Scaffold modelling captures the structure-function-dynamics relationship in brain microcircuits

Abstract

Modelling brain networks with complex configuration and cellular properties requires a set of neuroinformatic tools and an organized staged workflow. We have therefore developed the Brain Scaffold Builder (BSB), a new modeling framework embedding multiple strategies for cell placement and connectivity and a flexible management of cellular and network mechanisms. With BSB, for the first time, the mouse cerebellar cortex was reconstructed and simulated at cellular resolution, using morphologically realistic multi-compartmental single-neuron models. Embedded connection rules allowed BSB to generate the cerebellar connectome, unifying a collection of scattered experimental data into a coherent construct. Naturalistic background and sensory-burst stimulation were used for functional validation against recordings in vivo, monitoring the impact of subcellular mechanisms on signal propagation and spatio-temporal processing and providing a new ground-truth about circuit organization for the prediction of neural dynamics.

Cross-comparison of state of the art neuromorphological simulators on modern CPUs and GPUs using the Brain Scaffold Builder

“…Single cell benchmarks were also run on a single thread on an AMD Ryzen 7 3700X 8-Core Processor, 3593 MHz, integrating the mechanisms into a functional unit with validated electrophysiology (Masoli et al, 2020, 2021; Rizza et al, 2021; Masoli and D’Angelo, 2017) 5 . The Arborize 6 package was used to write cell model descriptions in a single format that can construct equivalent cells in both Arbor and NEURON.…”

“…We adapted a catalogue of 33 NEURON NMODL mechanisms 1 to the Arbor dialect 2 . The mechanisms 3 were collected from the granule, Golgi, Purkinje, basket and stellate cell models 4 used in the model of the cerebellar cortex (Masoli et al, 2020, 2021; Rizza et al, 2021; Masoli and D’Angelo, 2017).…”

“…We performed benchmarks for two sizes of cerebellar cortex model: a small model with 1, 000 cells and 10, 000 synapses, and a large model with 30, 000 cells and 1.5 million synapses (De Schepper et al, 2021). Each type of benchmark was repeated ten times for 1000 ms with an integration timestep of 0.025 ms.…”

“…The network models were constructed using the methods described in De Schepper et al (2021), resizing the network description to obtain suitable estimated computational loads for the benchmarks in terms of cell and synapse number. A BSB-to-Arbor adapter was developed which can create Arbor’s functional recipes .…”

“…In order to have more scalable simulation software, it is therefore important to develop a robust benchmarking strategy that allows insight into specific computational bottlenecks for models of realistic size and complexity. In this study, we demonstrate the use of the Brain Scaffold Builder (BSB; De Schepper et al, 2021) as a framework for performing such benchmarks. We perform a comparison between the well-known neuromorphological simulator NEURON (Carnevale and Hines, 2006), and Arbor (Abi Akar et al, 2019), a new simulation library developed within the framework of the Human Brain Project.…”