Scope and Direction

[huangziwei]

This thread is about the scope and directions of jaxley-mech.

jaxley-mech will be built on top of jaxley and offer wide-range of paper-specific implementations of ion channels, synapses and receptors.

That said, there's actually no strong reasons against simply adding them to jaxley directly. But my preference right now is to keep jaxley as an inference machine (with a few canonical examples of how to implement channels, synapses and receptors there), while offers paper-specific implementations in jaxley-mech.

UPDATE (20230130)

I think I will prioritize going through the NeuronC documentation and porting the relevant parts, probably for quite a while

[philippberens]

This sounds good to me. Why do you focus on papers as the organizing unit?

I think it would be nice to provide also a template implementation e.g. of channels we recommend when simulation mouse retina

[huangziwei]

the main rationale is that the rate constants for the gating variables vary from paper to paper. They are also just some parameters fitted by the experimental data (such as voltage clamp data) to determine a channel's kinetics, but in most papers they are presented as fixed constants.

[huangziwei]

maybe after implementing enough mechanisms, we can add some presets:

from jaxley_mech.presets import retina_mouse, retina_cat # etc 

it would also be nice to be able to generate a table of all mechanisms in the presets with reference.

[philippberens]

Ok, makes sense... do we have a strategy for synapses yet?

[huangziwei]

no. need to read some papers first.

[huangziwei]

a few receptors (AMPA, NMDA, GABAa/b) are now implemented as synapses:

https://github.com/jaxleyverse/jaxley-mech/blob/main/notebooks/synapses/dms98.ipynb

[michaeldeistler]

Sorry for the late reply, I had not activated notifications for this repo...but yes, keeping this as a separate repo sounds good to me!