Neural simulation and modelling

36 views
Skip to first unread message

Dilawar Singh

unread,
Jan 2, 2015, 8:18:20 AM1/2/15
to wncc...@googlegroups.com
Neurons are extremely complicated entities. They are approximated by various
models: a. Integrate and Fire Neuron, Hodgin-Huxeley comparment Models + Rall's
cable theory with active channels, Izhikevich Models, etc. etc.. (see
http://www.scholarpedia.org/article/Category:Models_of_Neurons for more
details).

For many people neuron is a black-box with spikes-in and spikes-out transfer
function with some randomness thrown in; such people are more into exploring
applications of neural networks in machine-learning, image processing,
neuromorphic computations etc. etc. etc.. For others, especially those who
study memory and computation in brain; neuron is a hugely complex entity with
synapse, ion-channels, and dendrites (parts of neurons) doing complex signalling
at scales from few nano-meters to centimeter range, and few microseconds to
second scales.

To get a grip over such complexity, people are trying to model web of such
complex phenomenon across scales: nano-seconds to days, nano-meters to meter.
Thought time and spatial locations are usually employed scales, one can
definitely think of any abstract scale. Different disciplines -- physics,
biology, environment sciences -- takes on multiscale modelling might differ in
practise; which is natural given that this is an emerging domain.

Various tools exists for neural simulations: genesis (legacy), neuron (most
popular), python-brian (easiest to use and slowest), MOOSE are to name some.
Genesis is almost abandoned now, Neuron is actively built and used. MOOSE was
the fork of genesis which is now a complete standalone simulator with
capabilities to do fine-grain simulator of synapses. It can work with GPU and
openMPI. MOOSE comes with python bindings for scripting and a alpha GUI for
chemical modelling.

If someone wants to use or try out MOOSE, details along with documentation can
be found here http://moose.ncbs.res.in .

Packages can be found at following places:

1. For ubuntu (>= 12.04 (LTS))

$ sudo -E add-apt-repository ppa:dilawar/moose
$ sudo apt-get update
$ sudo apt-get install moose-gui moose-python

You should be able to `import moose` into python. Package moose-gui is bulky and
can be dropped if interested only in python bindings.

For RPM checkout the Open Build Service page of moose:

http://software.opensuse.org/download.html?project=home%3Amoose&package=moose-all

Bugs can be reported at SourceForge: https://sourceforge.net/projects/moose/
A git fork is also available : https://github.com/dilawar/moose

PS: If someone wants to make his/her work available as debian/rpm packages,
OpenBuildService is a great place to host them for testing and sharing.

--
Dilawar
Grad. Student
Bhalla Lab, NCBS Bangalore
Reply all
Reply to author
Forward
0 new messages