Riassunto analitico
Introduction
In silico brain modeling is going to provide to the medical community innovative tools to investigate brain activity: from electrophysiological patterns in physiological and pathological conditions to pharmacodynamics and predictive neurosurgery. Unfortunately the remarkable complexity of the brain has limited the development of highly resolved models. The methods currently used for this purpose largely vary according to the desired level of detail. At a small scale, the focus is on the activity of individual cells, at a medium scale, on neuronal microcircuits, and at a large scale, on the activity of entire regions or eventually on the whole brain. Furthermore, two modeling approaches are typically adopted: bottom-up, which starts from elementary phenomena to replicate the complexity of certain brain functions, and top-down, which seeks to infer the fundamental causes of observable phenomena. In this work I illustrate a bottom-up method for modeling a medium-scale network of the regions CA1 and CA3 of the mouse hippocampus, describing its validation and potential applications to human.
Methods
The methodological approach relies on an algorithm that creates a network of single compartment neuronal models (point neurons) starting from morpho-anatomical data taken from literature. Neuronal classes are identified by studying their morphologies, which are then represented using abstracted geometric shapes, called probability clouds. The putative connections between neurons are estimated by intersecting their probability clouds in 3D space. The computed connections are then pruned to fit literature data. This process results in a connectivity map used to create a network of point neurons and connect them together. Finally, the electrical activity of the network is simulated on a supercomputer and validated according to literature data.
Results
The algorithm successfully modeled the CA1 and CA3 regions. Each brain region is constituted by approximately 300k neurons and one billion synapses. Adopting a realistic morpho-anatomical connection strategy, both networks were built using neuronal and synaptic models available on NEST simulator, a neuroinformatic tool developed to build and simulate neural circuit activity. The CA1 network was simulated on the Piz-Daint supercomputer at the Swiss National Supercomputer Center. In the simulation, the electrical activity of the neurons propagated in accordance with the literature, thus validating the model.
Discussion and Conclusion
This method, starting from detailed anatomical-morphological description allowed to faithfully reproduce the neuronal activity of the hippocampus . The adaptation of this model to the human hippocampus could be included in large-scale models such as The Virtual Brain to improve the resolution of specific areas involved in the generation of neurological diseases. This work has thus the ambition of creating personalized brain models employed to i) simulate epileptic activity, ii) predict the consequences of drug application, iii) estimate lesions pathophysiology or iv) simulate neurosurgical interventions therefore generating advanced tools for personalized medicine.
|
Abstract
Introduction
In silico brain modeling is going to provide to the medical community innovative tools to investigate brain activity: from electrophysiological patterns in physiological and pathological conditions to pharmacodynamics and predictive neurosurgery. Unfortunately the remarkable complexity of the brain has limited the development of highly resolved models. The methods currently used for this purpose largely vary according to the desired level of detail. At a small scale, the focus is on the activity of individual cells, at a medium scale, on neuronal microcircuits, and at a large scale, on the activity of entire regions or eventually on the whole brain. Furthermore, two modeling approaches are typically adopted: bottom-up, which starts from elementary phenomena to replicate the complexity of certain brain functions, and top-down, which seeks to infer the fundamental causes of observable phenomena. In this work I illustrate a bottom-up method for modeling a medium-scale network of the regions CA1 and CA3 of the mouse hippocampus, describing its validation and potential applications to human.
Methods
The methodological approach relies on an algorithm that creates a network of single compartment neuronal models (point neurons) starting from morpho-anatomical data taken from literature. Neuronal classes are identified by studying their morphologies, which are then represented using abstracted geometric shapes, called probability clouds. The putative connections between neurons are estimated by intersecting their probability clouds in 3D space. The computed connections are then pruned to fit literature data. This process results in a connectivity map used to create a network of point neurons and connect them together. Finally, the electrical activity of the network is simulated on a supercomputer and validated according to literature data.
Results
The algorithm successfully modeled the CA1 and CA3 regions. Each brain region is constituted by approximately 300k neurons and one billion synapses. Adopting a realistic morpho-anatomical connection strategy, both networks were built using neuronal and synaptic models available on NEST simulator, a neuroinformatic tool developed to build and simulate neural circuit activity. The CA1 network was simulated on the Piz-Daint supercomputer at the Swiss National Supercomputer Center. In the simulation, the electrical activity of the neurons propagated in accordance with the literature, thus validating the model.
Discussion and Conclusion
This method, starting from detailed anatomical-morphological description allowed to faithfully reproduce the neuronal activity of the hippocampus . The adaptation of this model to the human hippocampus could be included in large-scale models such as The Virtual Brain to improve the resolution of specific areas involved in the generation of neurological diseases. This work has thus the ambition of creating personalized brain models employed to i) simulate epileptic activity, ii) predict the consequences of drug application, iii) estimate lesions pathophysiology or iv) simulate neurosurgical interventions therefore generating advanced tools for personalized medicine.
|
