This repository contains 2 versions of the same original mechanism Ampa obtained from modeldb.

1. cat_gt

   This directory contains the "ground truth" version of the mechanism, written against arbor's mechanism ABI in C++. It can be used to generate a mechanism catalogue (which can then be loaded in arbor) using arbor-build-catalogue Ampa cat_gt/ --raw Ampa -g cuda -C.

   cat_gt/used_nmodl/AMPA.mod was used as the starting point for writing the ground truth mechanism. It differs from the original mechanism in the following (major) ways:

   • Arrays tspike and PRE are commented out, along with the definition and usage of the NTdiffusion function. This is because arbor's modcc doesn't support arrays and for loops.
   • on, t0, x, y, u are ASSIGNED instead of arguments to the NET_RECEIVE function. This means that these variables are not shared among connections to the same synapse. This is needed because arbor doesn't support connection-local variables.
   • delay is added as a new state variable. It is used to replace the net_send functionality. It decays linearly and acts as a timer delaying the effect of a spike, similar to net_send in the original code.

   cat_gt/used_nmodl/AMPA.mod was first compiled using modcc by running modcc -t gpu -t cpu -N arb::Ampa_catalogue -o Ampa AMPA.mod, then the generated files were editted in the following (major) ways:

   • The arrays tspike and PRE as well as the definition and usage of the NTdiffusion function were added again.
   • The code was simplified for readability.

   Writing mechanisms this way is not recommended as it requires C++ knowledge as well as sufficient knowledge of the mechanism ABI. However, it can be used to verify the performance of the third version of the mechanism AMPA_approx.mod

2. cat_approx

   This directory contains the "approximate" version of the mechanism, written in nmodl. It can be used to generate a mechanism catalogue (which can then be loaded in arbor) using arbor-build-catalogue Ampa cat_approx/ -g cuda -C.

   cat_approx/AMPA.mod differs from the original mechanism in the following (major) ways:

   • A kinetic scheme gstate is added to approximate the effect of the NTdiffusion function. It works on 3 state variables g_active, g_emission

    g_emission <-> g_active   (alpha, beta)
    g_active   <-> (gamma, 0)
   

   The first equation represents the diffusion of glutamate between the emission area and the active area. The second equation represents the diffusion of glutamate between the active area and the surrounds.

   Values for alpha, beta, and gammaneed to be determined using parameter optimization. ampa_cell.py and curve_fit.py are used to determine these parameters.