Describing Brian 2 Models with SONATA

Overview

Brian2 is a spiking neural network simulator. This document describes how a network is defined using SONATA, to be loaded by the Brian2 simulator.

Warning

This is a preliminary draft, and does not handle several facets of circuits. It is also a living document; as features are added, please update them. See Future Work below for what needs to be added.

Physiology

Brian2 describes neuron physiology with a set of equations supplied at NeuronGroup creation time.

Ex:

NeuronGroup( # create neurons
    model='''dv/dt = (v_0 - v + g) / t_mbr : volt (unless refractory)
            dg/dt = -g / tau               : volt (unless refractory)
            rfc                            : second
            ''',
    method,
    method_options,
    threshold,
    reset,
    refractory,
            [...]
)

Thus, for the model_template arguments, within params one can have keys: model, method, method_options, threshold, reset, refractory. Within params, multiline values can be wrapped in a list to make it more readable.

The Brian2 simulator is also particular about units, so they must be encoded when using the namespace keyword. For this encoding to work, the following pattern is used: ‘tau’: [5, ‘ms’],; the loader will then apply the correct units at instantiation time. Each different set of parameters and equations is stored in a .json file under the components::point_neuron_models_dir as specified in components.

The associated name is stored in the model_template nodes field. Per neuron parameters are noted in the dynamics_params JSON object, as name, unit keys. These values are then loaded from the nodes file.

Thus, an example model template would be:

{
   "params": {
     "model": ["dv/dt = (v_0 - v + g) / t_mbr : volt (unless refractory)",
               "dg/dt = -g / tau              : volt (unless refractory)",
               "rfc                           : second"],
     "method": "linear",
     "threshold": "v > v_th",
     "reset": "v = v_rst; w = 0; g = 0 * mV",
     "refractory": "rfc"
   },
   "namespace": {
     "t_mbr": [20, "ms"],
     "tau": [5, "ms"],
     "v_0": [-52, "mV"],
     "v_rst": [-52, "mV"]
   },
   "dynamics_params": {
     "v_th": "mV"
   }
 }

The edges (ie: synapses) are parameterized in the same way; there is a template per synapse_type. In this example, there is a dynamics_params/v_th attribute in the nodes.h5 file, and when the simulator reads the module template, it will use that data to parameterize the neurons.

Nodes

The following are required for all models - it has the type of brian2_point for purposes of the SONATA circuit config file.

Group

Field

Type

Description

/0

model_template

utf8

Reference a template or class describing the electrophysical properties and mechanisms.

/0

model_type

utf8

brian2_point

/0

synapse_class

utf8

Defines the synapse type of the node; whether the neuron is inhibitory or excitatory. “EXC” or “INH”.

/0/dynamics_params

…

float32

Values described in dynamics_params

Values in the model_template dynamics_params section are read from the dynamics_params, and a per node value is registered in the simulator. For the example model template above, there would be /0/dynamics_params/v_th.

Edges

The following are required for all models - it has the type of brian2_synapse for purposes of the SONATA circuit config file.

Group

Field

Type

Description

/0

synapse_type

utf8

Model template from the point_neuron_models_dir

/dynamics_params

…

float32

Per synapse values (ex: w, pre.delay)

Inputs

Poisson Spike

A new input_type::module is defined for Poisson Spike:

"inputs": {
    "poisson_sugar": {
        "input_type": "spikes",
        "module": "poisson",
        "node_set": "sugar_neurons",
        "rate": "POISSON_RATE_in_Hz",
        "weight": "POISSON_W",
    }
}

Examples

FlyWire

Physiology

For the model template, this is used:

{
  "params": {
    "model": [
      "dv/dt = (v_0 - v + g) / t_mbr : volt (unless refractory)",
      "dg/dt = -g / tau               : volt (unless refractory)",
      "rfc                            : second",
    ],
    "method": "linear",
    "threshold": "v > v_th",
    "reset": "v = v_rst; w = 0; g = 0 * mV",
    "refractory": "rfc"
  },
  "namespace": {
    "t_mbr": [20.0, "ms"],
    "tau": [5.0, "ms"],
    "v_0": [-52.0, "mV"],
    "v_th": [-45.0, "mV"],
    "v_rst": [-52.0, "mV"]
  },
  "initial": {
    "v": [-52.0, "mV"],
    "g": [0, "mV"],
    "rfc": [2.2, "ms"]
  }
}

For the synapse template, this is used:

{
  "params": {
    "model": "w : volt",
    "on_pre": "g += w",
    "delay": [1.8, "ms"]
  },
  "dynamics": {
    "w": "mV"
  }
}

Nodes

The nodes files doesn’t contain much information per node, since most of the values are constant and set by the model template.

/nodes/drosophila/0/@library/model_template Dataset {1}
/nodes/drosophila/0/@library/model_type Dataset {1}
/nodes/drosophila/0/flywire_id Dataset {127400}
/nodes/drosophila/0/model_template Dataset {127400}
/nodes/drosophila/0/model_type Dataset {127400}
/nodes/drosophila/node_type_id Dataset {127400}

Edges

Only the w (or weight) is different per synapses, so it is stored along with the information about which model_template to use.

/edges/drosophila__drosophila__brian2_synapse/0/@library/model_template Dataset {1}
/edges/drosophila__drosophila__brian2_synapse/0/model_template Dataset {14687178}
/edges/drosophila__drosophila__brian2_synapse/0/w Dataset {14687178}
/edges/drosophila__drosophila__brian2_synapse/edge_type_id Dataset {14687178}
/edges/drosophila__drosophila__brian2_synapse/source_node_id Dataset {14687178}
/edges/drosophila__drosophila__brian2_synapse/target_node_id Dataset {14687178}

Circuit Config

{
  "components": {
    "point_neuron_models_dir": "models"
  },
  "node_sets_file": "node_sets.json",
  "target_simulator": "Brian2",
  "networks": {
    "nodes": [
      {
        "nodes_file": "nodes.h5",
        "populations": {
          "drosophila": {
            "type": "brian2_point"
          }
        }
      }
    ],
    "edges": [
      {
        "edges_file": "edges.h5",
        "populations": {
          "drosophila__drosophila__brian2_synapse": {
            "type": "brian2_synapse"
          }
        }
      }
    ]
  }
}

Simulation Config

{
  "run": {
    "tstop": 1000,
    "dt": 0.1,
    "random_seed": 42
  },
  "target_simulator": "Brian2",
  "network": "../output/circuit_config.json",
  "inputs": {
      "poisson": {
          "input_type": "spikes",
          "module": "poisson",
          "node_set": "sugar",
          "delay": 0,
          "duration": 1000,
          "rate": 150,
          "weight": 68.75
      }
  }
}

Future Work