Negative weights at NetCon

Moderator: wwlytton

Post Reply
Krishna Chaitanya
Posts: 70
Joined: Wed Jan 18, 2012 12:25 am
Location: University of Pavia

Negative weights at NetCon

Post by Krishna Chaitanya » Thu May 07, 2015 4:06 am

Hi,

I am trying to give excitatory and inhibitory connections to Izhikevich model. When I create NetCon objects for these connections, how would the model know the incoming connection is excitatory or inhibitory? I tried setting negative weights for inhibitory connections but the simulation slowed down significantly taking longer duration of time.

How will the Net_Receive block in izhikevich file know the incoming connection is excitatory or inhibitory?

Kindly let me know.

Thank you.

ted
Site Admin
Posts: 5572
Joined: Wed May 18, 2005 4:50 pm
Location: Yale University School of Medicine
Contact:

Re: Negative weights at NetCon

Post by ted » Thu May 07, 2015 10:51 am

Krishna Chaitanya wrote:I am trying to give excitatory and inhibitory connections to Izhikevich model. When I create NetCon objects for these connections, how would the model know the incoming connection is excitatory or inhibitory?
Depends on whether your particular model implements the cells as artificial spiking cells, or whether it implements them as biophysical model cells. If they're artificial spiking cells, it's the sign of the NetCon's weight that determines whether a particular input is excitatory or inhibitory. If they're biophysical model cells, all NetCon weights must be positive and it's the event-driven synaptic mechanism's reversal potential that determines whether an input is excitatory or inhibitory.

Krishna Chaitanya
Posts: 70
Joined: Wed Jan 18, 2012 12:25 am
Location: University of Pavia

Re: Negative weights at NetCon

Post by Krishna Chaitanya » Fri May 08, 2015 1:47 am

Dear Ted,

I did use negative weights for my Izhikevich mod file which is point process. But the simulation time has increased significantly. Whether I use NONSPECIFIC_CURRENT variable or not, the simulation time takes a very long duration.

Kindly let me know what to do.
Thank you.

ted
Site Admin
Posts: 5572
Joined: Wed May 18, 2005 4:50 pm
Location: Yale University School of Medicine
Contact:

Re: Negative weights at NetCon

Post by ted » Fri May 08, 2015 1:49 pm

Krishna Chaitanya wrote:I did use negative weights for my Izhikevich mod file which is point process.
The question isn't whether or not your model cells are implemented with point processes. The question is whether or not they are biophysical model cells. A biophysical model cell is implemented with one or more sections, and it operates by the integration of one or more currents by the section's membrane capacitance; those currents are produced by density mechanisms and/or point processes that generate NONSPECIFIC currents or WRITE ionic currents. In NEURON, an ARTIFICIAL_CELL has a NET_RECEIVE block but has no BREAKPOINT block and is not associated with a section or numerical integrator--that is, it does not refer to v or any ions, and it does not have a POINTER. Its internal workings are entirely isolated from other mechanisms, and it depends on discrete events from the outside to affect it, and it can affect the outside only by sending discrete events.

There is no way to "attach" a synaptic mechanism to an ARTIFICIAL_CELL; instead, events are delivered directly to the model cell instance, and statements in the NET_RECEIVE block govern how those events affect the cell. Whether a given synaptic input is excitatory or inhibitory depends on whether the weight associated with the NetCon that delivers the event is > 0 or < 0. A biophysical model cell, however, needs to have a synaptic mechanism "attached" to it in order for it to receive synaptic input. For most synaptic mechanisms, arrival of an event with weight w perturbs the synaptic conductance by an amount that is proprotional to w. I have yet to see source code for a synaptic mechanism that prevents a weight < 0 from having an effect; in the absence of such safeguards, it is possible to drive synaptic conductance < 0, which would be a meaningless result and, if the synaptic conductance is sufficiently negative, could have bad consequences for simulation stability.

So once again, the question: what kind of model cells are you using? Biophysical model cells or ARTIFICIAL_CELLs?

Post Reply