| 1 |
PointNet |
| 2 |
======== |
| 3 |
|
| 4 |
.. figure:: _static/images/bmtk_architecture_pointnet_highlight.jpg |
| 5 |
:scale: 40% |
| 6 |
|
| 7 |
PointNet is a simulation engine that utilizes `NEST <http://www.nest-simulator.org/>`_ to run large-scale point |
| 8 |
neuron network models. Features including: |
| 9 |
|
| 10 |
* Run the same simulation on a single core or in parallel on an HPC cluster with no extra programming required. |
| 11 |
* Supports any spiking neuron models (rates models in development) available in NEST or with user contributed modules. |
| 12 |
* Records neuron spiking, multi-meter recorded variables into the optimized SONATA data format. |
| 13 |
|
| 14 |
Inputs |
| 15 |
------ |
| 16 |
Inputs can be specified in the "inputs" sections of the `simulation config <simulators.html#configuration-files>`_, |
| 17 |
following the rules specified in the |
| 18 |
`SONATA Data format <https://github.com/AllenInstitute/sonata/blob/master/docs/SONATA_DEVELOPER_GUIDE.md#simulation-input---stimuli>`_. |
| 19 |
|
| 20 |
Spike-Trains |
| 21 |
++++++++++++ |
| 22 |
Cells with ``model_type`` value ``virtual`` are equivalent to NEST’s spike_generator models which will play a |
| 23 |
pre-recorded series of spikes throughout the simulation. You may use either a |
| 24 |
`SONATA spike-train file <./analyzer.html#spike-trains>`_, an NWB file, or a space-separated csv file with columns |
| 25 |
**node_id**, **population**, and **timestamps**. Examples of how to create your own spike-train files can be found |
| 26 |
`here <./analyzer.html#creating-spike-trains>`_. |
| 27 |
|
| 28 |
.. code:: json |
| 29 |
|
| 30 |
{ |
| 31 |
"LGN_spikes": { |
| 32 |
"input_type": "spikes", |
| 33 |
"module": "sonata", |
| 34 |
"input_file": "./inputs/lgn_spikes.h5", |
| 35 |
"node_set": {"population": "lgn"} |
| 36 |
} |
| 37 |
} |
| 38 |
|
| 39 |
* module: either sonata, hdf5, csv, or nwb: depending on the format of the spikes file |
| 40 |
* `node_set <./simulators.html#node-sets>`_: used to filter which cells will receive the inputs |
| 41 |
* input_file: path to file contain spike-trains for one or mode node |
| 42 |
|
| 43 |
|
| 44 |
`Extracelluar ElectroPhysiology (ECEPhys) Probe Data (NWB 2.0) Spikes <ecephys_probe.html>`_ |
| 45 |
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ |
| 46 |
An increasing number of ECEPhys electrode experimental data is being release to the public in NWB format, such as the |
| 47 |
`Allen Visual Coding - Neuropixels <https://allensdk.readthedocs.io/en/latest/visual_coding_neuropixels.html>`_ dataset or through |
| 48 |
`DANDI <https://dandiarchive.org/>`_. While it is possible to manually convert this data into SONATA spike-trains to |
| 49 |
encorpate into your simulations, the `ecephys_probe` spikes module can do this automatically; fetching spikes from ECEPhys units |
| 50 |
and converting them to virtual cells for network input into your model. |
| 51 |
|
| 52 |
For example, using a session NWB downloaded using the AllenSDK, the below example wil randomly l map "LGd" cells from the session onto our |
| 53 |
"LGN" population, and filter out only spikes that occur between 10.0 and 12.0 seconds |
| 54 |
|
| 55 |
.. code:: json |
| 56 |
|
| 57 |
{ |
| 58 |
"inputs": { |
| 59 |
"LGN_spikes": { |
| 60 |
"input_type": "spikes", |
| 61 |
"module": "ecephys_probe", |
| 62 |
"input_file": "./session_715093703.nwb", |
| 63 |
"node_set": {"population": "LGN"}, |
| 64 |
"mapping": "sample_with_replacement", |
| 65 |
"units": { |
| 66 |
"location": "LGd" |
| 67 |
}, |
| 68 |
"interval": [10000.0, 12000.0] |
| 69 |
} |
| 70 |
} |
| 71 |
} |
| 72 |
|
| 73 |
See the `documentation <ecephys_probe.html>`_ for more information and advanced features. |
| 74 |
|
| 75 |
`Current Clamps <current_clamps.html>`_ |
| 76 |
+++++++++++++++++++++++++++++++++++++++ |
| 77 |
May use one step current clamp on multiple nodes, or have one node receive multiple current injections. |
| 78 |
|
| 79 |
.. code:: json |
| 80 |
|
| 81 |
{ |
| 82 |
"current_clamp_1": { |
| 83 |
"input_type": "current_clamp", |
| 84 |
"module": "IClamp", |
| 85 |
"node_set": "biophys_cells", |
| 86 |
"amp": 0.1500, |
| 87 |
"delay": 500.0, |
| 88 |
"duration": 500.0 |
| 89 |
} |
| 90 |
} |
| 91 |
|
| 92 |
See `documentation <current_clamps.html>`_ for more details on using current clamp inputs. |
| 93 |
|
| 94 |
|
| 95 |
Outputs |
| 96 |
------- |
| 97 |
|
| 98 |
Spikes |
| 99 |
++++++ |
| 100 |
By default all non-virtual cells in the circuit will have all their spikes at the soma recorded. |
| 101 |
|
| 102 |
|
| 103 |
Membrane and Intracellular Variables |
| 104 |
++++++++++++++++++++++++++++++++++++ |
| 105 |
Used to record the time trace of specific cell variables, usually the membrane potential (v). This is equivalent to NEST’s multimeter object. |
| 106 |
|
| 107 |
.. code:: json |
| 108 |
|
| 109 |
{ |
| 110 |
"membrane_potential": { |
| 111 |
"module": "multimeter_report", |
| 112 |
"cells": {"population": "V1"}, |
| 113 |
"variable_name": "V_m" |
| 114 |
"file_name": "cai_traces.h5" |
| 115 |
} |
| 116 |
} |
| 117 |
|
| 118 |
* module: either mutlimeter_report or membrane_report, both the same |
| 119 |
* variable_name: name of variable being recorded, will depend on the nest cell model. |
| 120 |
* cells: a `node_set <./simulators.html#node-sets>`_ filter out what cells to record. |
| 121 |
* file_name: name of file where traces will be recorded, under the “output_dir”. If not specified the the report title |
| 122 |
will be used, eg “calcium_concentration.h5” and “membrane_potential.h5” |
| 123 |
|
| 124 |
|
| 125 |
Recording Synaptic Weights |
| 126 |
++++++++++++++++++++++++++ |
| 127 |
Used to record the changes to synaptic weight changes throughout the simulation lifetime. Useful for measuring changes plastic synapse models like |
| 128 |
"stdp_synapse" or "tsodyks_synapses" (can be used for static synapses though values will never change). To create a recorder add the following |
| 129 |
section to the "reports" section in the simulation config json: |
| 130 |
|
| 131 |
.. code:: json |
| 132 |
|
| 133 |
{ |
| 134 |
"reports": { |
| 135 |
"<name>": { |
| 136 |
"module": "weight_recorder", |
| 137 |
"nest_model": "<original-nest-model>", |
| 138 |
"model_template": "<recorder-name>", |
| 139 |
"file_name": "<file-name>.csv" |
| 140 |
} |
| 141 |
} |
| 142 |
} |
| 143 |
|
| 144 |
Which will create a special synpase model called "<recorder-name>", which is just a version of *<original-nest-model>* that will save a trace |
| 145 |
of synapic changes to the csv file *output/<file-name>.csv*. Just set **model_template** property value to "<recorder-name>" in the edge-types file. |
| 146 |
|
| 147 |
For example, to record the changes to a subset of the *stdp_synapse* type NEST models, add the following to the configuration: |
| 148 |
|
| 149 |
.. code:: json |
| 150 |
|
| 151 |
{ |
| 152 |
"reports": { |
| 153 |
"weight_recorder_stdp_1": { |
| 154 |
"module": "weight_recorder", |
| 155 |
"nest_model": "stdp_synapse", |
| 156 |
"model_template": "stdp_synapse_recorder", |
| 157 |
"file_name": "stdp_weights.csv", |
| 158 |
} |
| 159 |
} |
| 160 |
} |
| 161 |
|
| 162 |
Then make changes to **edge_types.csv** file |
| 163 |
|
| 164 |
.. list-table:: |
| 165 |
:widths: 25 25 25 25 |
| 166 |
:header-rows: 1 |
| 167 |
|
| 168 |
* - edge_type_id |
| 169 |
- model_template |
| 170 |
- dynamics_params |
| 171 |
- ... |
| 172 |
* - 100 |
| 173 |
- stdp_synapse_recorder |
| 174 |
- stdp_params.json |
| 175 |
- ... |