Skip to content

Latest commit

 

History

History
468 lines (367 loc) · 15.7 KB

README.md

File metadata and controls

468 lines (367 loc) · 15.7 KB
Raw

Neuron++ is a wrapper for NEURON (http://www.neuron.yale.edu) with easy to use Python objects. The key intention behind this framework was to perform tedious tasks in few lines of code with the object-oriented paradigm.

NEURON allows to create simulations of Biological Neural Networks. The Neuron++ framework was designed to match the simplicity of libraries such as Keras library for Artificial Neural Networks. With Neuron++ you can easily create group of cells, stack them together as populations, then stimulate them with external input and collect readouts to perform any task.

Key features

  • Use for fast prototyping of neural models in NEURON simulator, using Python interface with the object-oriented paradigm (OOP)

  • Precisely define single cell models and connect them together to create a network

  • Auto-compilation of all MOD files on the fly

  • Auto-load SWC/ASC or HOC morphology for each cell

  • Upload HOC defined models and adapt them to your needs

  • Define in vitro experimental protocols (eg. STDP paradigms)

  • Manage synaptic signaling

  • Debug synapses and point processes RANGE values in real time with interactive stimulation from the keyboard

  • Create predefined dendritic spines and synapses with ease

  • Define populations of neurons and connect them together

  • Provide helpful exception messages and guidelines of how to use NEURON functions with Neuron++ wrapper without errors

This is the Alpha version.

Prerequisites

  1. Python >= 3.5 (recommended is Python 3.8)
  2. Install requirements.txt:
pip install -r requirements.txt

Repository

https://github.com/ziemowit-s/neuronpp

Neuron++ installation

  • Locally:
python setup.py bdist_wheel
  • Through pip and GitHub:
pip install git+https://github.com/ziemowit-s//neuronpp
  • Bear in mind that if you install the library through the pip you will have access to all features of NEURON++, however additionally provided cell models from other publications (listed below) will not work correctly, unless you download the 'commons/' folder from the GitHub repository and change paths in 'examples/' indicating the 'commons/'.

So if you want to work with those predefined models it is recommended to clone the repository from GitHub rather than install through pip.

Predefined Cell models

This repository contains the basic cell objects:

  • Cell class
  • HocCell - the experimental `HOC class which loads HOC based cell model.

The repository also contains some predefined cell models from ModelDB (https://senselab.med.yale.edu/modeldb)

  • All of those models are located in the cells/ folder.
  • If you want to create your own model it is recommended to use:
    • Cell
    • HocCell - if you want to reuse existing HOC model

The list of predefined cell models:

MOD files of all of those models are located in the commons/mods/ folder. Combe 2018 model and Graham 2014 model additionally have HOC files located in the commons/hocmodels/ folder.

Automatic MOD compilation

Automatic (recommended)

Cell() object has a compile_path param which allows to specify paths which contain MOD files.

You don't need to compile files externally, if you provided appropriate pathways it will be done automatically before each run.

Introduction to NEURON++

All examples are located in the examples/ folder

Please bear in mind that due to substantial updates some examples may not work as described here in the README, also some additional features have been added which are not listed here. We will update readme soon.

Basics

  • create a cell:
 cell = Cell(name="mycell")
  • load SWCor ASC morphology:
 cell.load_morpho(filepath='commons/morphologies/asc/cell2.asc')
  • load HOC-based cell (and HOC-based morphology) model to the Cell object:
    • allows to work with most modelDB single cell models
    • it will auto load all sections and point processes for further usage and/or filtering
    • This is an experimental feature, so currently works ONLY with HOC models which define a single cell
 cell = HocCell(name="mycell")
 cell.load_hoc("your_cell_model.hoc")
  • if the HOC cell model is defined as a Template - just specify the cell_template_name param:
   cell = HocCell(name="mycell")
   cell.load_hoc("Ebner2019_minimum_load/load_model.hoc", hoc_template_name="L5PCtemplate")
  • create and connect sections:
 cell.add_seg("soma", diam=20, l=20, nseg=10)
 cell.add_seg("dend", diam=2, l=100, nseg=10)
 cell.connect_secs(child="dend", parent="soma")
  • add NEURON mechanisms (default or MOD-based):
 cell.insert("pas")
 cell.insert("hh")
  • define simulation and run:
 sim = Simulation(init_v=-65, warmup=20)
 sim.run(runtime=500)
  • add IClamp:
sections = cell.filter_secs("soma")
soma = sections[0]

ic = IClamp(segment=soma(0.5))
ic.debug(delay=100, dur=10, amp=0.1)

Easy filters

You can obtain any part of the cell by string or regular expression filtering

  • filter section of the cell by string:
  # Assuming you have sections dend[0]...dend[100] it will return all of them
  sections = cell.filter_secs(name="dend")
  • filter by string separated by coma:
  # Each coma function as OR between strings
  sections = cell.filter_secs(name="apic[1],apic[50]")
  • filter section of the cell by regular expression:
  # Assuming you have sections dend[0]...dend[100] and apic[0]...apic[100] it will return all of them
  sections = cell.filter_secs(name="regex:(apic)|(basal)")
  • return synapses of type 'ExpSyn' located in all heads sections (of the dendritic spines)
 cell.filter_synapses(mod_name="ExpSyn", name="head")
  • custom function-based filtering:
    • pass function to the obj_filter param. eg. (lambda expression) returns sections which name contains 'apic' or their distance > 1000 um from the soma:
 soma = cell.filter_secs("soma")
 cell.filter_secs(obj_filter=lambda o: 'apic' in o.name or h.distance(soma(0.5), o(0.5)) > 1000)
  • field-based filtering with custom function. eg. (lambda expression) returns sections which parent's name contains less than 10 characters
cell.filter_secs(parent=lambda o: len(o.parent.name) < 10)

All filter functions available in default cell object Cell:

  • filter_secs - obtain sections
  • filter_point_processes - obtain point processes
  • filter_netcons - obtain netcons
  • filter_synapses - obtain single synapse, which is a point_process and netcon wrapper
  • filter_synaptic_group - obtain a group of synapses which is many single synapses grouped together

Define Synapses

If you define source param - it will stimulate synapse based on the stimulation from the source. If source is None - there will be no source and no netcon, however you can add those later or use such synapse for the Experiment.

  • add single synapse:
 cell = Cell(name="cell")
 soma = cell.filter_secs(name="soma")
 cell.add_sypanse(source=None, mod_name="Syn4P", seg=soma(0.5), netcon_weight=0.01, delay=1)
  • add many spines to the provided sections:
cell = Cell(name="cell")
dendrites = cell.filter_secs(name="dend")
cell.add_randuniform_spines(spine_number=10, head_nseg=10, neck_nseg=10, secs=dendrites)
  • add many synapses with spines (1 synapse/spine) in a single function to the provided sections:
 cell = Cell(name="cell")
 dendrites = cell.filter_secs(name="dend")
 syns = cell.add_random_synapses_with_spine(source=None, secs=dendrites, mod="ExpSyn",
                                            netcon_weight=0.01, delay=1, number=10)
  • define NetStim (or VecStim) and pass it to synapses as a source while creating:
  netstim = NetStimCell(name="netst")
  stim = netstim.add_netstim(start=300, number=5, interval=10)
  
  cell = Cell(name="cell")
  soma = cell.filter_secs(name="soma")
  cell.add_sypanse(source=stim, seg=soma(0.5), mod_name="ExpSyn", netcon_weight=0.01, delay=1)
  • send external input to the synapse by calling synaptic event:
    • every synapse can be stimulated by making event, however a good practice is to define source=None for such synapses.
 cell = Cell(name="cell")
 soma = cell.filter_secs(name="soma")
 syns = cell.add_sypanse(source=None, seg=soma(0.5), mod_name="ExpSyn", 
                         netcon_weight=0.01, delay=1)
                           
 sim = Simulation(init_v=-55, warmup=20)
 
 syns[0].make_event(10)
 syns[0].make_event(20)
 
 sim.run(runtime=100)
  • Add another source to the previously created synapse:
  synapse.add_netcon(source=None, weight=0.035, threshold=15, delay=2)

Record and plot

  • make spike detector for the cell:
  cell.make_spike_detector()

  sim = RunSim(init_v=-65)
  sim.run(runtime=500)

  cell.plot_spikes()
  • record variables from sections and point_processes:
 # record section's voltage
 soma = cell.filter_secs(name="soma")
 rec_v = Record(soma(0.5), variables="v")

 # record synaptic (point_process) wariables (weight 'w')
 point_processes = cell.filter_point_processes(mod_name="Syn4P", name="dend")
 rec_syn = Record(point_processes, variables="w") 

 sim = Simulation(init_v=-65, warmup=20, with_neuron_gui=True)
 sim.run(runtime=500)
  • make plots and export recorded variables to CSV:
 rec_v.plot()
 rec_syn.plot()
 plt.show()
 rec_v.to_csv("vrec.csv")
  • make shape plot of the cell in NEURON GUI:
# show 'cai' propagation in range 0-0.01 uM 
make_shape_plot(variable="cai", min_val=0, max_val=0.01)

# show 'v' propagation in range -70-40 mV
make_shape_plot(variable="v", min_val=-70, max_val=40)

Replicate in vitro experiments

  • define experimetal protocols, eg. STDP protocol:
  soma = cell.filter_secs("soma")
  syn = cell.filter_synapses(tag="my_synapse")

  experiment = Experiment(iti=40)
  experiment.add_epsp(num=3, synapse=syn, init=20, interval=20, weight=0.02)
  experiment.add_iclamp(num=3, segment=soma(0.5), init=60, interval=20, dur=3, amp=1.6)
  experiment.build()

Populations of neurons

Create a population of many neurons of the same type and connect them between populations:

  • Create a template cell function:

      def cell_function():
      cell = Cell(name="cell")
      morpho_path = os.path.join(path, "..", "commons/morphologies/swc/my.swc")
      cell.load_morpho(filepath=morpho_path)
      cell.insert("pas")
      cell.insert("hh")
      cell.make_spike_detector(seg=cell.filter_secs("soma")(0.5))
      return cell
    • Create stimulation:
      # Create NetStim
  netstim = NetStimCell("stim").add_netstim(start=21, number=100, interval=2)

  # Define weight distribution for both: NetStim->population1 and population1->population2
  weight_dist = NormalTruncatedDist(mean=0.1, std=0.2)
    • Define population 1:
      pop1 = Population("pop_1")
  pop1.add_cells(num=3, cell_function=cell_function)

  # create 10 synapses on population 2 per NetStim object (single NetStim here)
  connector = pop1.connect(syn_num_per_cell_source=10)
  connector.set_source(netstim)

  # choose all dendrites as potential targets for synaptic choice
  targets = [d(0.5) for c in pop1.cells for d in c.filter_secs("dend")]
  connector.set_target(targets)

  # Make synapse
  syn_adder = connector.add_synapse("Exp2Syn")
  syn_adder.add_netcon(weight=weight_dist)
  # change tau1 and tau2 for Exp2Syn synapses
  syn_adder.add_point_process_params(tau1=0.1, tau2=2)
    • Build connections and define that you want to record from the population
      • By default record() method make records of: voltage variable in soma(0.5)
      connector.build()
  pop1.record()
    • Create population 2:
      pop2 = Population("pop_2")
  pop2.add_cells(num=3, cell_function=cell_function)
    • Define connections between pop1 and pop2 where weights will be chosen from the Normal Truncated Distribution:
      # create 5 synapses per single cell in population 1
  connector = pop2.connect(syn_num_per_cell_source=5)

  source = [c.filter_secs("soma")(0.5) for c in pop1.cells]
  connector.set_source(source)

  # choose all dendrites as potential targets for synaptic choice
  targets = [d(0.5) for c in pop2.cells for d in c.filter_secs("dend")]
  connector.set_target(targets)

  # Make synapse
  syn_adder = connector.add_synapse("Exp2Syn")
  syn_adder.add_netcon(weight=weight_dist)
  # change tau1 and tau2 for Exp2Syn synapses
  syn_adder.add_point_process_params(tau1=0.1, tau2=2)
    • Build connections and define that you want to record from the population
    • By default record() method make records of: voltage variable in soma(0.5)
      connector.build()
  pop2.record()
    • Ryn simulation and plot activities:
      sim = Simulation(init_v=-70, warmup=20)
  for i in range(1000):
      sim.run(runtime=1)
      pop1.plot(animate=True)
      pop2.plot(animate=True)
    • Create (experimental) interactive graph of connected populations which allows you to see and move nodes in the web browser:
    show_connectivity_graph(pop1.cells + pop2.cells)

Network Graph

Debug synapses and point processes

Debug any cell and synapse on interactive plot.

  • By pressing a key defined as stim_key param (default is w) you can stimulate synapses provided to the Debugger
  • It allows to easily plot synaptic weight (defined as MOD's RANGE variable) to see how the plasticity behaves in real time
  from neuronpp.cells.cell import Cell
  from neuronpp.utils.synaptic_debugger import SynapticDebugger
  
  # Prepare cell
  cell = Cell("cell")
  soma = cell.add_sec("soma", diam=20, l=20, nseg=100)
  cell.insert("pas")
  cell.insert("hh")
  
  syn1 = cell.add_synapse(source=None, netcon_weight=0.002, seg=soma(0.1), mod_name="Exp2Syn")
  syn2 = cell.add_synapse(source=None, netcon_weight=0.002, seg=soma(0.9), mod_name="Exp2Syn")
  syn3 = cell.add_synapse(source=None, netcon_weight=0.002, seg=soma(0.5), mod_name="Exp2Syn")
  
  # Debug
  debug = SynapticDebugger(init_v=-70, warmup=10, delay_between_steps=15)
  debug.add_syn(syn1, key_press='1', plot=True, syn_variables='i')
  debug.add_syn(syn2, key_press='2')
  debug.add_syn(syn3, key_press='3')
  debug.add_seg(soma(0.5))
  debug.debug_interactive()

Debugger Example of Ebner et al. 2019 model of synaptic weight (variable w) changing based on synaptic stimulation on demand by pressing key "w" on the keyboard:

  • variable w: weight on the synapse.
  • variable v: voltage on the soma.
  • Pressing "w" on the keyboard produce a synaptic event.