 were best described by deeper layers of the model. Together, these results illustrate that kinematic state estimation is a fundamental computational goal of the proprioceptive pathway and reveal important factors in how proprioceptive signals are processed.
Despite the computational tour-de-force, several questions remain open: (i) although the task-optimized models were more effective in describing neural activity than passive movement in CN and S1, it remains unclear whether How models can represent passives. Movements (ii) The experiments limited the monkeys' workspace to small movements, raising questions about whether the results generalize to larger, more complex workspaces. (iii) Proprioceptive signals are usually combined with other sensory signals such as visual and tactile information. Modeling the integration of these diverse inputs remains an unresolved challenge.
The final question may be the most speculative: Where might all this lead us? The normative framework introduced by Marin Vargas and Bisi et al. Holistic sensory processing has the potential to expand our understanding of the integration of multiple sensory domains, such as proprioception, vision, and touch. Additionally, the study holds the promise of potentially disruptive developments in the field of neuroprosthetics. Despite significant advances in controlling robotic arms, until recently, these movements lacked a key feature: sensory feedback. While direct stimulation now allows contact recreation,5 Effectively simulating the corresponding proprioceptive sensation remains a challenge. The work of Marin Vargas and Bisi et al. represents an important step towards this goal.