The problem for coreneuron to trace V_soma

[Kim1230]

Hello,
I find a question for plotting the trace of V_soma of each cell. When I execute Neuron without Coreneuron, I can see the trace of V_soma. But when I use Coreneuron, the V_soma doesn't show. So, whether coreneuron doesn't have the function to record the V_soma?
Hope for your answer!

Thanks!

[nrnhines]

From https://nrn.readthedocs.io/en/latest/python/modelspec/programmatic/network/parcon.html#ParallelContext.nrncore_run the intention is for CoreNEURON to send back all the trajectories at the end of each run so that they will be shown on the GUI (and in all the vector.record instances). The negative about that is that shape plot and space plot movies won't display. There is provision with pc.nrncore_run(argstr, 1) to send GUI info back every time step but I don't expect that is relevant to your issue (and in any case, will have to become a neuron.coreneuron parameter for the user to set).
Anyway, let me verify that GUI behavior is correct when CoreNEURON is run via pc.psolve and, if so, can look more into the details of how your model is using it.

[Kim1230]

@nrnhines Thanks for your response. I use neuron with coreneuron to execute my model. And when I try to plot the fluctuating of v of each cell. It shows nothing. So, if I understand well, if I want to trace the v of my model. I have to use neuron without coreneuron ?

[nrnhines]

I was mistaken and the doc about nrncore_run is misleading. When not in per time step mode, only the Vector.record trajectories are sent back to NEURON. There are several paths I can take to remedy the issue. The correct but longer path is to record all the GUI trajectories on the CoreNEURON side and send them back at the end of a run for plotting, just as is done with Vector.record. A short term but partial fix that I'll do first is to give the standard pc.psolve style the same behavior as nrncore_run by allowing the user to ask for per time step return values for the GUI trajectories. For now, you have only two work arounds. 1) replace your pc.solve(tstop) call with

if  coreneuron.enable:
    pc.nrncore_run(coreneuron.nrncore_arg(tstop), 1)
else:
    pc.psolve(tstop)

2. Use Vector.record for each trajectory you want to plot and then pop up a Graph and loop over all the vectors with Vector.plot(Graph_instance, tvec) to see the trajectories.

[Kim1230]

Hello@nrnhines, Thanks for your response. I wanna use your solution to solve the problem. The pc.solve() I will replace is in the python file of the model(such as ringtest model) which I used not to modify the souce of neuron and coreneuron. And it will work to record each time step of v per cell. But the result is that it shows nothing again. What else should I do ?

By the way, I follow and debug the instruction of the coreneuron part. I can clearly see the spike record instruction in each time step. And I can't see the record instruction of the v. So I wander whether coreneuron has the ability to record the v in its computation. Maybe I am wrong.

And for the step 2, what should I do or where should I modify?
Hope for your help!

[nrnhines]

Yes. Line 100 of https://github.com/neuronsimulator/ringtest/blob/master/ringtest.py
But my assumption is that you are seeing correct graph output when you are running with coreneuron.enable=False
Is that assumption correct?

[Kim1230]

Hello, @nrnhines I try to use neuron with my model with tstop=1500ms and it absolutely be ok to show the figure like:

And the model I used is
https://github.com/suny-downstate-medical-center/S1_netpyne/tree/coreneuron (564,7 MB)
And netpyne I used is :
https://github.com/suny-downstate-medical-center/netpyne/releases/tag/v1.0.2.1

for the coreneuron, I have changed the code as:

Hope for your help

[iomaganaris]

I did some digging into NetPyNE and NEURON and it seems that the issue is that NEURON doesn't pass the trajectory requests to CoreNEURON to return the needed data.
The way the reports are instantiated in NetPyNE is: https://github.com/suny-downstate-medical-center/netpyne/blob/development/netpyne/cell/cell.py#L302
Also we call CoreNEURON from NetPyNE from here: https://github.com/suny-downstate-medical-center/netpyne/blob/development/netpyne/sim/run.py#L169
I managed to run a slimmed down version of the S1_netpyne model doing these changes:

diff --git a/sim/cfg.py b/sim/cfg.py
index 7323b12e..83184dab 100644
--- a/sim/cfg.py
+++ b/sim/cfg.py
@@ -27,7 +27,7 @@ cfg.random123 = True
 #------------------------------------------------------------------------------
 # Run parameters
 #------------------------------------------------------------------------------
-cfg.duration = 6.0*1e4 ## Duration of the sim, in ms
+cfg.duration = 10 ## Duration of the sim, in ms
 cfg.dt = 0.025
 cfg.seeds = {'conn': 4321, 'stim': 4321, 'loc': 4321}
 cfg.hParams = {'celsius': 34, 'v_init': -65}
@@ -188,11 +188,11 @@ cfg.analysis['plotTraces'] = {'include': cfg.recordCells, 'oneFigPer': 'cell', '
 #------------------------------------------------------------------------------
 # Network
 #------------------------------------------------------------------------------
-cfg.scale = 1.0 # reduce size
+cfg.scale = 0.05 # reduce size
 cfg.sizeY = 2082.0
 cfg.sizeX = 420.0 # r = 210 um and hexagonal side length = 230.9 um
 cfg.sizeZ = 420.0
-cfg.scaleDensity = 1.0 # Number of cells = 31346
+cfg.scaleDensity = 0.003 # Number of cells = 31346

 #------------------------------------------------------------------------------
 # Spontaneous synapses + background - data from Rat

To run the model I think it's enough to follow the instructions in the README of https://github.com/suny-downstate-medical-center/S1_netpyne/tree/coreneuron

[Kim1230]

@iomaganaris Thanks for your dedication. I know the issue now. And how I can fix it to satisfy my requirement to trace the V in the coreneruon mode ?

[iomaganaris]

Hello @Kim1230 ,

I think I found the issue. CoreNEURON doesn't support returning back record vector if they have a custom recording dt (recordStep). There is some work needed to be done in https://github.com/neuronsimulator/nrn/blob/master/src/nrncvode/netcvode.cpp#L5791 and maybe other places to support that.
Until this is implemented I would suggest removing the recordStep from https://github.com/suny-downstate-medical-center/netpyne/blob/development/netpyne/cell/cell.py#L303, so changing this line to:

sim.simData[key]['cell_'+str(self.gid)].record(ptr)

If you do that for consistency I think you need to set the recordStep in cfg.py to:

cfg.recordStep = cfg.dt

Let me know if this works for you

[Kim1230]

@iomaganaris Thanks for your help. It works!

[iomaganaris]

Great! Thanks for letting us know. Btw with this change you can also keep using pc.psolve to launch the simulation. There is no need to call nrncore_run so that you can avoid the overhead of transferring back and forth the data at each timestep