Latent inputs improve estimates of neural encoding in motor cortex

Area of Science:

• Neuroscience
• Computational Neuroscience
• Motor Control

Background:

• Tuning curves in motor cortex traditionally link neural firing rates to observed actions, assuming motor command equals motor act.
• Perturbations (external or internal) and adaptation can decouple motor intent from actual movement, complicating tuning curve interpretation.
• Existing methods struggle to accurately estimate tuning curves when neural representations diverge from observed actions.

Purpose of the Study:

• To develop a novel method for inferring latent, unobserved inputs into neuronal populations.
• To improve the estimation of neural tuning curves under conditions of perturbation and during motor learning.
• To differentiate between changes in neural input and processing during motor adaptation.

Main Methods:

• Developed a method to infer latent inputs driving neuronal populations.
• Applied the method to neural data from nonhuman primates in brain-computer interface (BCI) tasks.
• Utilized BCI learning experiment data with intentionally incorrect decoding.

Main Results:

• Tuning curves derived from inferred latent directions showed a better fit than those based on actual movements.
• The method successfully differentiated aspects of motor adaptation in a BCI learning paradigm.
• Demonstrated the utility of latent input estimation for understanding neural processing changes.

Conclusions:

• Inferring latent inputs provides a more accurate representation of neural tuning curves, especially during perturbations and adaptation.
• This approach offers a powerful tool for dissecting the mechanisms underlying motor learning and neural plasticity.
• The method advances our understanding of how the motor cortex represents and adapts movement intentions.