Computational Investigation of Contributions from Different Subtypes of Interneurons in Prefrontal Cortex for Information Maintenance

Abstract

Interneurons play crucial roles in neocortex associated with high-level cognitive functions; however, the specific division of labor is still under investigation. Interneurons are exceptionally diverse in their morphological appearance and functional properties. In this study, we modify a prefrontal multicolumn circuit in which five subtypes of inhibitory interneurons play distinct roles in the maintenance of transient information. These interneurons are classified according to the extending range of axonal projections. Our work simplifies the division of labor between different types of interneurons for the maintenance of information and the principle of functional redundancy of the brain from the perspective of computational modeling. This model presents a framework to understand the cooperation between different interneurons in a recurrent cortical circuit.Most of the neocortical neurons (70-80%) are excitatory neurons, and the remaining are mostly inhibitory interneurons 1,2 . Although the proportion is limited, synaptic inhibition plays a crucial role in supporting cognitive function of the cortical cortex 3,4 . Recent years have witnessed a dramatic accumulation of our knowledge about the diverse morphological, physiological, molecular, and synaptic characteristics of interneurons 5-8 . A mechanistic understanding of these interneurons and how they support cognition remains to be further demonstrated.The computational models can integrate different levels of neuroscience data, helping us to explore the unknown. We present here, a multicolumn prefrontal cortex (PFC) circuit to elucidate distinct operations carried out by diverse interneurons. We have modified the model of a detailed data-driven single prefrontal column proposed by Durstewitz and colleagues [ModelDB (http://senselab.med.yale.edu/ModelDB/)] 9 , and used the adaptive exponential integrate-and-fire (aEIF) neuron model. The sources of anatomical structure include ferret, rodent, and primate PFC experiment 10-12 . The neuron parameters in this PFC network are derived from the experimental literature related to PFC and other areas of the neocortex 8,9,13 . Therefore, this model is a PFC network based entirely on biological experimental data.The axons of interneurons usually arborize within a cortical column and can project laterally across columns 14,15 . Thus, our network incorporates five interneuron subtypes, according to the extending range of axonal projections. We explore the division of labor between different types of interneurons for information maintenance and prove that cortex network is modular 16,17 . It contains connector hubs that have connections distributed diversely across communities 18,19 . This structure can compensate for the loss of function when partial short-range connected interneurons are damaged.Understanding the circuit mechanisms of signal transmission and information maintenance associated is expected to increase our understanding of high-level cognitive functions, such as working memory. The computational models hel...

BrainCog: A Spiking Neural Network based Brain-inspired Cognitive Intelligence Engine for Brain-inspired AI and Brain Simulation

“…This change improves the efficiency of information transmission, which is consistent with the results of biological experiments [161]. This data-driven PFC column model provides an effective simulation-validation platform to study other high-level cognitive functions [162]. 2) Cortical Column: A mammalian thalamocortical column is constructed in BrainCog, which is based on detailed anatomical data [73].…”

Comparison Between Human and Rodent Neurons for Persistent Activity Performance: A Biologically Plausible Computational Investigation

“…In this model, we used many types of interneurons, across which the division of labor in information maintenance differed ( Zhang et al, 2020 ). The network is composed of numerous neurons, so changing the parameters of neurons will break the original homeostasis of the network ( Wu et al, 2020 ).…”

Brain-inspired neural circuit evolution for spiking neural networks

“…Unlike real-value-based ANNs, spiking neurons have rich spatiotemporal dynamics by receiving presynaptic information (spatial), accumulating membrane potential over time (temporal), and releasing spikes when the threshold is exceeded. Based on the above characteristics of SNNs, they outperform shallow ANNs in adversarial attack ( 13 , 14 ), noise robustness ( 15 , 16 ), continuous learning ( 17 , 18 ), and simulation of brain cognitive functions ( 19 , 20 ). However, compared with the deep ANNs, there is still a performance gap in terms of performance such as image classification ( 21 – 24 ) and object detection ( 25 ).…”