Reconstructions and validations of neurological circuits for topological analysis
This project provides a method to work with networks as if they were biological networks. The primary use case is neurological networks, particularly EPFL-BlueBrain neocortical microcircuit. Results of experiments run with this code are available at the Neuro-Topology group website. This project is related to several other projects:
- NEST simulator for the simulation environment
- flagser for counting active simplices during the simulation
The main goals are for this project to be:
- open-source,
- executable on a personal computer,
- easily modifiable.
We will assume you are working in an Anaconda environment. First install NEST simulator and the relevant Python packages:
conda install -c conda-forge nest-simulator
conda install numpy scipy pandas pickle h5py
Next, clone this project. You only need the bbmc2.py file for everything to work, but the other files are necessary for more complex constructions and experiments.
Simply call python bbmc2.py. This will run a simulation for 100ms on the 31346-neuron mc2 Blue Brain column, with lots of strong stimuli sent to the circuit. A spike file with a name likebbmc2_constant_1591291009.npy will be output. To make a visual overview of the experiment, open the file analysis.py and paste the name of this output spike file, adjusting the other parameters as necessary. The end of the file should look like this:
nnum = 31346
spikes = 'bbmc2_constant_1591291009.npy'
time = 250
flagser='/home/user/flagser/flagser'
t1 = 5
t2 = 10
The location of flagser is necessary to compute the sizes of the transmission response active graphs (in the function flag_tr). If you are not computing this, you do not need flagser. Then call python analysis.py to analyze the output spike file. This will also compute the transmission response of the experiment for t1=5 and t2=10, compute the homology of the flagged active graph, and plot pairs of betti curves against each other.
The following options can be given the Python call initianting the file, as --option_name=option_value. Default values of options are indicated below, which (along with descriptions) are in the main file.
| Option | Type | Default | Description |
|---|---|---|---|
--fibres |
string | fibres.npy |
Filename of fibres (must be in folder "stimuli"). List of lists of GIDs. Length is number of thalamic fibers. |
--stimulus |
string | constant |
Firing pattern of stimulus (must be in folder "stimuli"), must be one of "n5","n15","n30","constant". |
--time |
integer | 100 |
Length, in milliseconds, of experiment. |
The following flags can be given the Python call initianting the file, as --flag_name. Default behavior when flags are not called is indicated below.
| Flag | If not called (default) | If called |
|---|---|---|
--no_mc2approx |
Approximate structure (distances between neurons, weights / delays / failures of synapses) is used. | Nothing beyond mc2 neuron connections is used. |
--shuffle |
Adjacency matrix is used as given. | Rows and columns of adjacency matrix are shuffled. |
ToDo.
Takes about 2 minutes to run a 250 ms simulation. The line nest.SetKernelStatus({"local_num_threads": 8}) indicates that the computer being used has 8 threads. This can (should) be adjusted for your own system.