Generate equations from small multicompartmental models

[thesamovar]

The idea is that you might sometimes want to have a large network of neurons with a small number of compartments. Rather than writing out all the equations by hand, you could use the existing multicompartmental modelling framework (which only works for a single neuron) to generate those equations.

From an email exchange:

@thesamovar: A quick way to make this sort of thing possible would be to use the existing syntax to set up a multicompartmental neuron, but have a method that just returns the generated equations which can then be used in a standard NeuronGroup. You could also give specific variable names to specific variables in specific compartments.

@romainbrette: It's possible but requires a little bit of work. You would need basically to duplicate the equations for all compartments, then add a membrane equation that includes the lateral currents.

[mstimberg]

I'm a bit undecided about the best approach for this. On the one hand, it would be nice if you could just use the Morphology classes and then call some convenience function to create the equations for you, but on the other hand, I'm a bit worried that too much parallelism with the SpatialNeuron will make users think that you can use all of its features. E.g. when you create a SpatialNeuron, you can record the membrane potential from a specific compartment like this:

mon_R = StateMonitor(neuron, 'v', record=morpho.R[75*um])

If neuron is just a NeuronGroup with complex equations, this can't work (it would have to record v_R_17 or something like that). But maybe this is just a documentation issue.

Note that there was support for this kind of equation generation in Brian1 (MembraneEquation and Compartments classes): http://brian.readthedocs.org/en/1.4.1/librarymodels.html

[thesamovar]

I think you'd need to create names for specific compartments when you created it, e.g.

eqs = morpho.create_equations(r=morpho.R[75*um])
...
G = NeuronGroup(N, eqs, ...)
mon_r = StateMonitor(G, 'r')

Actually since threshold and reset are probably part of it, maybe we should directly create a NeuronGroup from morpho rather than just the equations, e.g. G=morpho.create_group(N, ...).

I would say that it could be a documentation issue in this case. Yes, users might well make mistakes because of not reading the docs but in this case those mistakes won't lead to silent problems, they'll raise errors, so it's not so bad.

I wouldn't like to have an entirely parallel system like the library models from Brian 1: too much duplication.

[romainbrette]

It should probably be kept separate from SpatialNeuron. For the naming: actually the morphology object already has (optionally) names for the branches, so those could be used. And the compartments could be numbered of subscripted with position. Reset and threshold should be added by the user.

Le 10/04/2015 13:38, Dan Goodman a écrit :

I think you'd need to create names for specific compartments when you created it, e.g.

eqs = morpho.create_equations(r=morpho.R[75*um]) ... G = NeuronGroup(N, eqs, ...) mon_r = StateMonitor(G, 'r')

Actually since threshold and reset are probably part of it, maybe we should directly create a |NeuronGroup| from |morpho| rather than just the equations, e.g. |G=morpho.create_group(N, ...)|.

I would say that it could be a documentation issue in this case. Yes, users might well make mistakes because of not reading the docs but in this case those mistakes won't lead to silent problems, they'll raise errors, so it's not so bad.

I wouldn't like to have an entirely parallel system like the library models from Brian 1: too much duplication.

— Reply to this email directly or view it on GitHub https://github.com/brian-team/brian2/issues/443#issuecomment-91527004.

Twitter: https://twitter.com/RomainBrette Blog: http://romainbrette.fr/category/blog/

New paper: Laudanski J, Zheng Y, Brette R (2014). A structural theory of pitch. eNeuro. DOI: 10.1523/ENEURO.0033-14.2014

[mstimberg]

I agree that we don't want to have a parallel system as in Brian 1, and we should certainly keep it separate from SpatialNeuronGroup. On the other hand, the most convenient thing from the user point of view would be to have another class that could be created like SpatialNeuronGroup, i.e. it takes the morphology, a transmembrane current etc. It would basically be a wrapper around NeuronGroup, putting the equations and everything else together. Still, connecting with synapses or recording with monitors would work differently from SpatialNeuronGroup. I'd be happy to help someone working on this (it shouldn't be very difficult in fact), but I think for us it is not a high priority right now.

[romainbrette]

I think you just need something to output the equations and then you can use NeuronGroup. This could be a method of Morphology.

Le 10/04/2015 15:51, Marcel Stimberg a écrit :

I agree that we don't want to have a parallel system as in Brian 1, and we should certainly keep it separate from |SpatialNeuronGroup|. On the other hand, the most convenient thing from the user point of view would be to have another class that could be /created/ like |SpatialNeuronGroup|, i.e. it takes the morphology, a transmembrane current etc. It would basically be a wrapper around |NeuronGroup|, putting the equations and everything else together. Still, connecting with synapses or recording with monitors would work differently from |SpatialNeuronGroup|. I'd be happy to help someone working on this (it shouldn't be very difficult in fact), but I think for us it is not a high priority right now.

— Reply to this email directly or view it on GitHub https://github.com/brian-team/brian2/issues/443#issuecomment-91564228.

Twitter: https://twitter.com/RomainBrette Blog: http://romainbrette.fr/category/blog/

New paper: Laudanski J, Zheng Y, Brette R (2014). A structural theory of pitch. eNeuro. DOI: 10.1523/ENEURO.0033-14.2014

[mstimberg]

Let's leave this to dendrify.