neuroml-synapses Mailing List for NeuroML

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Hi,

These are a few points that came up in discussion with Padraig over 
the Synapses proposal. Since it is still in an early stage, they are 
more just things to bear in mind for the next version, rather than 
concrete suggestions, though I hope other people will have ideas to add.

For synapse specifications, I'm not a fan of the notion of a synapse
as a black box that reads the postsynaptic membrane potential and
tells the postsynaptic model what current to inject. This in effect
forces the simulator of the postsynaptic neuron to use an explicit
integration method.  Instead, for an ohmic current, the black box
should tell the postsynaptic simulator the conductance and driving
force and let it compute the current (which it could then do
implicitly, taking into account changes to the potential during the
step, hence being more stable at large timesteps etc etc.)
If the conductance law is more complicated, I'd favor having the
synapse specification tell the simulator what expression to apply
rather than having the synapse model offer to compute it though
the latter could be useful as a fall-back position no doubt.

Another issue that needs some clarification is the combination of 
axonal delays and access to the presynaptic potential. There are 
several different cases which risk being mixed up by the single use of 
"V" in the current specification:

1) a synapse model which is present on a modeled axon, so it has
    access to the actual membrane potential at the presynaptic
   terminal.  There should be no concept of axoanl delay with such a
    model.

2) a synapse model that receives an event when the presnyaptic cell
    fires. The delay is applied by the presynaptic cell model somehow,
    and the synapse gets an event at the right time. It doesn't have a
    concept of delay or presynaptic potential (unless it introduces the
    latter by, say, applying a precomputed voltage profile in response
    to the event).

3)  a synapse model that is given access to the soma of the
     presynapytic cell and a delay and then does its own thing (eg
     for a network of point neurons with complicated synapse models).
     It makes sense to use the V in a threshold sensor, but not expose
     the presynaptic machinery to this potential.

I'm not really sure what to do here  but it would be good to
distinguish the two V's perhaps as V_local and V_remote and somehow
make it hard to do silly things like take a synapse model defined
to depend on V_local and attach it to the soma with a non-zero delay.

Cheers,

Robert

-- 

Robert Cannon, A.nnotate.com,  (Textensor Ltd - www.textensor.com)
ro...@te...  tel: +44 (0)131 2082026  fax: +44 (0)131 4644881

neuroml-synapses Mailing List for NeuroML

neuroml-synapses — A mailing list for descriptions of synaptic mechanisms in NeuroML