How should I initialize the conductances of the synapses , lets say between (1 - 8 Micro S)?

[Henry Miller]

Hi Marcel, thank you for your valuable insights and kind assistance in the previous problems I have faced. I apologize if I was unclear in describing the problem, especially on the focus.

Here are 3 files that I have attached to summarize the problem:

description.png - The model that I am trying to simulate with my experimental values
my_simulation.py - The code that I use for my simulation
output.png - Output from running the code

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Problem Summary:

1. The curves in the asymmetric STDP window function are fitted based on experimental values (description.png)
2. The initial conductances for the left ( delta t negative ) and right (delta t positive ) are different (1 MicroS and 8 MicroS respectively) (description.png)

3. I am trying to model how the distribution of initial conductances will evolve over time under the (conductance based synapse) model with input Poisson neuron and output LIF neuron (Refer to description.png)
4 .I ran the simulation with gmax = 0.01 and S.w (The initial conductances) between (0,gmax) and all the weights seem to blow up at gmax after 100 second
5. This is likely due to the mismatch of numerical order between gmax and dApre, dApost
6. If so, how should I initialise S.w if I want them to make sense in accordance with my experimental values ( between 1 to 9 Micro S conductance)?

7. Which parts of the codes should I change?

Thank you for your assistance and sorry for the trouble!

[Marcel Stimberg]

Hi Henry,

I am afraid I cannot answer all these questions. It seems that most of them are not about the use of Brian but rather on how to model your experiment in general (which, as I said earlier, would rather be a discussion to have with a supervisor or collaborator). A few comments, though:

* Are you sure that your STDP window is 12 microseconds? I don't think you can have a pre spike trigger a post spike with such a short delay. In particular, the Brian simulation uses a time step of 100µs, so all spike times are forced to that grid and you will never encounter spike time differences above 0 but below 100µs. You could change Brian's timestep by setting defaultclock.dt, but it doesn't seem reasonable to me to simulate at that time scale for a neuron with a membrane time constant of 20ms.

* I am not sure what you mean by the "initial conductances" for the left/right parts of the STDP curve.

* In your model formulation, ge and gmax are multiples of the membrane conductance. You could use a different model formulation where you have the membrane capacitance and membrane conductance as separate variables instead of a single membrane time constant tau. Then, you can set your synaptic conductances to use actual units as well.

Two general suggestions:

* Make sure that the model parts are reasonable individually before putting everything in (e.g. does the cell something reasonable without STDP, e.g. does it fire at all but not at every time step?)

* Do not expect to be able to directly put experimentally measured values into a highly simplified model without adjusting them. I guess your experimental cells do not have exactly 1000 synapses spiking at 14Hz either.

Best,

  Marcel

[Henry Miller]

Ah thank you for your kind assistance! I realised that I might have been looking at the wrong direction. Poisson neuron spikes might be more for the 'stimulus' sense, whereas what I am trying to simulate is using 'rectangular EPSP' with poisson- distributed initiation times ( The rectangular-voltage-time waveform) as pre-spikes to trigger the output neuron to generate a post 'rectangular EPSP', eventually to change the synaptic weights over time using this pre-post pair.  I will look more into it, much thanks!

Regards,
Henry