www.neuron.yale.edu

Hi,

I am trying to recode a NEURON model in MATLAB (Thalamocortical and Thalamic Reticular Network (Destexhe et al 1996) http://senselab.med.yale.edu/modeldb/Sh ... model=3343).

The portion of the model I am trying to recode consists of single compartment RE and TC cells, with GABAa, AMPA, and GABAb synapses. Most all of the channel conductances are in mS/cm^2 (in some cases uS/cm^2, but I've converted these to mS/cm^2). However, the Kleak conductances and the synaptic conductances are listed simply in nS or mS. For example, the AMPA synapse to the RE cell has a conductance of 0.2 uS. In the paper they list the surface area of the RE cell as being 14,300 um^2. I converted this surface area to cm^2. Then I simply divided the conductance (.0002 mS) by the area (1.4300e-04 cm^2) to get the conductance in mS/cm2.

Should this work? Is this the right way to do it? Will the AMPA, Alpha and Beta values for the synaptic kinetics still be appropriate?

It appears to be the case that the AMPA current is too small, this may be due to the fraction of receptors that are active, with the values of Alpha and Beta used or because the conductance values are wrong. I find that if I either increase the AMPA conductance by quite a bit (a factor of ten) or increase the Alpha value that my results look more similar (but not the same) to the output of the model in NEURON; I suspect this is because the others synaptic currents may not be working exactly as they should be either.

Any help would be greatly appreciated as I have been trying to get this to work for quite a while now.

Thanks!

neuron999 wrote:I am trying to recode a NEURON model in MATLAB (Thalamocortical and Thalamic Reticular Network (Destexhe et al 1996) http://senselab.med.yale.edu/modeldb/Sh ... model=3343).

It's always good to see a published model be reused. If your work leads to a new publication, could you please do ModelDB the great favor of mentioning that you got the NEURON source code from model entry 3343?

The portion of the model I am trying to recode consists of single compartment RE and TC cells, with GABAa, AMPA, and GABAb synapses. Most all of the channel conductances are in mS/cm^2 (in some cases uS/cm^2, but I've converted these to mS/cm^2). However, the Kleak conductances and the synaptic conductances are listed simply in nS or mS.

Authors of empirically-based computational models of neurons tend to fall into two camps: those who use absolute units for all membrane currents and membrane conductances, and those who use density units (current/area, conductance/area) for such variables. NEURON models use density units (S/cm2, mA/cm2) to specify channels and currents that are distributed over the surface of a cell, and absolute units (uS, nA) for channels and currents that are concentrated in a zone that is much smaller than a length constant and so can be treated as "point sources" (e.g. synapses). This all works out because NEURON knows the length and diameter of each compartment and is smart enough to factor compartment surface area into its calculations as necessary.

If you want your model to use density units for all conductances and currents, just divide each synaptic conductance by the surface areas of the compartment to which it is attached, as you appear to have done.


There are, of course, some caveats.

1. The surface area of a compartment in NEURON is PI*diameter*length, where diameter and length are the diameter and length of the compartment. In other words, the area of a cylinder but omitting the ends.

2. There is a tendency for parameters published in a paper to differ from the actual parameters in modelers' code. Believe the code.

3. Most source code is hard to understand, and some modelers use complex schemes and/or custom initializations for setting parameter values.* In many cases it is necessary to execute the code in order to determine parameter values. For example, with NEURON it (can be helpful | may be necessary) to retrieve the zip file from ModelDB, expand it, compile the mod files with mknrndll (or nrnivmodl for UNIX/Linux users), execute mosinit.hoc, and run the model (or at least run its initialization code). THEN one can use NEURON's ModelView tool (see tutorial at http://www.neuron.yale.edu/neuron/docs) to discover parameter values.

*--Destexhe's code tends to be particularly well structured and easy to read.

Will the AMPA, Alpha and Beta values for the synaptic kinetics still be appropriate?

Why would they not be? You're only changing the magnitude of a number that specifies the amplitude of a waveform. The rate constants will not be affected. But if you got the values of Alpha and Beta from ampa.mod, there is always a possibility that these were changed by some statement in the authors' own hoc code (see caveat 3) during model setup.

It appears to be the case that the AMPA current is too small, this may be due to the fraction of receptors that are active, with the values of Alpha and Beta used or because the conductance values are wrong.

If in doubt, get the NEURON code and execute it, and use ModelView to discover the parameter values. The ModelView tool can also export a text file that contains a more-or-less human-readable description of what's in a model (it can also export NeuroML, but I find its plain text output to be most informative).