Multistable bimodal perceptual coding within the ventral premotor cortex

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The integration of multimodal sensory information by neurons in the ventral premotor cortex (VPC) is not well understood.
• Investigating how the brain processes and combines different sensory inputs is crucial for understanding perception and decision-making.

Purpose of the Study:

• To elucidate how neurons in the VPC integrate and process multimodal sensory information.
• To compare neural dynamics in biological VPC networks with artificial recurrent neural networks (RNNs) during a bimodal detection task (BDT).

Main Methods:

• Recorded VPC neuronal activity in monkeys performing a tactile or acoustic BDT.
• Performed single-cell and population analyses of VPC network activity.
• Trained an RNN on the same BDT and analyzed its latent dynamics.
• Utilized low-dimensional modeling to understand neural trajectory transitions.

Main Results:

• Identified diverse neuronal response types: purely tactile, purely acoustic, bimodal, and decision-mnemonic.
• Observed strikingly similar latent dynamics between the VPC network and the trained RNN.
• Demonstrated that neural trajectories diverge along modality-specific paths during stimuli presentation and converge into distinct attractors for decision maintenance.
• Showcased a universal mechanism for sensory processing to memory retention transition.

Conclusions:

• Ventral premotor cortex neurons effectively encode bimodal information and integrate competing sensory inputs.
• The VPC network exhibits a dynamical motif for transitioning from sensory processing to maintaining multimodal perceptual decisions.
• The study proposes a unified mechanism explaining neural dynamics in both biological and artificial systems.