Simulating networks¶
Simulations can be run through the CLI:
bsb simulate my_network.hdf5 my_sim_name
or through the bsb library for more
control. When using the CLI, the framework sets up a “hands off” simulation workflow:
Read the network file
Read the simulation configuration
Translate the simulation configuration to the simulator
Create all cells, connections and devices
Run the simulation
Collect all the output
When you use the library, you can set up more complex workflows, such as parameter sweeps:
# A module to read HDF5 data
# A module to run NEURON simulations in isolation
import nrnsub
from bsb import from_storage
# This decorator runs each call to the function in isolation
# on a separate process. Does not work with MPI.
@nrnsub.isolate
def sweep(param):
# Open the network file
network = from_storage("my_network.hdf5")
# Here you can set whatever simulation parameter you want.
network.simulations.my_sim.devices.my_stim.rate = param
# Run the simulation
results = network.run_simulation("my_sim")
# Tjese are the recorded spiketrains and signals
print(results.spiketrains)
print(results.analogsignals)
for i in range(11):
# Sweep parameter from 0 to 1 in 0.1 increments
sweep(i / 10)
Parallel simulations
To parallelize any BSB task prepend the MPI command in front of the BSB CLI command, or the Python script command:
mpirun -n 4 bsb simulate my_network.hdf5 my_sim_name
mpirun -n 4 python my_simulation_script.py
Where n is the number of parallel nodes you’d like to use.
Configuration¶
Each simulation config block needs to specify which simulator they use. Valid
values are arbor, nest or neuron. Also included in the top level block are the
duration, resolution and temperature attributes:
{
"simulations": {
"my_arbor_sim": {
"simulator": "arbor",
"duration": 2000,
"resolution": 0.025,
"temperature": 32,
"cell_models": {
},
"connection_models": {
},
"devices": {
}
}
}
}
The cell_models are the simulator specific representations of the network’s
cell types, the connection_models of the
network’s connectivity types and the
devices define the experimental setup (such as input stimuli and recorders).
All of the above is simulation backend specific and is covered per simulator below.
Arbor¶
Cell models¶
The keys given in the cell_models should correspond to a cell type in the
network. If a certain cell type does not have a corresponding cell model then no
cells of that type will be instantiated in the network. Cell models in Arbor should refer
to importable arborize cell models. The Arborize model’s .cable_cell factory will
be called to produce cell instances of the model:
{
"cell_models": {
"cell_type_A": {
"model": "my.models.ModelA"
},
"afferent_to_A": {
"relay": true
}
}
}
Note
Relays will be represented as spike_source_cells which can, through the connectome
relay signals of other relays or devices. spike_source_cells cannot be the target of
connections in Arbor, and the framework targets the targets of a relay instead, until
only cable_cells are targeted.
Connection models¶
todo: doc
{
"connection_models": {
"aff_to_A": {
"weight": 0.1,
"delay": 0.1
}
}
}
Devices¶
spike_generator and probes:
{
"devices": {
"input_stimulus": {
"device": "spike_generator",
"explicit_schedule": {
"times": [1,2,3]
},
"targetting": "cell_type",
"cell_types": ["mossy_fibers"]
},
"all_cell_recorder": {
"targetting": "representatives",
"device": "probe",
"probe_type": "membrane_voltage",
"where": "(uniform (all) 0 9 0)"
}
}
}
todo: doc & link to targetting
NEST¶
Additional root attributes:
modules: list of NEST extension modules to be installed.
{
"simulations": {
"first_simulation": {
"simulator": "nest",
"duration": 1000,
"resolution": 1.0,
"modules": ["cerebmodule"],
"cell_models": {
},
"connection_models": {
},
"devices": {
}
},
"second_simulation": {
}
}
}
Cell models¶
In the cell_models block, you specify the simulator representation
for each cell type. Each key in the block can have the following attributes:
model: NEST neuron model, See the available models in the NEST documentationconstants: neuron model parameters that are common to the NEST neuron models that could be used, including:t_ref: refractory period duration [ms]C_m: membrane capacitance [pF]V_th: threshold potential [mV]V_reset: reset potential [mV]E_L: leakage potential [mV]
Example¶
Configuration example for a cerebellar Golgi cell. In the eglif_cond_alpha_multisyn
neuron model, the 3 receptors are associated to synapses from glomeruli, Golgi cells and
Granule cells, respectively.
{
"cell_models": {
"golgi_cell": {
"constants": {
"t_ref": 2.0,
"C_m": 145.0,
"V_th": -55.0,
"V_reset": -75.0,
"E_L": -62.0
}
}
}
}
Connection models¶
Devices¶
NEURON¶
Cell models¶
By default the NEURON adapter uses an ArborizedCellModel, which loads
external arborize definition to instantiate cells.
{
"cell_models": {
"cell_type_A": {
"model": "dbbs_models.GranuleCell"
},
"cell_type_B": {
"model": "dbbs_models.PurkinjeCell"
}
}
}
This example dictates that during simulation setup, any member of
cell_type_A should be created by importing and using
dbbs_models.GranuleCell. Documentation incomplete, see arborize docs ad
interim.
Connection models¶
Once more the connection models are predefined inside of arborize and they
can be referenced by name:
{
"connection_models": {
"A_to_B": {
"synapses": ["AMPA", "NMDA"]
}
}
}
Devices¶
Devices send input, or record output.
Here we’ll place a spike generator and a voltage clamp:
{
"devices": {
"stimulus": {
"device": "spike_generator",
"targetting": {
"strategy": "cell_model",
"cell_models": ["cell_type_A"]
},
"synapses": ["AMPA"],
"start": 500,
"number": 10,
"interval": 10,
"noise": true
},
"voltage_clamp": {
"device": "voltage_clamp",
"targetting": {
"strategy": "cell_model",
"cell_models": ["cell_type_A"]
"count": 1,
},
"location": {
"strategy": "soma"
},
"parameters": {
"delay": 0,
"duration": 1000,
"after": 0,
"voltage": -63
}
}
}
}