Connectionists: CNS*14 Tutorial on Neuronal Model Parameter Search Techniques

[Cengiz Günay]

Dear colleagues,

You are invited to attend our shortly upcoming full-day tutorial session at
CNS*2014 in Quebec City.

*Where:* Organization for Computational Neurosciences (OCNS) Conference,
Quebec City Conference Center, Room 2105
*When:* Saturday July 26, 2014 between 9am - 4:30pm


*T9: Neuronal Model Parameter Search Techniques*


Parameter tuning of model neurons to mimic biologically realistic activity
is a non‐trivial task. Multiple models may exhibit similar dynamics that
match experimental data – i.e., there is no single “correct” model. To
address this issue, the ensemble modeling technique proposes to
represent properties
of living neurons with a set of neuronal models. Several approaches to
ensemble modeling have been proposed over the years, but the two most
prevalent parameter tuning methods are systematic “brute‐force” searches
[1, 2] and various evolutionary algorithms‐based techniques [3, 4, 5, 6].
Both approaches relay on traversing a very large parameter space (with
thousands
to millions of model instances), but utilize diametrically different ways
to accomplish that. In both cases, however, entire collections of
biologically realistic models are generated, whose neural activity
characteristics can then be cataloged and studied using a database [1, 2].

The tutorial covers “tips and tricks,” as well as various pitfalls in all
stages of model construction, large‐scale simulations on high performance
computing clusters [S2], database construction and analysis of neural data,
along with a discussion about the strengths and weaknesses of the two parameter
search techniques. We will review software implementations for each
technique: PANDORA Matlab Toolbox [7][S1] for the brute force method and
NeRvolver (i.e., evolver of nerve cells) for evolutionary algorithms.
PANDORA was used in recent projects for tuning models of rat globus
pallidus neurons [2][M1], lobster pyloric network calcium sensors [8][M2],
leech heart interneurons [9][M3,S3] and hippocampal O‐LM interneurons
(Skinner Lab, TWRI/UHN and Univ. Toronto). NeRvolver is a prototype of a
computational intelligence‐based system for automated construction, tuning,
and analysis of neuronal models, which is currently under development in
the Computational Intelligence and Bio (logical) informatics Laboratory at
Delaware State University [10]. Through the utilization of computational
intelligence methods (i.e., Multi‐Objective Evolutionary Algorithms and
Fuzzy Logic), the NeRvolver system generates classification rules
describing biological  phenomena discovered during the process of model
creation or tuning. Thus in addition to producing neuronal models,
NeRvolver provides–via such rules–insights into the functioning of the
biological
neurons being modeled. In the tutorial, we will present basic
functionalities of the system and demonstrate how to analyze the results
returned by the software.

We will allocate enough time for Q&A and if participants bring a laptop
pre‐loaded with Matlab, they can follow some of our examples.

*Lecturers/Organizers:*

Cengiz Günay, Anca Doloc­-Mihu (Emory University, USA)

Vladislav Sekulić (University of Toronto, Canada),

Tomasz G. Smolinski (Delaware State University, USA)



*References *
[1]   Astrid A. Prinz, Cyrus P. Billimoria, and Eve Marder. Alternative to
hand‐tuning conductance‐based models: Construction and analysis of
databases of model neurons. J Neurophysiol, 90:3998–4015, 2003.

[2]   Cengiz Günay, Jeremy R. Edgerton, and Dieter Jaeger. Channel density
distributions explain spiking variability in the globus pallidus: A
combined physiology and computer simulation database approach. J.
Neurosci., 28(30):7476–91, July 2008.

[3]   Pablo Achard and Erik De Schutter. Complex parameter landscape for a
complex neuron model. PLoS Comput Biol, 2(7):794–804, Jul 2006.

[4]   Tomasz G. Smolinski and Astrid A. Prinz. Computational intelligence
in modeling of biological neurons: A case study of an invertebrate
pacemaker neuron. In Proceedings of the International Joint Conference on
Neural Networks, pages 2964–2970, Atlanta, GA, 2009.

[5]   Tomasz G. Smolinski and Astrid A. Prinz. Multi‐objective evolutionary
algorithms for model neuron parameter value selection matching biological
behavior under different simulation scenarios. BMC Neuroscience, 10(Suppl
1):P260, 2009.

[6]   Damon G. Lamb and Ronald L. Calabrese. Correlated conductance
parameters in leech heart motor neurons contribute to motor pattern
formation. PLoS One, 8(11):e79267, 2013.

[7]   Cengiz Günay, Jeremy R. Edgerton, Su Li, Thomas Sangrey, Astrid A.
Prinz, and Dieter Jaeger. Database analysis of simulated and recorded
electrophysiological datasets with PANDORA’s Toolbox. Neuroinformatics,
7(2):93–111, 2009.

[8]   Cengiz Günay and Astrid A. Prinz. Model calcium sensors for network
homeostasis: Sensor and readout parameter analysis from a database of model
neuronal networks. J Neurosci, 30:1686–1698,
Feb 2010. NIHMS176368,PMC2851246.

[9]   Anca Doloc‐Mihu and Ronald L. Calabrese. A database of computational
models of a half‐center oscillator for analyzing how neuronal parameters
influence network activity. J Biol Phys, 37(3):263–283, Jun 2011.

[10]   Emlyne Forren, Myles Johnson‐Gray, Parth Patel, and Tomasz G.
Smolinski. Nervolver: a computational intelligence‐based system for
automated construction, tuning, and analysis of neuronal models. BMC
Neuroscience, 13(Suppl 1):P36, 2012.
-Cengiz

--
Cengiz Gunay
Postdoctoral Fellow, Dept. of Biology
Visiting Faculty, Dept. of Math & CS
Emory University
cgunay at emory.edu
http://www.biology.emory.edu/research/Prinz/Cengiz/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20140724/cc6c6f8c/attachment.html>