#
# MSN model
#
# Robert Lindroos
#
# Original version by
# Alexander Kozlov <akozlov@kth.se>
# Kai Du <kai.du@ki.se>
#
from __future__ import print_function, division
import mpi4py as MPI
import sys
import json
import numpy as np
from neuron import h
from math import exp
import pickle
import os
mod = "./mod/"
defparams = "./params-rob.json"
morphology = "./morphology/"
with open('./substrates.json') as file:
SUBSTRATES = json.load(file)
#h.nrn_load_dll('/pdc/vol/neuron/7.4-py27/x86_64/.libs/libnrnmech.so')
h.load_file('stdlib.hoc')
h.load_file('import3d.hoc')
# global result dict
RES = {}
# Distributions:
'''
T-type Ca: (1.0/(1+np.exp((v-70)/-4.5)))
naf (den): (0.1 + 0.9/(1 + np.exp((v-60.0)/10.0)))
'''
# 1, 0.8, 2.5,60,-25
#calculate_distribution(d3, dist, a4, a5, a6, a7, g8)
def calculate_distribution(d3, dist, a4, a5, a6, a7, g8):
# d3 is the distribution type:
# 0 linear, 1 sigmoid, 2 exponential
# 3 step for absolute distance (in microns)
# dist is the somatic distance
# a4-7 is distribution parameters
# g8 is the maximal conductance
if d3 == 0:
value = a4 + a5*dist
elif d3 == 1:
value = a4 + a5/(1 + exp((dist-a6)/a7) )
elif d3 == 2:
value = a4 + a5*exp((dist-a6)/a7)
elif d3 == 3:
if (dist > a6) and (dist < a7):
value = a4
else:
value = a5
if value < 0:
value = 0
value = value*g8
return value
def alpha(tstart, gmax, tau):
# calc and returns a "magnitude" using an alpha function -> used for [DA] in cascade
v = 1 - (h.t - tstart) / tau
e = exp(v)
mag = gmax * (h.t - tstart) / tau * e
return mag
def save_vector(t, v, outfile):
with open(outfile, "w") as out:
for time, y in zip(t, v):
out.write("%g %g\n" % (time, y))
def calc_rand_Modulation(mod_list, range_list=False):
'''
calc random modulation values between 0 and 2 (i.e. max +/- 100%).
the random values are drawn in such a way that low modulations are more likely than high.
this is achived by drawing two uniform random numbers and subtracts one from the other,
giving values ranging from -1 to 1. By adding one we end upp at the wanted value.
If a range_list is supplied the range of the values can be shifted from [0,2] to any
other range. The range list must have the same length as mod_list and hold lists of
[min, max] values.
'''
mod_factors = []
A=0
B=2
for i,channel in enumerate(mod_list):
factor = 1.0 + ( np.random.uniform() - np.random.uniform() )
if range_list:
a = range_list[i][0]
b = range_list[i][1]
factor = (b-a) / (B-A) * (factor-A) + a
mod_factors.append(factor)
return mod_factors
# 'save/'+
def save_obj(obj, name ):
with open(name + '.pkl', 'wb') as f:
pickle.dump(obj, f, pickle.HIGHEST_PROTOCOL)
def load_obj(name ):
with open('save/' + name + '.pkl', 'rb') as f:
return pickle.load(f)
def getSpikedata_x_y(x,y):
'''
There's probably a Neuron function for this--use instead?
getSpikedata_x_y(x,y) -> return count
Extracts and returns the number of spikes from spike trace data.
# arguments
x = time vector
y = vm vector
# returns
count = number of spikes in trace (int)
# extraction algorithm
-threshold y and store list containing index for all points larger than 0 V
-sorts out and counts the index that are the first one(s) crossing the threshold, i.e.
the first index of each spike. This is done by looping over all index and check if
the index is equal to the previous index + 1. If not it is the first index of a
spike.
If no point is above threshold in the trace the function returns 0.
'''
count = 0
spikes = []
# pick out index for all points above zero potential for potential trace
spikeData = [i for i,v in enumerate(y) if v > 0]
# if no point above 0
if len(spikeData) == 0:
return spikes
else:
# pick first point of each individual transient (spike)...
for j in range(0, len(spikeData)-1):
if j==0:
count += 1
spikes.append(x[spikeData[j]])
# ...by checking above stated criteria
elif not spikeData[j] == spikeData[j-1]+1:
count += 1
spikes.append(x[spikeData[j]])
return spikes
def make_random_synapse(ns, nc, Syn, sec, x, \
Type='glut', \
NS_start=1, \
NS_interval=1000.0/18.0, \
NS_noise=1.0, \
NS_number=1000, \
S_AN_ratio=1.0, \
S_tau_dep=100, \
S_U=1, \
S_e=-60, \
S_tau1=0.25, \
S_tau2=3.75, \
NC_delay=1, \
NC_conductance=0.6e-3, \
NC_threshold=0.1 ):
# create/set synapse in segment x of section
if Type == 'glut':
key = sec
Syn[key] = h.tmGlut(x, sec=sec)
Syn[key].nmda_ratio = S_AN_ratio
Syn[key].tauR = S_tau_dep
Syn[key].U = S_U
elif Type in ['expSyn2', 'tmgabaa', 'gaba']:
key = sec.name() + '_gaba'
if Type == 'expSyn2':
Syn[key] = h.Exp2Syn(x, sec=sec)
Syn[key].tau1 = S_tau1
Syn[key].tau2 = S_tau2
elif Type == 'tmgabaa':
Syn[key] = h.tmGabaA(x, sec=sec)
Syn[key].e = S_e
# create NetStim object
ns[key] = h.NetStim()
ns[key].start = NS_start
ns[key].interval = NS_interval # mean interval between two spikes in ms
ns[key].noise = NS_noise
ns[key].number = NS_number
# create NetCon object
nc[key] = h.NetCon(ns[sec],Syn[sec])
nc[key].delay = NC_delay
nc[key].weight[0] = NC_conductance
nc[key].threshold = NC_threshold
def set_rand_synapse(channel_list, base_mod, max_mod, range_list=[[0.75,1.5],[0.75,1.5]]):
syn_fact = calc_rand_Modulation(channel_list, range_list=range_list)
# normalize factors to max-value of pointer substrate
normalized_factors = []
for i,mech in enumerate(channel_list):
normalized_factors.append( (syn_fact[i] - 1) / (max_mod - base_mod) )
return syn_fact, normalized_factors
# ======================= the MSN class ==================================================
class MSN:
def __init__(self, params=defparams, factors=None):
Import = h.Import3d_SWC_read()
Import.input(morphology + 'latest_WT-P270-20-14ak.swc')
imprt = h.Import3d_GUI(Import, 0)
imprt.instantiate(None)
h.define_shape()
# h.cao0_ca_ion = 2 # default in nrn
h.celsius = 35
self._create_sectionlists()
self._set_nsegs()
self.v_init = -80
for sec in self.allseclist:
sec.Ra = 150
sec.cm = 1.0
sec.insert('pas')
#sec.g_pas = 1e-5 # set using json file
sec.e_pas = -70 # -73
for sec in self.somalist:
sec.insert('naf')
sec.insert('kaf')
sec.insert('kas')
sec.insert('kdr')
sec.insert('kir')
sec.ena = 50
sec.ek = -85 # -90
sec.insert('cal12')
sec.insert('cal13')
sec.insert('car')
sec.insert('cadyn')
sec.insert('caldyn')
sec.insert('sk')
sec.insert('bk')
sec.insert('can')
#sec.kb_cadyn = 200.
for sec in self.axonlist:
sec.insert('naf')
#sec.insert('kaf')
sec.insert('kas')
#sec.insert('kdr')
#sec.insert('kir')
sec.ena = 50
sec.ek = -85 # -90
for sec in self.dendlist:
sec.insert('naf')
sec.insert('kaf')
sec.insert('kas')
sec.insert('kdr')
sec.insert('kir')
sec.ena = 50
sec.ek = -85 # -90
sec.insert('cal12')
sec.insert('cal13')
sec.insert('car')
sec.insert('cadyn')
sec.insert('caldyn')
sec.insert('sk')
sec.insert('bk')
sec.insert('cat32')
sec.insert('cat33')
with open(params) as file:
par = json.load(file)
self.distribute_channels("soma", "g_pas", 0, 1, 0, 0, 0, float(par['g_pas_all']['Value']))
self.distribute_channels("axon", "g_pas", 0, 1, 0, 0, 0, float(par['g_pas_all']['Value']))
self.distribute_channels("dend", "g_pas", 0, 1, 0, 0, 0, float(par['g_pas_all']['Value']))
self.distribute_channels("soma", "gbar_naf", 0, 1, 0, 0, 0, float(par['gbar_naf_somatic']['Value']),factors=factors)
self.distribute_channels("soma", "gbar_kaf", 0, 1, 0, 0, 0, float(par['gbar_kaf_somatic']['Value']))
self.distribute_channels("soma", "gbar_kas", 0, 1, 0, 0, 0, float(par['gbar_kas_somatic']['Value']))
self.distribute_channels("soma", "gbar_kdr", 0, 1, 0, 0, 0, float(par['gbar_kdr_somatic']['Value']))
self.distribute_channels("soma", "gbar_kir", 0, 1, 0, 0, 0, float(par['gbar_kir_somatic']['Value']))
self.distribute_channels("soma", "gbar_sk", 0, 1, 0, 0, 0, float(par['gbar_sk_somatic']['Value']))
self.distribute_channels("soma", "gbar_bk", 0, 1, 0, 0, 0, float(par['gbar_bk_somatic']['Value']))
#self.distribute_channels("axon", "gbar_naf", 0, 1, 0, 0, 0, float(par['gbar_naf_somatic']['Value']),factors=factors)
self.distribute_channels("axon", "gbar_naf", 3, 1, 1.1, 30, 500, float(par['gbar_naf_axonal']['Value']),factors=factors)
#self.distribute_channels("axon", "gbar_naf", 3, 1, 1.1, 20, 500, float(par['gbar_naf_axonal']['Value']))
#self.distribute_channels("dend", "gbar_naf", 1, 1, 1.2, 30, -5, float(par['gbar_naf_axonal']['Value']))
self.distribute_channels("axon", "gbar_kas", 0, 1, 0, 0, 0, float(par['gbar_kas_axonal']['Value']))
self.distribute_channels("dend", "gbar_naf", 1, 0.1, 0.9, 60.0, 10.0, float(par['gbar_naf_basal']['Value']),factors=factors)
self.distribute_channels("dend", "gbar_kaf", 1, 1, 0.5, 120, -30, float(par['gbar_kaf_basal']['Value']))
#self.distribute_channels("dend", "gbar_naf", 0, 1, -0.0072, 0, 0, float(par['gbar_naf_basal']['Value']))
#self.distribute_channels("dend", "gbar_kaf", 0, 1, 0.0167, 0, 0, float(par['gbar_kaf_basal']['Value']))
self.distribute_channels("dend", "gbar_kas", 2, 1, 9.0, 0.0, -5.0, float(par['gbar_kas_basal']['Value']))
self.distribute_channels("dend", "gbar_kdr", 0, 1, 0, 0, 0, float(par['gbar_kdr_basal']['Value']))
self.distribute_channels("dend", "gbar_kir", 0, 1, 0, 0, 0, float(par['gbar_kir_basal']['Value']))
self.distribute_channels("dend", "gbar_sk", 0, 1, 0, 0, 0, float(par['gbar_sk_basal']['Value']))
self.distribute_channels("dend", "gbar_bk", 0, 1, 0, 0, 0, float(par['gbar_bk_basal']['Value']))
self.distribute_channels("soma", "pbar_cal12", 0, 1, 0, 0, 0, 1e-5)
self.distribute_channels("soma", "pbar_cal13", 0, 1, 0, 0, 0, 1e-6)
self.distribute_channels("soma", "pbar_car", 0, 1, 0, 0, 0, 1e-4)
self.distribute_channels("soma", "pbar_can", 0, 1, 0, 0, 0, 3e-5)
#self.distribute_channels("soma", "kb_cadyn", 0, 1, 0, 0, 0, 200.0)
self.distribute_channels("dend", "pbar_cal12", 0, 1, 0, 0, 0, 1e-5)
self.distribute_channels("dend", "pbar_cal13", 0, 1, 0, 0, 0, 1e-6)
self.distribute_channels("dend", "pbar_car", 0, 1, 0, 0, 0, 1e-4)
self.distribute_channels("dend", "pbar_cat32", 1, 0, 1.0, 70.0, -4.5, 1e-7)
self.distribute_channels("dend", "pbar_cat33", 1, 0, 1.0, 70.0, -4.5, 1e-8)
def _create_sectionlists(self):
self.allsecnames = []
self.allseclist = h.SectionList()
for sec in h.allsec():
self.allsecnames.append(sec.name())
self.allseclist.append(sec=sec)
self.nsomasec = 0
self.somalist = h.SectionList()
for sec in h.allsec():
if sec.name().find('soma') >= 0:
self.somalist.append(sec=sec)
if self.nsomasec == 0:
self.soma = sec
self.nsomasec += 1
self.axonlist = h.SectionList()
for sec in h.allsec():
if sec.name().find('axon') >= 0:
self.axonlist.append(sec=sec)
self.dendlist = h.SectionList()
for sec in h.allsec():
if sec.name().find('dend') >= 0:
self.dendlist.append(sec=sec)
def _set_nsegs(self):
for sec in self.allseclist:
sec.nseg = 2*int(sec.L/40.0)+1
for sec in self.axonlist:
sec.nseg = 2 # two segments in axon initial segment
def _max_dist(self, axon_excluding=True):
h.distance(sec=self.soma)
dmax = 0
for sec in self.allseclist:
if axon_excluding and sec.name().find('axon') == 0: continue
dmax = max(dmax, h.distance(1, sec=sec))
return dmax
def distribute_channels(self, as1, as2, d3, a4, a5, a6, a7, g8, factors=None):
h.distance(sec=self.soma)
dmax = self._max_dist()
for sec in self.allseclist:
if sec.name().find(as1) >= 0:
for seg in sec:
dist = h.distance(seg.x, sec=sec)
val = calculate_distribution(d3, dist, a4, a5, a6, a7, g8)
cmd = 'seg.%s = %g' % (as2, val)
exec(cmd)
#names = ['mVhalf_naf', 'hVhalf_naf', 'mSlope_naf', 'hSlope_naf']
names = ['taum_naf', 'tauh_naf', 'taun_naf']
if factors:
for i,factor in enumerate(factors):
print('updating factors! ', factors, factor, names[i])
# add minus here if running with slope and half activation; cmd = 'seg.%s = %g' % (names[i], -factor)
cmd = 'seg.%s = %g' % (names[i], factor)
exec(cmd)
#=========================================================================================
def main(par="./params-msn.json", \
sim='vm', \
amp=0.265, \
run=None, \
modulation=1, \
simDur=7000, \
stimDur=900, \
factors=None, \
section=None, \
randMod=None, \
testMode=False, \
target=None, \
chan2mod=['naf', 'kas', 'kaf', 'kir', 'cal12', 'cal13', 'can'] ):
print(locals())
# initiate cell
cell = MSN(params=par, factors=factors)
# set cascade ---- move to MSN def?
casc = h.D1_reduced_cascade2_0(0.5, sec=cell.soma) # other cascades also possible...
if target:
cmd = 'pointer = casc._ref_'+target
exec(cmd)
base_mod = SUBSTRATES[target][0]
max_mod = SUBSTRATES[target][1]
else:
pointer = casc._ref_Target1p #Target1p #totalActivePKA (if full cascade used)
base_mod = casc.init_Target1p
max_mod = 2317.1
# cAMP; init: 38.186016
# set edge of soma as reference for distance
h.distance(1, sec=h.soma[0])
# set current injection
stim = h.IClamp(0.5, sec=cell.soma)
stim.amp = amp
stim.delay = 100
stim.dur = stimDur # 2ms 2nA to elicit single AP, following Day et al 2008 Ca dyn
# record vectors
tm = h.Vector()
tm.record(h._ref_t)
vm = h.Vector()
vm.record(cell.soma(0.5)._ref_v)
# substrates
pka = h.Vector()
pka.record(casc._ref_Target1p)
camp = h.Vector()
camp.record(casc._ref_cAMP)
gprot = h.Vector()
gprot.record(casc._ref_D1RDAGolf) #D1RDAGolf
gbg = h.Vector()
gbg.record(casc._ref_Gbgolf) #Gbgolf
# peak n dipp parameters
da_peak = 500 # concentration [nM]
da_tstart = 1000 # stimulation time [ms]
da_tau = 500 # time constant [ms]
tstop = simDur # [ms]
# all channels to modulate
mod_list = ['naf', 'kas', 'kaf', 'kir', 'cal12', 'cal13', 'can' ]
not2mod = [] #['kaf']
# find channels that should not be modulated
for chan in mod_list:
if chan not in chan2mod:
not2mod.append(chan)
# for random modulation: modValues = np.arange(0.1, 2.0, 0.1) -------------------------
if randMod == 1:
# new factors every run
mod_fact = calc_rand_Modulation(mod_list, range_list=[[0.6,0.8], \
[0.65,0.85], \
[0.78,0.82], \
[0.85,1.25], \
[1.0,2.0], \
[1.0,2.0], \
[0.0,1.0]] )
# keep old factors
'''
if run == 0:
mod_fact = calc_rand_Modulation(mod_list)
else:
mod_fact = RES['factors']
'''
else:
mod_fact = [ 0.8, 0.8, 0.8, 1.25, 2.0, 2.0, 0.5 ]
# noormalize factors to target values seen in simulation
factors = []
for i,mech in enumerate(mod_list):
factor = (mod_fact[i] - 1) / (max_mod - base_mod) #2317.1
factors.append(factor)
#print(mech, mod_fact[i], factor) # --------------------------------------------------------
# set pointers
for sec in h.allsec():
for seg in sec:
# naf and kas is in all sections
h.setpointer(pointer, 'pka', seg.kas )
h.setpointer(pointer, 'pka', seg.naf )
if sec.name().find('axon') < 0:
# these channels are not in the axon section
h.setpointer(pointer, 'pka', seg.kaf )
h.setpointer(pointer, 'pka', seg.cal12 )
h.setpointer(pointer, 'pka', seg.cal13 )
h.setpointer(pointer, 'pka', seg.kir )
#h.setpointer(pointerc, 'pka', seg.car )
if sec.name().find('soma') >= 0:
# can is only distributed to the soma section
h.setpointer(pointer, 'pka', seg.can )
# synaptic modulation ================================================================
if sim == 'synMod':
# draw random modulation factors (intervals given by range_list[[min,max]]
glut_f, glut_f_norm = set_rand_synapse(['amp', 'nmd'], base_mod, max_mod, \
range_list=[[0.9,1.6], [0.9,1.6]] )
gaba_f, gaba_f_norm = set_rand_synapse(['gab'], base_mod, max_mod, \
range_list=[[0.6,1.4]] )
syn_fact = glut_f + gaba_f
I_d={}
ns = {}
nc = {}
Syn = {}
for sec in h.allsec():
if sec.name().find('dend') >= 0:
# create a glut synapse
make_random_synapse(ns, nc, Syn, sec, 0.5, \
NS_interval=1000.0/17.0, \
NC_conductance=0.15e-3, \
S_tau_dep=100 )
# create a gaba synapse
make_random_synapse(ns, nc, Syn, sec, 0.0, \
Type='tmgabaa', \
NS_interval=1000.0/4.0, \
NC_conductance=0.45e-3 )
# set pointer(s)
h.setpointer(pointer, 'pka', Syn[sec])
h.setpointer(pointer, 'pka', Syn[sec.name()+'_gaba'])
# set (random?) modulation
Syn[sec].base = base_mod
#randMod?
if randMod == 1:
Syn[sec].f_ampa = glut_f_norm[0]
Syn[sec].f_nmda = glut_f_norm[1]
else:
Syn[sec].f_ampa = 0
Syn[sec].f_nmda = 0
if randMod == 2:
Syn[sec.name()+'_gaba'].base = base_mod
Syn[sec.name()+'_gaba'].f_gaba = gaba_f_norm[0]
else:
Syn[sec.name()+'_gaba'].f_gaba = 0
'''
# record synaptic current from synapse
I_d[sec.name()] = h.Vector()
I_d[sec.name()].record(Syn[sec]._ref_i)
I_d[sec.name()+'_gaba'] = h.Vector()
I_d[sec.name()+'_gaba'].record(Syn[sec.name()+'_gaba']._ref_i)
'''
elif sec.name().find('axon') >= 0:
continue
if randMod == 1:
for seg in sec:
for mech in seg:
if mech.name() in not2mod:
mech.factor = 0.0
print(mech.name(), 'and channel:', not2mod, mech.factor, sec.name())
elif mech.name() in mod_list:
mech.base = base_mod
index = mod_list.index( mech.name() )
mech.factor = factors[index]
# dynamical modulation
elif sim == 'modulation':
print('inne ', sim)
for sec in h.allsec():
for seg in sec:
for mech in seg:
# if this first statement is active the axon will not be modulated
'''if sec.name().find('axon') >= 0 \
and mech.name() in mod_list:
mech.factor = 0.0
print(sec.name(), seg.x, mech.name() )'''
if mech.name() in not2mod:
mech.factor = 0.0
print(mech.name(), 'and channel:', not2mod, mech.factor, sec.name())
elif mech.name() in mod_list:
mech.base = base_mod
index = mod_list.index( mech.name() )
mech.factor = factors[index]
# static modulation
elif sim == 'directMod':
print('inne ', sim)
for sec in h.allsec():
for seg in sec:
for mech in seg:
if mech.name() in mod_list:
if sec.name().find('axon') < 10: # 0 no axon modulated; 10 all sections
factor = mod_fact[mod_list.index(mech.name() )]
if mech.name() in not2mod:
mech.factor = 0.0
elif mech.name()[0] == 'c':
pbar = mech.pbar
mech.pbar = pbar * factor
#print(''.join(['setting pbar ', mech.name(), ' ', str(factor) ]) )
else:
gbar = mech.gbar
mech.gbar = gbar * factor
#if seg.x < 0.2:
#print(''.join(['setting gbar ', mech.name(), ' ', str(factor), ' ', str(gbar), ' ', str(factor*gbar) ]) )
else:
print(sec.name(), seg.x, sec.name().find('axon'))
# solver------------------------------------------------------------------------------
cvode = h.CVode()
h.finitialize(cell.v_init)
# run simulation
while h.t < tstop:
if modulation == 1:
if h.t > da_tstart:
# set DA and ACh values (using alpha function)
casc.DA = alpha(da_tstart, da_peak, da_tau)
#casc.ACh = ach_base - alpha(ach_tstart, ach_base, ach_tau)
h.fadvance()
# save output
if sim in ['vm', 'directMod', 'modulation', 'synMod']:
if testMode:
mod_fact = mod_fact + syn_fact
ID = ''
for i,mech in enumerate(mod_list+syn_mod):
ID = ID + mech + str( int(mod_fact[i]*100) )
save_vector(tm, vm, ''.join(['./spiking_', str(run), '_', ID, '.out']) )
names = ['Target1p', 'cAMP', 'Gbgolf', 'D1RDAGolf']
data = {}
for i,substrate in enumerate([pka, camp, gbg, gprot]):
save_vector(tm, substrate, './substrate_'+names[i]+'.out' )
print(min(substrate), max(substrate))
data[names[i]] = [min(substrate), max(substrate)]
print(ID)
with open('substrates.json', 'w') as outfile:
json.dump(data, outfile)
#for key in I_d:
#save_vector(tm, I_d[key], ''.join(['./I_', key, '_', str(run), '.out']) )
spikes = getSpikedata_x_y(tm,vm)
RES[run] = {'factors': mod_fact + syn_fact, 'spikes': spikes}
# if run from terminal... ===============================================================
if __name__ == "__main__":
sys.argv = ['a', 'synMod']
print('starting sim')
factors = [1] #list(itertools.product(taum, tauh, taun))
if sys.argv[1] in ['vm', 'directMod']: #for current in currents:
mod_list = ['naf', 'kas', 'kaf', 'kir', 'cal12', 'cal13', 'can']
randMod = 1
currents = [0] #np.arange(300,360,10)
for n in range(200):
RES = {}
for i,current in enumerate(currents):
main( par="./params-rob.json", \
amp=current*1e-3, \
run=i, \
modulation=0, \
simDur=3000, \
stimDur=3000, \
sim=sys.argv[1], \
randMod=randMod, \
chan2mod=mod_list )
mod_fact = RES['factors']
ID = ''
for i,mech in enumerate(mod_list):
ID = ID + mech + str( int(mod_fact[i]*100) )
save_obj(RES, ''.join(['DRM-', ID]) )
elif sys.argv[1] in 'synMod': #for current in currents:
mod_list = ['naf', 'kas', 'kaf', 'kir', 'cal12', 'cal13', 'can']
syn_mod = ['amp', 'nmd', 'gab']
currents = [0] #np.arange(300,360,10)
RES = {}
testMode = False
if testMode:
n_runs = 4
else:
n_runs = 401
# ['Target1p', 'cAMP', 'Gbgolf', 'D1RDAGolf']
for n in range(n_runs):
main( par="./params-rob.json", \
amp=0.0, \
run=n, \
simDur=2000, \
stimDur=3000, \
sim=sys.argv[1], \
modulation=1, \
randMod=1, \
testMode=testMode, \
target='Target1p', \
chan2mod=mod_list )
if not testMode and n % 20 == 0:
print('in save loop')
mod_fact = RES[0]['factors']
ID = ''
for i,mech in enumerate(mod_list+syn_mod):
ID = ID + mech + str( int(mod_fact[i]*100) )
save_obj(RES, ''.join(['T1p_noGABAmod-', ID]) )