GitHub - ModelDBRepository/245563: Parameter optimization using CMA-ES (Jedrzejewski-Szmek et al 2018)

ModelDBRepository / 245563 Public

Notifications You must be signed in to change notification settings
Fork 1
Star 0

Parameter optimization using CMA-ES (Jedrzejewski-Szmek et al 2018)

modeldb.science/245563

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
ajustador-1.1		ajustador-1.1
moose_nerp-1		moose_nerp-1
D1_voltagecalcium.png		D1_voltagecalcium.png
D2_voltagecalcium.png		D2_voltagecalcium.png
Figure_1.png		Figure_1.png
Figure_6.png		Figure_6.png
Voltage-1.png		Voltage-1.png
Voltage.png		Voltage.png
readme.html		readme.html

Repository files navigation

<html>
<p>This is the readme for the model and methods associated with the paper:
</p>
<p>Jedrzejewski-Szmek Z, Abrahao KP, Jedrzejewska-Szmek J, Lovinger DM, Blackwell KT (2018) Parameter Optimization Using Covariance Matrix Adaptation-Evolutionary Strategy (CMA-ES), an Approach to Investigate Differences in Channel Properties Between Neuron Subtypes. Front Neuroinform 12:47
doi: <a href="https://doi.org/10.3389/fninf.2018.00047">https://doi.org/10.3389/fninf.2018.00047</a>
</p>
<p>The model and method are available in both modeldb and github however the associated data is only on github.  In modeldb the model is in this archives mosse-nerp folder and the method is in the ajustador folder (optimization of python code).
Please see the readme's in those folders.
</p>
Example model run:<p/>
cd Jedrzejewski-SzmekEtAl2018/moose_nerp-1<br/>
python -m moose_nerp.d1d2<p/>
after a minute or two will display the following figures:<p/>
<img src="./Figure_1.png" alt="Figure_1.png" width="550"><p/>
<img src="./Voltage.png" alt="Voltage.png" width="550"><p/>
<img src="./Voltage-1.png" alt="Voltage-1.png" width="550"><p/>
<img src="./D1_voltagecalcium.png" alt="D1_voltagecalcium.png" width="550"><p/>
<img src="./D2_voltagecalcium.png" alt="D2_voltagecalcium.png" width="550"><p/>
<img src="./Figure_6.png" alt="Figure_6.png" width="550"><p/>
	
<p/> Below excerpts from the paper indicate the contents of the github repositories.  Please see the paper for more details.
<hr/>
<p>The parameters are
optimized using the covariance matrix adaptation evolutionary
strategy (<a href="https://github.com/CMA-ES/pycma">https://github.com/CMA-ES/pycma</a>).
</p>
<p>To facilitate reproducibility, re-use and extension, the declarative
model specification uses a modular format. The ion channel
kinetics are specified in one file: (<a href="https://github.com/neurord/moose_nerp/blob/master/moose_nerp/d1d2/param_chan.py">https://github.com/neurord/moose_nerp/blob/master/moose_nerp/d1d2/param_chan.py</a>

</p>
<p>Both the morphology file (either standard GENESIS .p files
or .swc files are supported by MOOSE) and conductances (in
units of Siemens/m2) are specified in a separate file (<a href="https://github.com/neurord/moose_nerp/blob/master/moose_nerp/d1d2/param_cond.py">https://github.com/neurord/moose_nerp/blob/master/moose_nerp/d1d2/param_cond.py</a>)
</p>
<p>Explicit spines, calcium dynamics, and synaptic channels are
each optional and specified in separate files. Calcium dynamics
can be specified either with a single time constant of decay or
utilizing various mechanisms such as calcium buffers, pumps
and diffusion. Model stimulation, creation of output elements
and model simulation are clearly and explicitly separated from
the model creation. Parameter specification files are imported in
<a href="https://github.com/neurord/moose_nerp/blob/master/moose_nerp/d1d2/__init__.py">https://github.com/neurord/moose_nerp/blob/master/moose_nerp/d1d2/__init__.py</a>

</p>
<p>The
complete model specification is available at <a href="https://github.com/neurord/moose_nerp/">https://github.com/neurord/moose_nerp/</a>
, with moose_nerp/d1d2 specifying the SP
model parameters and moose_nerp/gp specifying the GPe model
parameters.
</p>
<p>The experimental data used
for the optimizations are part of the python package waves,
available at <a href="https://github.com/neurord/waves">https://github.com/neurord/waves</a>. The data consists
of recordings from identified external globus pallidus neurons
and unidentified striatal spiny projection neurons. As the data
was collected for other purposes, the current injection protocol
was implemented only once per neuron.
</p>
<p>Feature extraction
functions are specified in <a href="https://github.com/neurord/ajustador/blob/master/ajustador/features.py">https://github.com/neurord/ajustador/blob/master/ajustador/features.py</a>
, and they are combined into
a fitness function in <a href="https://github.com/neurord/ajustador/blob/master/ajustador/fitnesses.py">https://github.com/neurord/ajustador/blob/master/ajustador/fitnesses.py</a>.
</p>
<p>Features are further described in
the online documentation (<a href="https://neurord.github.io/ajustador/features.html">https://neurord.github.io/ajustador/features.html</a> and <a href="https://neurord.github.io/ajustador/fitnesses.html">https://neurord.github.io/ajustador/fitnesses.html</a>

</p>
<p>For the simulations reported, the initial values
were conductances from a roughly hand-tuned model to start
the optimization in an area that exhibits spiking behavior
(Supplementary Figure 1, also available at <a href="https://github.com/neurord/ajustador/tree/master/FrontNeuroinf/FigSuppl_initialconditions.jpg">https://github.com/neurord/ajustador/tree/master/FrontNeuroinf/FigSuppl_initialconditions.jpg</a>

</p>
<p>Experimental data can be specified using one of two different
file formats: Igor binaries or comma separated values. The
traces for the experiments are placed in a separate subdirectory,
e.g., gpedata_experimental, and the class Param in the python
package waves (<a href="https://github.com/neurord/waves">https://github.com/neurord/waves</a>) specifies the
onset and offset time of the injection current, as well as the time
frame for measuring baseline membrane potential and steady
state depolarization. Since the data specification is a separate
module, adding support for other file formats is straightforward.
</p>
<p>The python package ajustador is available at <a href="https://github.com/neurord/ajustador">https://github.com/neurord/ajustador</a>
, and the scripts used to run the
simulations are at <a href="https://github.com/neurord/ajustador/tree/master/FrontNeuroinf">https://github.com/neurord/ajustador/tree/master/FrontNeuroinf</a>.
</p>
<p>The convergence was determined from
the change in mean fitness: the slope of the mean fitness
across 25 generations must be &lt;0.002 and the standard
deviation of mean fitness across 25 generations must be
less than 0.06 (implemented in <a href="https://github.com/neurord/ajustador/tree/master/ajustador/helpers/converge.py">https://github.com/neurord/ajustador/tree/master/ajustador/helpers/converge.py</a>).
</p>
</html>