synapses as pointprocess, no cell

[rllin]

I am trying to simulate an isopotential compartment with some excitatory and some inhibitory synapses and custom input spikes and ability to record output.

I was told to use only pointprocesses. Do I have to make sure these synapses created with just

Code: Select all

//20 June 2011 Adds IntFire4 synapses to artifical cell

	objref synlist, synob

obfunc addSynapses() { 
	num_synapses = $1
	objref synob
	synlist = new List()
	for i=0, num_synapses {
		synob = new IntFire4()
		synlist.append(synob)
	}
	return synlist
}

are at the same point? Only one synapse will be for input and one will be for output. Thanks!

[ted]

If you are dealing with a single compartment model, unless these Exp4Syns have different reversal potentials or time constants, there's no need to have more than 1. If all excitatory synapses are to have the same reversal potential and time constants, use one Exp4Syn with those params. Ditto for the inhibitory synapses.

[rllin]

By Exp4Syn, you mean IntFire4, right?

Also, I can record output just by recording a new NetCon with same input source and a nil right? How would I see the output spikes? It wouldn't be the default v(0.5) right?

Thanks!


EDIT: I think I figured I can see spikes by plotting .M ? Anyway I am still unsure as to why there cannot be more identical synapses on a single compartmental model. Just for context, I am trying to reproduce figure 2 from The Information efficacy of a synapse http://www.ncbi.nlm.nih.gov/pubmed/11896396. They have 400 excitatory and 100 inhibitory synapses with identical parameters respectively on a single compartment. They pick one excitatory to be the "input to be analyzed" and see how the other synapses affect this input. Am I understanding your reply or the paper wrong? Thanks!

[ted]

I'm sorry. Was answering while distracted by other things, and got all messed up.

Back to your original query--

I am trying to simulate an isopotential compartment with some excitatory and some inhibitory synapses and custom input spikes and ability to record output.

"Isopotential compartment" implies to me a biophysical model cell. IntFire4 is an artificial spiking cell class, not a biophysical model cell, and definitely not a synaptic mechanism. So the first question is, which do you want? And the second is, what on earth are "custom input spikes"?

Finally, the code example you provide is irrelevant to both alternatives. A biophysical model cell that consists of a single compartment needs only ExpSyn to emulate ampaergic synaptic input, regardless of how many presynaptic cells drive the model, unless you require that the inputs have different reversal potentials or time constants for synaptic conductance decay. And only one Exp2Syn would be needed to represent inhibitory synaptic input, regardless of how many inhibitory interneurons innervate the cell, unless you want the inhibitory synapses to differ from each other in reversal potential or time course.

Regarding your second message, your query about my first response is moot since my first response was rubbish.

I can record output just by recording a new NetCon with same input source and a nil right? How would I see the output spikes? It wouldn't be the default v(0.5) right?

What output do you mean? Is your model cell excitable or passive? What do you want: spike times, or to see actual time course of membrane potential?

I think I figured I can see spikes by plotting .M ?

No. M has nothing to do with anything because IntFire4 is irrelevant to your original query (see above--it's not a synaptic mechanism, and it's not a single compartment biophysical model cell).

Code: Select all

Anyway I am still unsure as to why there cannot be more identical synapses on a single compartmental model.

Who said there can't? The point is that you don't need more than one event-driven synaptic mechanism to represent the effect of multiple identical synapses attached to a single compartment biophysical model. One is entirely sufficient. More just clutters up the code, consumes memory, and for no benefit. But either way is capable of producing "correct" results.

Just for context, I am trying to reproduce figure 2 from The Information efficacy of a synapse http://www.ncbi.nlm.nih.gov/pubmed/11896396. They have 400 excitatory and 100 inhibitory synapses with identical parameters respectively on a single compartment.

I can't comment on code that I haven't seen, or on why the implementers of a particular model did what they did. For all I know they might have had some perfectly valid reason (other than efficiency and simplicity of implementation) to do something other than what would be simplest and most efficient. Are you quite sure about the implementational details, i.e. have you actually seen their source code or did you just form your own interpretation of a verbal description in their paper?

[rllin]

I see there is much I don't understand completely. Thanks for the thorough reply.

Yes, this is purely from interpretation so, as you can see, I am not clear about many things.

And the second is, what on earth are "custom input spikes"?

By custom I mean, for example, the input spikes are from a data file. I guess this doesn't affect the question too much though.

Since I actually have no "cell," I'm not sure whether it is excitable or passive. I'm guessing excitable since I would like an affected output?

Ah, I think I see what you mean when you say to use ExpSyn and Exp2Syn. They do seem to fit better. Just for conceptual reasons though, could you explain more why IntFire4 wouldn't work though? (for a synapse that is)

Instead of multiple synapses maybe multiple connections to the same synapse? Does that make sense?


Thanks!

[ted]

To facilitate further discussion it would be very helpful if you would read the following:
1. "The NEURON simulation environment" (overviewforhbtnn2e.pdf)
or
"The NEURON eimulation environment"
An earlier paper with the same title, expanded from our first article in Neural Computation.
These introduce the basic concepts that underlie the design of NEURON and the implementation of models with NEURON, with an emphasis on biophysical models of cells.
2. Hines, M.L. and Carnevale, N.T. Discrete event simulation in the NEURON environment. Neurocomputing 58-60:1117-1122, 2004.
This discusses how artificial spiking cells are implemented in NEURON and introduces the event delivery system.

You might also find it helpful to read the section about NEURON in
Brette et al. Simulation of networks of spiking neurons: a review of tools and strategies. J. Comput. Neurosci. 23:349-398, 2007.
which may help you understand something about how spike driven synaptic transmission is implemented in NEURON.

All of these articles are available from links at http://www.neuron.yale.edu/neuron/nrnpubs

Spike driven synaptic transmission is implemented with events that are conveyed from a presynaptic source to a postsynaptic target by instances of the NetCon (Network Connection) class. Sources may be artificial spiking cells or biophysical model cells. NEURON's built in artificial spiking cells and synaptic mechanism may receive events from one or more presynaptic sources.

The IntFire cell classes are artificial spiking cell models (other kinds of artificial spiking cell models are also possible in NEURON). Artificial spiking cell models in NEURON are "point neurons" described by very simple equations that have analytical solutions and do not require numerical integration. They are not synaptic mechanism models. They receive events from other cells (via NetCons) and generate events that may be deliverd to other cells (via NetCons).

ExpSyn and Exp2Syn are generic conductance change synaptic mechanisms that may be attached to biophysical model cells (model cells that are implemented with one or more "section"s (think "neurite" or "a length of unbranched cable")). Other kinds of synaptic mechanisms are also possible in NEURON. All are targets of events from presynaptic sources, and the events are delivered by NetCons.

By custom I mean, for example, the input spikes are from a data file.

It is easy to use such data to control the generation of spike events in NEURON.

Since I actually have no "cell," I'm not sure whether it is excitable or passive. I'm guessing excitable since I would like an affected output?

The indispensible prerequisite for any computational model is to start with a clear and explicit conceptual model. The only reason to use computational modeling is when one has a conceptual model that is too complex to allow one to draw useful inferences by intuition alone.