Allen Institute Brain Modeling Tool Kit (BMTK) + Gap Junctions

Post Reply
bll5z6
Posts: 27
Joined: Thu May 26, 2016 10:27 am

Allen Institute Brain Modeling Tool Kit (BMTK) + Gap Junctions

Post by bll5z6 »

Hey all,

I have been contributing to BMTK (https://github.com/alleninstitute/bmtk), which uses NEURON to model biophysical neural networks. I talked briefly with Ted at SfN about this issue I've been having implementing gap junctions so thought I would formalize that here to see if anyone can help.

BMTK has two independent steps to creating a model; building and simulation. Building the model actually doesn't use NEURON at all, it simply creates h5 files that store information about the cells, connectivity, etc. The simulation step loads these files and creates a NEURON model out of them.

I have provided a minimal working example of a synaptic connection (chemical synapse) between two cells here:

https://github.com/latimerb/BLA_SingleC ... apJunction

The readme file in the top level of the repository explains how to run the BMTK simulation.

Digging through the BMTK code is a bit of a chore, but we managed to identify the place where synapses are made:

https://github.com/latimerb/bmtk/blob/d ... ll.py#L161

Our thought was to simply handle the gap junction here through point processes as we have seen done in the past:

https://github.com/latimerb/BLA_SingleC ... on/test.py

However, when we tried that, we got the message "segmentation fault (core dumped)" when h.run() was called from BMTK. This is where the simulation gets initialized and run:

https://github.com/latimerb/bmtk/blob/d ... or.py#L228

I noticed that pc.psolve() is not used in BMTK, could that be a problem? Is there anything to do with how the simulation is initialized and run that would conflict with the gap junction method outlined above? Any guidance is greatly appreciated.

Ben
ted
Site Admin
Posts: 5809
Joined: Wed May 18, 2005 4:50 pm
Location: Yale University School of Medicine
Contact:

Re: Allen Institute Brain Modeling Tool Kit (BMTK) + Gap Junctions

Post by ted »

First, just to help isolate the cause of the problem--

NOT using BMTK
1. have you successfully implemented and executed simulations of a toy network model that involves gap junctions and is NOT parallelized?
2. have you successfully implemented and executed simulations of the same network model, but parallelized and using Parallel Transfer https://www.neuron.yale.edu/neuron/stat ... l-transfer to communicate membrane potentials between hosts?

Next, assuming that the answers to 1 and 2 are yes, on to questions that probe ways to work around BMTK's current limitation--

When using BMTK, is there a clear separation between model setup and simulation execution? If so, can that separation be exploited by changing the normal sequence

BMTK-controlled model setup
BMTK-controlled simulation execution

to

BMTK-controlled model setup
user-specified additional custom setup
BMTK-controlled simulation execution

?

If yes, then I'd suggest doing all gap junction setup in the "user-specified addtional custom setup" stage. That avoids having to tinker with BMTK's internals (although you'll have to be able to discover or calculate where BMTK put your cells of interest).
Post Reply