Reimplementation of the Potjans-Diesmann cortical microcircuit model: from NEST to Brian

Abstract

1This work targets the replicability of computational models to provide the community with tested and proven simulator and obtain the same results presented in the reference article. We did not replicate analyses that 7 involve changes in the network structure. Our replicated network model presents activity dynamic patterns very 8 similar to the ones observed in the original model, with comparisons made in terms of firing rates and synchrony 9 and irregularity measures. In conclusion, the Potjans-Diesmann model was successfully replicated in a different 10 platform than the one in which it was originally implemented. Most theoretical studies of cortical activity are based on networks of randomly connected units [2,6,7,12] or 13 with architectures artificially built from random networks [10]. In spite of the usefulness of these models, in 14 order to understand the interplay between network structure and cortical dynamics it is essential to have 15 computational models which accurately represent the cortical network architecture. Recently, Potjans and 16 Diesmann [8] developed a network model of the local cortical microcircuit based on extensive experimental data 17 on the intrinsic circuitry of striate cortex [1,9]. The model contains two cell types (excitatory and inhibitory) 18 distributed over four layers, L2/3, L4, L5, and L6, and represents the cortical network below a surface area of 1 19 mm 2 (a scheme is shown in Fig. 1). 20The original implementation was based on the NEST simulator [4] and the source code is available at the Methods 24In this work, we replicated in Brian 2 every detail of the Potjans-Diesmann model as described in their original 25 article [8]. Hereafter, we will refer to the original NEST implementation of the Potjans-Diesmann model [8] as 26 reference (or original) article. In this section we explain how this reimplementation was done. Further statistical 27 analyses were performed using SciPy, NumPy, and Matplotlib libraries for the Python language.

Ultrafast Simulation of Large-Scale Neocortical Microcircuitry with Biophysically Realistic Neurons

“…Although several attempts have been made to reduce the demanding computational load of neuronal simulations 32, 33, 88–92 , the most widely used approach is parallelization, both at the level of single cells 93 and network models 94, 95 . In case of moderately sized 96–100 and full-scale networks 19–21, 101 this hinders the possibility of running these models on widely available computational resources, instead users and developers are forced to employ supercomputers. However, ANNs offer a possible solution to this problem.…”

“…Although several attempts have been made to reduce the demanding computational load of neuronal simulations 32,33,[88][89][90][91][92] , the most widely used approach is parallelization, both at the level of single cells 93 and network models 94,95 . In case of moderately sized [96][97][98][99][100] and full-scale networks [19][20][21]101 45 . Rett syndrome occurs in ~1:10,000 births worldwide, resulting in intellectual disability, dysmorphisms, declining cortical and motor function, stereotypies, and frequent myoclonic seizures mostly in girls [104][105][106][107][108][109][110] .…”

“…Simulation environments such as NEURON rely on compartment-specific mathematical abstractions of active and passive biophysical mechanisms 54 , which results in high computational load in increasingly complex circuit models. In the case of moderately sized [80][81][82][83][84] and full-scale networks [19][20][21]85 this hinders the possibility of running these models on non-specialized computational resources. Although several attempts have been made to reduce the demanding computational load of neuronal simulations 32,33,[86][87][88][89][90] , the most commonly used approach is parallelization, both at the level of single cells 91 and network models 92,93 .…”

Modelos de redes de neurônios para o neocórtex e fenômenos emergentes observados

“…Além das sinapses recorrentes advindas dos neurônios da rede, cada neurônio recebe entradas externas modeladas como processos de Poisson independentes com taxa ν ext = 8 Hz. Uma descrição mais detalhada da implementação computacional do modelo de Potjans-Diesmann pode ser vista na recente replicação do modelo em um simulador diferente do usado originalmente [69].…”

Modelos de redes neurais com neurônios estocásticos e diferentes topologias: construção e análise

“…The nervous system is composed of billions of neurons interacting with each other by means of about a trillion of synapses (Lent, 2004;Herculano-Houzel, Catania, Manger, & Kaas, 2015) forming several interconnected regions (Hagmann et al, 2008;Cordes et al, 2000). In particular, cortical neurons that encode similar features of the external stimuli in the primary sensory cortices can be seen as segregated in what is known as cortical microcolumns (Mountcastle, 1997;Potjans & Diesmann, 2012;Shimoura et al, 2018). These local connectivities aggregate to give rise to different cortical regions, which, by their turn, communicate among themselves by means of long range fiber projections, creating pathways by which information can be conveyed across the whole cortex.…”