Task-driven neural network models predict neural dynamics of proprioception

Cell, Journal Year: 2024, Volume and Issue: 187(7), P. 1745 - 1761.e19

Published: March 1, 2024

Proprioception tells the brain state of body based on distributed sensory neurons. Yet, principles that govern proprioceptive processing are poorly understood. Here, we employ a task-driven modeling approach to investigate neural code neurons in cuneate nucleus (CN) and somatosensory cortex area 2 (S1). We simulated muscle spindle signals through musculoskeletal generated large-scale movement repertoire train networks 16 hypotheses, each representing different computational goals. found emerging, task-optimized internal representations generalize from synthetic data predict dynamics CN S1 primates. Computational tasks aim limb position velocity were best at predicting activity both areas. Since task optimization develops better during active than passive movements, postulate is top-down modulated goal-directed movements.

Language: Английский

---

Decoding the brain: From neural representations to mechanistic models

Cell, Journal Year: 2024, Volume and Issue: 187(21), P. 5814 - 5832

Published: Oct. 1, 2024

Language: Английский

---

Contrasting action and posture coding with hierarchical deep neural network models of proprioception

eLife, Journal Year: 2023, Volume and Issue: 12

Published: May 31, 2023

Biological motor control is versatile, efficient, and depends on proprioceptive feedback. Muscles are flexible undergo continuous changes, requiring distributed adaptive mechanisms that continuously account for the body's state. The canonical role of proprioception representing body We hypothesize system could also be critical high-level tasks such as action recognition. To test this theory, we pursued a task-driven modeling approach, which allowed us to isolate study proprioception. generated large synthetic dataset human arm trajectories tracing characters Latin alphabet in 3D space, together with muscle activities obtained from musculoskeletal model model-based spindle activity. Next, compared two classes tasks: trajectory decoding recognition, train hierarchical models decode either position velocity end-effector one's posture or character (action) identity firing patterns. found artificial neural networks robustly solve both tasks, networks' units show tuning properties similar neurons primate somatosensory cortex brainstem. Remarkably, uniformly directional selective only action-recognition-trained not trajectory-decoding-trained models. This suggests encoding additionally associated higher-level functions recognition therefore provides new, experimentally testable hypotheses how aids control.

Language: Английский

---

Acquiring musculoskeletal skills with curriculum-based reinforcement learning

Neuron, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 1, 2024

Language: Английский

---

MouseNet: A biologically constrained convolutional neural network model for the mouse visual cortex

PLoS Computational Biology, Journal Year: 2022, Volume and Issue: 18(9), P. e1010427 - e1010427

Published: Sept. 6, 2022

Convolutional neural networks trained on object recognition derive inspiration from the architecture of visual system in mammals, and have been used as models feedforward computation performed primate ventral stream. In contrast to deep hierarchical organization primates, mouse has a shallower arrangement. Since mice primates are both capable visually guided behavior, this raises questions about role computation. work, we introduce novel framework for building biologically constrained convolutional network model cortex. The structural parameters derived experimental measurements, specifically 100-micrometer resolution interareal connectome, estimates numbers neurons each area cortical layer, statistics connections between layers. This is constructed support detailed task-optimized cortex, with populations that can be compared specific corresponding brain. Using well-studied image classification task our working example, demonstrate computational capability mouse-sized network. Given its relatively small size, MouseNet achieves roughly 2/3rds performance level ImageNet VGG16. combination large scale Allen Brain Observatory Visual Coding dataset, use representational similarity analysis quantify extent which recapitulates representation Importantly, provide evidence optimizing does not improve biological beyond certain point. We distributions some physiological quantities closer observed brain after training. encourage by making code freely available.

Language: Английский

---

A computational study of how an α- to γ-motoneurone collateral can mitigate velocity-dependent stretch reflexes during voluntary movement

Proceedings of the National Academy of Sciences, Journal Year: 2024, Volume and Issue: 121(34)

Published: Aug. 8, 2024

The primary motor cortex does not uniquely or directly produce alpha motoneurone (α-MN) drive to muscles during voluntary movement. Rather, α-MN emerges from the synthesis and competition among excitatory inhibitory inputs multiple descending tracts, spinal interneurons, sensory inputs, proprioceptive afferents. One such fundamental input is velocity-dependent stretch reflexes in lengthening muscles, which should be inhibited enable It remains an open question, however, extent unmodulated disrupt movement, whether how they are limbs with numerous multiarticular muscles. We used a computational model of Rhesus Macaque arm simulate movements feedforward commands only, added reflex feedback. found that caused movement-specific, typically large variable disruptions movements. These were greatly reduced when modulating feedback (i) as per commonly proposed (but yet clarified) idealized alpha-gamma (α-γ) coactivation (ii) alternative collateral projection homonymous γ-MNs. conclude collaterals physiologically tenable propriospinal circuit mammalian fusimotor system. could still collaborate α-γ coactivation, few skeletofusimotor fibers (β-MNs) mammals, create flexible ecosystem By locally automatically regulating highly nonlinear neuro-musculo-skeletal mechanics limb, these critical low-level enabler learning, adaptation, performance via higher-level brainstem, cerebellar, cortical mechanisms.

Language: Английский

---

Encoding of limb state by single neurons in the cuneate nucleus of awake monkeys

Journal of Neurophysiology, Journal Year: 2021, Volume and Issue: 126(2), P. 693 - 706

Published: May 19, 2021

The cuneate nucleus (CN) is among the first sites along neuraxis where proprioceptive signals can be integrated, transformed, and modulated. objective of study was to characterize representations in CN. To this end, we recorded from single CN neurons three monkeys during active reaching passive limb perturbation. We found that many exhibited responses were tuned approximately sinusoidally movement direction, as has been for other sensorimotor neurons. distribution their preferred directions (PDs) highly nonuniform resembled muscle spindles within individual muscles, suggesting typically receive inputs only a muscle. also tended modestly amplified movements compared perturbations, contrast cutaneous whose not systematically different conditions. Somatosensory thus seem subject "spotlighting" relevant sensory information rather than uniform suppression suggested previously.

Language: Английский

---

A leg to stand on: computational models of proprioception

Current Opinion in Physiology, Journal Year: 2021, Volume and Issue: 22, P. 100426 - 100426

Published: March 20, 2021

Language: Английский

---

Acquiring musculoskeletal skills with curriculum-based reinforcement learning

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 25, 2024

Efficient musculoskeletal simulators and powerful learning algorithms provide computational tools to tackle the grand challenge of understanding biological motor control. Our winning solution for inaugural NeurIPS MyoChallenge leverages an approach mirroring human skill learning. Using a novel curriculum approach, we trained recurrent neural network control realistic model hand with 39 muscles rotate two Baoding balls in palm hand. In agreement data from subjects, policy uncovers small number kinematic synergies even though it is not explicitly biased towards low-dimensional solutions. However, by selectively inactivating parts signal, found that more dimensions contribute task performance than suggested traditional synergy analysis. Overall, our work illustrates emerging possibilities at interface physics engines, reinforcement neuroscience advance

Language: Английский

---

Clinical neuroscience and neurotechnology: An amazing symbiosis

iScience, Journal Year: 2022, Volume and Issue: 25(10), P. 105124 - 105124

Published: Sept. 16, 2022

In the last decades, clinical neuroscience found a novel ally in neurotechnologies, devices able to record and stimulate electrical activity nervous system. These technologies improved ability diagnose treat neural disorders. Neurotechnologies are concurrently enabling deeper understanding of healthy pathological dynamics system through stimulation recordings during brain implants. On other hand, neurosciences not only driving neuroengineering toward most relevant issues, but also shaping neurotechnologies thanks advancements. For instance, etiology disease informs location therapeutic stimulation, way patterns should be designed more effective/naturalistic. Here, we describe cases fruitful integration such as Deep Brain Stimulation cortical interfaces highlight how this symbiosis between neurotechnology is closer integrated framework than simple interdisciplinary interaction.

Language: Английский

---