Model overview
==============

[Vogels et al 2011](http://www.sciencemag.org/content/334/6062/1569.short) provides an extension of the model described in [Vogels and Abbott 2005](http://www.jneurosci.org/content/25/46/10786.short) mainly in order to evaluate the role of inhibitory synapse plasticity in recovering an asynchronous network state after a manifestation of some memory pattern (i.e., highly correlated activity of some neuron group).

A brief video illustrating model features can be found [here](http://lcn.epfl.ch/~vogels/index_movie.html).

Downloading and running a simulation
====================================

To get local clone of this project [Install Git](http://www.opensourcebrain.org/projects/gitintro/wiki/Wiki), go to the directory in which the project will be cloned and type:

> git clone https://github.com/OpenSourceBrain/VogelsEtAl2011.git

In order to install PyNN see http://neuralensemble.org/trac/PyNN/wiki/Installation. Make sure to use the latest v0.8 version from [GitHub](https://github.com/NeuralEnsemble/PyNN), which includes the required Vogels2011Rule synapse. Note: NEURON is the only supported simulator for this model at the moment.

Go to the PyNN subdirectory:

> cd VogelsEtAl2011/PyNN

To run a simulation type:

> python vogelsEtAl2011.py neuron

Version of the project
======================

A Matlab code using the original model implementation can be found [here](http://senselab.med.yale.edu/ModelDB/ShowModel.asp?model=143751).

The project described here is still under development and NEURON is the only supported simulator for this model at the moment.

Our aim is to simulate the model using the protocol that generates Figure 4 in [Vogels et al 2011](http://www.sciencemag.org/content/334/6062/1569.short) obtaining a result matching with the original. However, there are difficulties associated with the required simulation load and the specification of some model parameters.

The simulation of the full network using the original protocol times is quite time consuming using conventional PCs and a single processor core. In order to handle that, we added to the main script a parameter called “downscaleFactor” by which each simulation time is divided. In addition, in an attempt to interfere the less possible in the results we rescaled the learning rate of the inhibitory synapses using the same “downscaleFactor”.

The figure below shows the results of a simulation of the full network with downscaleFactor equal to 100. The subsequent table shows the description of each state in the simulation.


table{border:1px solid black}.
{background:\#ddd}. |*=.State|*=.Description|\_=.Original simulation time|
|=. **pre**|The asynchronous irregular (AI) network dynamics of the model published in Vogels and Abbott (2005) without inhibitory plasticity|1 min (60000 ms)|
|={background:\#eee}. **A** |{background:\#eee}. Inhibitory to excitatory synapses was turned to 0 efficacy. The network is forced out of the AI regime and begins to fire at high rates.
Inhibitory plasticity is turned on.|{background:\#eee}. Not specified|
|=. **B** |Inhibitory plasticity has restored AI dynamics|60 min (60\*60\*1000 ms)|
|={background:\#eee}. **C** |{background:\#eee}.The excitatory non-zero weights of the two designated memory patterns are increased ad-hoc by a factor of 5.
The neurons of the subset begin to exhibit elevated and more sychronized activity|{background:\#eee}.5 sec (5000 ms)|
|=. **D** |Inhibitory plasticity has succesfully suppressed elevated activity from the pattern and restored the global background state|60 min (60\*60\*1000 ms)|
|={background:\#eee}. **E** |{background:\#eee}.By delivering an additional, 1 s long stimulus as described above to 25% of the cells within one memory pattern, the whole pattern is activated.
Activity inside the pattern stays asynchronous and irregular, and the rest of the network, including the other pattern, ramains nearly unaffected|{background:\#eee}.5 sec (5000 ms)|