Neuron_Reduce provides an analytical method for reducing neuron model complexity. It enables the mapping of synapses and active ion channels to a computationally simpler model while accelerating simulation speed by up to 200-fold for inputs consisting of thousands of dendritic synapses.
Full details are available in the accompanied paper: An efficient analytical reduction of detailed nonlinear neuron models. Nat. Commun., 11 (2020), p. 288 - https://www.nature.com/articles/s41467-019-13932-6
pip install --user neuron_reduce
The following code show the main function that is used to reduce a complex cell.
complex_cell # The model cell
synapses_list # A list of all synapse on this cell
netcon_list # A list of all netcons for the synapses on the cell
import neuron_reduce
reduced_cell, synapses_list, netcons_list = neuron_reduce.subtree_reductor(complex_cell, synapses_list, netcons_list)
Copy example folder from github
git clone https://github.com/orena1/neuron_reduce.git
Go to example folder
cd neuron_reduce
cd example
nrnivmodl mod #compile the mod files
Open python and run the following code
from __future__ import division
from neuron import gui,h
import numpy as np
import neuron_reduce
import time
import matplotlib.pyplot as plt
#Create a L5_PC model
h.load_file('L5PCbiophys3.hoc')
h.load_file("import3d.hoc")
h.load_file('L5PCtemplate.hoc')
complex_cell = h.L5PCtemplate('cell1.asc')
h.celsius = 37
h.v_init = complex_cell.soma[0].e_pas
#Add synapses to the model
synapses_list, netstims_list, netcons_list, randoms_list = [], [], [] ,[]
all_segments = [i for j in map(list,list(complex_cell.apical)) for i in j] + [i for j in map(list,list(complex_cell.basal)) for i in j]
len_per_segment = np.array([seg.sec.L/seg.sec.nseg for seg in all_segments])
rnd = np.random.RandomState(10)
for i in range(10000):
seg_for_synapse = rnd.choice(all_segments, p=len_per_segment/sum(len_per_segment))
synapses_list.append(h.Exp2Syn(seg_for_synapse))
if rnd.uniform()<0.85:
e_syn, tau1, tau2, spike_interval, syn_weight = 0, 0.3, 1.8, 1000/2.5, 0.0016
else:
e_syn, tau1, tau2, spike_interval, syn_weight = -86, 1, 8, 1000/15.0, 0.0008
#set synaptic varibales
synapses_list[i].e, synapses_list[i].tau1, synapses_list[i].tau2 = e_syn, tau1, tau2
#set netstim variables
netstims_list.append(h.NetStim())
netstims_list[i].interval, netstims_list[i].number, netstims_list[i].start, netstims_list[i].noise = spike_interval, 9e9, 100, 1
#set random
randoms_list.append(h.Random())
randoms_list[i].Random123(i)
randoms_list[i].negexp(1)
netstims_list[i].noiseFromRandom(randoms_list[i])
#set netcon varibales
netcons_list.append(h.NetCon(netstims_list[i], synapses_list[i] ))
netcons_list[i].delay, netcons_list[i].weight[0] = 0, syn_weight
#Simulate the full neuron for 1 seconds
soma_v = h.Vector()
soma_v.record(complex_cell.soma[0](0.5)._ref_v)
time_v = h.Vector()
time_v.record(h._ref_t)
h.tstop = 1000
st = time.time()
h.run()
print('complex cell simulation time {:.4f}'.format(time.time()-st))
complex_cell_v = list(soma_v)
#apply Neuron_Reduce to simplify the cell
reduced_cell, synapses_list, netcons_list = neuron_reduce.subtree_reductor(complex_cell, synapses_list, netcons_list, reduction_frequency=0, total_segments_manual=-1)
for r in randoms_list:r.seq(1) #reset random
#Running the simulation again but now on the reduced cell
st = time.time()
h.run()
print('reduced cell simulation time {:.4f}'.format(time.time()-st))
reduced_celll_v = list(soma_v)
#plotting the results
plt.figure()
plt.plot(time_v, complex_cell_v, label='complex cell')
plt.plot(time_v, reduced_celll_v, label='redcued cell')
plt.show()
O. Amsalem, G. Eyal, N. Rogozinski, M. Gevaert, P. Kumbhar, F. Schürmann, I. Segev. An efficient analytical reduction of detailed nonlinear neuron models. Nat. Commun., 11 (2020), p. 288