More advanced IntFire neuron

[steve]

I am simplifying a complex neuron model at the moment, and think I can basically replicate it for the purpose of my project, with a simple IntFire neuron with added channels. How should I best solve this problem, work from the IntFire source?

[ted]

NEURON's artificial spiking cells have neither membrane capacitance nor conductance,
so it is not possible to "add channels" to them in the usual sense. For the sake of
computational efficiency, they avoid numerical integration entirely. Instead, since the
analytical solutions to their system equations are known, the future trajectory of any
cell can be computed directly from its initial condition. This computation need only be
performed when that cell receives an event. Run time is therefore proportional to the
number of events delivered, and independent of the number of cells, number of
connections, or problem time. For further information, you may want to read chapter 10
in The NEURON Book, or at least this paper
Hines, M.L. and Carnevale, N.T. Discrete event simulation in the NEURON
environment. Neurocomputing 58-60:1117-1122, 2004
which is available from http://www.neuron.yale.edu/neuron/bib/nrnpubs.html.

That said, I should mention that many enhancements are possible beyond the three
basic classes of artificial spiking cells that are built into NEURON; Bill Lytton has
developed special classes of artificial spiking cells with more complex dynamics
for his own work, so you may wish to contact him.

--Ted

[steve]

I understand that. But maybe you could give me advice on the following:
I am researching membrane potential bistability in the visual cortex - although I am able to reproduce it with complex cells I want to simplify it as much as possible to make the simulation faster - that is why I was thinking of
inserting non-linearity in the most efficient models - the artificial neurons. So do you think my best way to achieve this is to contact Bill Lytton?

[ted]

You can "insert" any "nonlinearity" you like into an artificial spiking cell model, as
long as you can come up with an analytic solution that predicts the future course of
the model's states from their present values. A pretty tough constraint.

I am researching membrane potential bistability in the visual cortex

If the bistability depends on membrane biophysical properties (i.e. active currents)
then I don't see how artificial spiking cell representations are relevant. If it is
entirely caused by the balance between excitatory and inhibitory synaptic inputs,
then maybe you can get away with linearized membrane properties, but you'd
still want to have membrane capacitance and resistance.

When you say your models are "complex" do you mean anatomically complex, or
biophysically/kinetically complex? If the former, have you used the d_lambda method
for specifying the spatial grid (to achieve accuracy while preserving efficiency)? If the
latter, are the active currents specified in a way that makes them compatible with
adaptive integration?

--Ted