Application Practicalities

The application is freely available and details can be found at http://www.neuroConstruct.org. It is a desktop based application written in Java, and so can be run on a wide range of machine architectures and operating systems. Automatic installers are available for Windows, Mac OS and Linux. The main interface is shown in Figure 5.1A. The network models are organized into projects containing all of the constituent elements (cell types, 3D region information, network connections, etc.). While networks can be built and analyzed through the interface alone, the simulation is run on one of the existing simulators. The script files generated by neuroConstruct are in the native language of the chosen simulation platform, making them open to modification and refinement by experienced users.

FIVE MAIN FUNCTIONALITY AREAS The functionality of neuroConstruct is divided into five main areas (see Figure 5.1B).

Importation and validation of morphologies

Many biologically realistic models of neurons use reconstructed morphologies (De Schutter and Bower, 1994; Mainen et al., 1995; Migliore et al., 1995; Rapp et al., 1996; Destexhe and Pare, 1999; Vetter et al., 2001; Poirazi et al., 2003; Schaefer et al., 2003; Hanson et al., 2004; Golding et al., 2005; Jarsky et al., 2005) and public databases have been produced which contain examples of these in a number of different formats (Cannon et al., 1998; Ascoli, 2006). neuroConstruct supports importation of morphological data files in various formats including Neurolucida's *.asc files (Glaser and Glaser, 1990), GENESIS readcell compatible format (*.p), most NEURON/ntscable based morphology files (*.nrn or *.hoc) and Cvapp (*.asc) format (Cannon et al., 1998). Morphology files can also be imported and exported in MorphML format (Crook et al., 2007), which is part of the NeuroML standards, as outlined later. The imported files can be validated and checked for errors or artifacts such as zero length or diameter segments of the dendritic tree, which commonly occur during reconstruction. When network models with large numbers of cells are to be generated, representations of cells with reduced numbers of compartments are often used to speed simulations while preserving some aspects of cell region

Neuroconstruct

neuroConstruct

Importation and Validation of Morphologies

Network Generation

Simulation Management

Importation and Validation of Morphologies

Network Generation

Simulation Management

Automatic export to external analysis packages

Igor Pro/ NeuroMatic

FIGURE 5.1 Application interface and main functionality. (A) The main interface to neuroConstruct, showing a visualization of a network model of the cerebellar granule cell layer in 3D, with a single Golgi cell highlighted to display its pre- and postsynaptic targets. (B) The main functional areas of the application. Modified from Gleeson et al., 2007, with permission. (See Plate 1 in color plate section.)

Automatic export to external analysis packages

Igor Pro/ NeuroMatic

FIGURE 5.1 Application interface and main functionality. (A) The main interface to neuroConstruct, showing a visualization of a network model of the cerebellar granule cell layer in 3D, with a single Golgi cell highlighted to display its pre- and postsynaptic targets. (B) The main functional areas of the application. Modified from Gleeson et al., 2007, with permission. (See Plate 1 in color plate section.)

FIGURE 5.2 Connectivity schemes possible in neuroConstruct. (A) Morphology based connection. Connections can be made between specified subregions on the presynaptic granule cell (i) and postsynaptic Purkinje cells (ii). Various other parameters can be set including maximum/minimum length of connections and number of pre-/postsynaptic connections allowed per cell. The connections are shown as green (presynaptic) and red (postsynaptic) spheres connected by lines. (iii) (B) Volume based connection. A region can be specified around presynaptic cells where axonal connections can potentially be made (i). Postsynaptic cells have regions specified associated with this type of connection (ii). Intersections between the 3D axonal regions and permitted dendritic sections can lead to connections, in line with the other rules on numbers of connections per cell (iii). Reproduced from Gleeson et al., 2007, with permission. (See Plate 2 in color plate section.)

specificity of synaptic connections and inhomogeneous channel density (Santhakumar et al., 2005; Traub et al., 2005). These more abstract cells can be created manually in neuroConstruct (e.g. see Figure 5.2). For all types of cells, groups of cell sections can be defined, allowing specification of axons, apical and basal dendrites, etc.

Creation of cell models

To give a cell model realistic electrophysiological behavior, mechanisms representing voltage and ion concentration dependent conductances must be added to the cell membrane. This is a core feature of packages like NEURON and

GENESIS and distinguishes them from more abstract neural network simulators. Models of ion channel conductances, as well as internal ion concentration and synaptic mechanisms, can be created in neuroConstruct in one of two ways. Script files in the native language of one of the supported simulators can be reused, allowing simulation only on that platform, or the parameters for the mechanism can be specified in a simulator independent format (ChannelML, see Section on methodologies below), which can be mapped onto the language of the simulators. Electrophysiological cell models in neuroConstruct can be created by specifying the conductance densities of channel mechanisms at various locations on the cell membrane. Passive properties of cells can also be set (e.g. axial resistance and specific capacitance) as can the potential location for synaptic connections on different regions of the cell.

Network generation

Once a number of cells have been added to the project, these can be arranged in 3D space according to a number of packing patterns. These can vary from abstract arrangements to cells packed in a random manner, avoiding somata of other cells and reproducing closely the measured anatomical cell densities (see Section on methodologies below). Connections can be made between cell groups taking into account allowed synaptic locations for each cell and a number of other factors such as maximum and minimum connection length, number of synapses per pre-/postsynaptic cells, etc. Generated networks can be analyzed visually (see Figure 5.1A) and there are a number of inbuilt tools to display connection length distributions, synaptic convergence/divergence ratios, etc. Electrical stimulation of various forms can be applied to subsets of cells of the network.

Simulation management

The simulator independent representation of the cell and network structures can be mapped onto script files for instantiating the models in NEURON or GENESIS. Additional parameters such as simulation duration, time step and method of numerical integration are specified through the interface, as are lists of variables to record or plot during the simulation (e.g. membrane potentials, synaptic conductances, internal ion concentrations). Simulations are initiated through the interface and run on standard versions of the simulators. There is no interaction between neuroConstruct and the individual simulators during the simulation run. Data from the simulation are saved in a simple text format and there is a simulation browsing interface in neuroConstruct which lists recorded simulations.

Network analysis

The simulation results can be loaded back into neuroConstruct for analysis. The network behavior can be visualized by replaying the simulation with, for example, cells color coded according to membrane potential or spiking frequency. Plots of individual cell traces can be displayed and there are extra functions present for network wide analysis, e.g. generation of rasterplots of cell groups, spiking histograms, correlation functions between individual cells and other cells in the same group, etc. Importation of simulation data into numerical analysis packages such as MATLAB and IGOR Pro is also possible.

+1 0

Post a comment