Conversion to Neosim

Converting a model to use Neosim

This document is about the conversion of an initialy c++ programmed model into a Neosim module. It explains the difference between the original source code and the new source code that uses Neosim interfaces as well as the new programming approach of the model.

Contents

The new programming approach of the model
Specific features
The NeuroML parametrisation
Running the simulation

How to use NeuroML to parametrize a simulation :

The best is to use the sample file as a template.
The steps to describe a simulation are :

defining the entities
defining the network
- the populations
- the projections
defining the source and destination methods
defining the simulation's controls
defining the external modules used

Defining the entities


<neuroml class="NeuromlPkg" description="Slow wave sleep cortical network" author="Maxim Bazenhov" keywords="hodgkin-huxley cortical network sleep wave">
   <models>
      <model name="our_cells" class="CellPkg">
         <solutions>
         </solutions>
         <sectionProperties>
         </sectionProperties>
         <neurons>
            <neuron name="CXcell" class="ModuleNeuron">
              <module module="Maximjar">
		 <parameters class="maxim.CXIN_Neosim" Cell-class="CX_Neosim"
                  Cell-type="CX"
                  G_Ampa="0.00015"
                  G_AmpaMin="0.00015"
                  G_Nmda="0.00001"
                  G_GabaA="0.00005"
                  G_GabaB="0.00000"
                  G_Ext="-0.010"
                  h="0.02"/>
	      </module>
            </neuron>
         </neurons>
         <structures>
         </structures>
      </model>
   </models>
</neuroml>

name : name of the entity in the list. This name will be the one to which populations refer
class : NeuroML provides different templates to define neurons. the one we use here is "ModuleNeuron" which defines simple neurons using external code module with specific parameters
module : this is the link to the module code. It should refer to a module name in the module list
parameters : the tag <parameters> defines a list of custom parameters the user can freely specify. Those parameters will be given to the module coding the entity, so they must reflect the actual entity parameters.

Defining the network


<neuroml class="NeuromlPkg" description="Slow wave sleep cortical network" author="Maxim Bazenhov"
	  keywords="hodgkin-huxley cortical network sleep wave">
   <models>
      <model name="the_networks" class="NetworkPkg">
         <networks>
            <network name="cortical_net">
               <elements>
               </elements>
               <projections>
               </projections>
            </network>
         </networks>
      </model>
   </models>
</neuroml>

The tag <networks> is a list of networks, each network being described by a <network ... > tag.
In the current model we define only one network. The attributes of a network are :

elements : a list of elements (populations and entities)
projections : a list of projections

Defining a network is then defining its populations and projections :

defining the populations
defining the projections

Defining the populations


<neuroml class="NeuromlPkg" description="Slow wave sleep cortical network" author="Maxim Bazenhov" keywords="hodgkin-huxley cortical network sleep wave">
   <models>
      <model name="the_networks" class="NetworkPkg">
         <networks>
            <network name="cortical_net">
               <elements>

                  <element name="ExtcellA" elementClass="Extcell">
                     <structure class="Grid1DStructure" size="1"/>
                  </element>

                <element name="CXcells" elementClass="CXcell">
                     <structure class="Grid3DStructure" xsize="5" ysize="1" zsize="1"/>
                  </element>

               </elements>
               <projections>
               </projections>
            </network>
         </networks>
      </model>
   </models>
</neuroml>

The list of elements can contain as many populations as wanted. The attributes of a element are :

name : name of the population, to which the projections will refer
elementClass : reference to the element's name
structure : the structure of a population

The structure of a population can be of many kind according to the needs.
Here are shown two different kinds of structures used in the current model : the first one just specifies its number of elements without any spatial constraints, whereas the second one is an 3D structure able to emulate a three dimensional array.

The attributes of the Grid1DStructure structure are :

class : the class of the structure
size : the number of elements

The attributes of the Grid3DStructure structure are :

class : the class of the structure
xsize,ysize,zsize : the number of elements per dimension

Defining the projections


<neuroml class="NeuromlPkg" description="" author="" keywords="">
   <models>
      <model name="the_networks" class="NetworkPkg">
         <networks>
            <network name="cortical_net">
               <elements>
               </elements>
               <projections>

                  <projection name="ExtToIN"
		     class="ModuleProjection" source="ExtcellB" dest="INcells">
                     <sourceMethod module="modules.toAllSrc">
   	                <parameters class="tutorial.sourcemethod.SpikeSourceMethod" srcPort="out"/>
                     </sourceMethod>
                     <destMethod module="modules.toAllDest">
                      <parameters class="tutorial.destmethod.SpikeDestMethod" dstPort ="in" delay="2.0"/>                 
                     </destMethod>
                  </projection>

                  <projection name="CXToCX"
		     class="ModuleProjection" source="CXCells" dest="CXCells">
                     <sourceMethod module="modules.Sphere3DSrc">
                        <parameters class="projection.Sphere3DSrc"
		            srcPort="out"
                            autoconnect="false"
                            boundmethod="mirror"
                            radius="2"
                            xrange="2"
                            yrange="1"
                            zrange="1"/>
                     </sourceMethod>
                     <destMethod module="modules.Sphere3DDst">
                        <parameters class="projection.Sphere3DDest"
                          dstPort="in"
			  delay="2.0"
			  Cell-type="CX"
                     </destMethod>
                  </projection>

               </projections>
            </network>
         </networks>
      </model>
   </models>
</neuroml>

The projections to be defined are also part of a list, called projections.
We use here the default projection class which allow to target a source and a destination method, to which a list of parameters will be given to properly create the connections.

The attributes of the ModuleProjection are :

name : name of the projection
source : reference to the source population name
dest : reference to the destination population name
sourceMethod : definition of the source method
destMethod : definition of the destination method

Defining the destination and source methods

module : reference to the name of the code module link.
parameters : the method's parameters. These parameters are exaclty those of the code module coding the projection methods.

toAllSrc and toAllDest : connect each cell to every cells of the destination population
Sphere3DSource and Shere3DDest : connects each cell to a spherical area of destination cells. This module has been specilay written to fit the current model needs.

Defining the simulation's controls


<neuroml class="NeuromlPkg" description="Slow wave sleep cortical network" author="Maxim Bazenhov" keywords="hodgkin-huxley cortical network sleep wave">
   <models>
   </models>
   <environment>
      <io>
      </io>
      <schedule class="ControlledSchedule" module="Controljar">
         <parameters class="maxim.MaximControl"
            pop1name="CXcells"
            pop1ODEnumber="13"
            pop2name="INcells"
            pop2ODEnumber="13"
            pop3name="ExtcellA"
            pop4name="ExtcellB"
            time="150.0"
            scaletime="10.0"
            printtime="2000.0"
            filetime="1000.0"/>
      </schedule>
   </environment>
   <modules>
   </modules>
</neuroml>

The simulation environment is defined in this section.
The environement attributes are :

inputs : list of non-biological entities used to send intput into the model.
outputs : list of non-biological probes used to get outputs from the model.
schedule : control of the simulation's run.

The schedule's default class allows only to set the simulation duration. The schedule class used here is a ControlledSchedule giving the control to a code module. The attributes of the ControlledSchedule are:

class : Name of the schedule class used. ControlledSchedule is not the default class for schedules and therefore must be explicitly named.

module : reference to the name of the code module link.

parameters : parameters of this control module, of basic class SimpleParameter. For the details of the actual parameters used here, see MaximControl module.

Defining the external modules used


<neuroml class="NeuromlPkg" description="" author="" keywords="">
   <models>
   </models>
   <environment>
   </environment>
   <modules>
      <module name="toAllSrc" class="SourceMethodModule"
            location="http://www.anc.ed.ac.uk/~paulro/jars/projToAllSrc.jar"/>
      <module name="toAllDest" class="DestMethodModule"
            location="http://www.anc.ed.ac.uk/~paulro/jars/projToAllDst.jar"/>
      <module name="Sphere3DSrc" class="SourceMethodModule"
            location="http://www.anc.ed.ac.uk/~paulro/jars/proj3DSphereSrc.jar"/>
      <module name="Sphere3DDst" class="DestMethodModule"
            location="http://www.anc.ed.ac.uk/~paulro/jars/proj3DSphereDst.jar"/>
      <module name="Maximjar" location="http://www.anc.ed.ac.uk/~paulro/jars/maxim.jar"/>
      <module name="Controljar" class="ControlModule" location="http://www.anc.ed.ac.uk/~paulro/jars/mcontrol.jar"/>
   </modules>
</neuroml>

The actual paths for code modules are not explicitely given everytime a module has to be specified.
Instead a reference is given, targeting a module name placed in a module list.

The list of modules contain the modules name and location.