Dear all.
Thank you very much for the feedback ! Very much appreciated ! (ans sorry for the late reply).
This is a very interesting discussion, I can totally see that it points to old discussions and I agree with you @dan and @mstimberg, it doesn’t need to be included within the software. My motivation was really just about finding a good/standard/robust way to implement the mechanisms, and find the way to make this implementation as “Brian2-ish” as possible.
So @mstimberg your proposal sounds very nice to me, working on those two parallel lines would be great. 1) some (minor) tweaks in the Equation object, e.g. the EquationTemplate of your Issue’s discussion with @dan seems very close to what I would like to have and 2) just releasing the standards of way to implement those mechanisms. For me it can be of any form: a dedicated website, even just the “from_papers/” directory of the repo would be good already. And yes, benefiting from the NeuroML material would indeed be very valuable !
For point 1), there might also be indeed be some conceptual work to do as well. E.g. you pointed out:
The main problem is the “scoping” issue, e.g. how to avoid clashes between two gating variables “m”,
Even more tricky things should be caught, like: “[Ca2+]-dependent currents should be inserted only with a model for Calcium dynamics (pump, …)”, or the redundancy problem of @rth, etc … It might be good to have a place to list the constraints that we would like to be caught in an ideal implementation, maybe we could start an “Issue” for that ?
On my side, I’ll set up a github repository repository with this material, I will add a few additional things that might be useful for cellular physiology in the cortex (e.g. pointing to nicely formatted morphology datasets that load straight into brian2, …).
I will post the link of the repo here if useful for anyone (I’ll try to do this soon).
And just sharing some thoughts on the discussion about the Frankenstein models. Personally, I totally agree with the opinion of @rbrette. In my scientific work in general, I would personally prefer to stick to study-specific low-dimensional models. But other investigators put more emphasis on “you can not look at this phenomenon without the full high-dimensional picture, so please use the Frankenstein model ! And yes, if only this T-channel model is available, plug in the thalamic relay cell model in your Layer 2/3 V1 pyramidal dendrite model, even if the parameters are wrong, at least you’ll get a degree of the complexity”. My experience is that the field of theoretical neuroscience is divided into two groups of people: those that fear high-dimensionality (usually people with maths/physics background) and those that fear the lack of complexity (usually more biology backgrounds). To some extent, I find it constructive to have those antagonistic approaches in the field, so it could make sense to have a simulator that would allow those people to interact easily, but that’s just my opinion. And anyway, the only way to discuss/refine/discard the Frankenstein models is to benefit from good implementations 