Inhibitory Components in Muscle Synergies Factorized by the Rectified Latent Variable Model From Electromyographic...

Area of Science:

• Motor Neuroscience
• Computational Neuroscience
• Biophysics

Background:

• Non-negative matrix factorization (NMF) is standard for muscle synergy analysis from electromyographical signals (EMGs).
• NMF cannot identify negative muscle components, potentially missing inhibitory neural control.
• Inhibitory spinal interneurons play a role in motor control but are not captured by NMF.

Purpose of the Study:

• To introduce and validate the Rectified Latent Variable Model (RLVM) for muscle synergy extraction.
• To overcome NMF limitations in identifying negative muscle components.
• To explore the role of inhibitory spinal interneurons in muscle synergies.

Main Methods:

• Utilized an autoencoder neural network architecture for RLVM.
• Compared RLVM with NMF on simulated and experimental EMG datasets.
• Applied RLVM to macaque monkey upper-limb EMG and spinal premotor interneuron recordings during grasping.

Main Results:

• RLVM demonstrated superior performance in identifying muscle-synergy subspaces in simulations.
• RLVM identified small negative muscle components (NegCps) in experimental data, absent in NMF.
• RLVM's NegCps correlated with inhibitory muscle fields of spinal premotor interneurons.

Conclusions:

• RLVM is a feasible method for extracting potential inhibitory muscle-synergy components from EMGs.
• The findings suggest RLVM can reveal neural control mechanisms previously undetectable.
• This advances the understanding of motor control involving inhibitory pathways.