__author__ = 'Aaron D. Milstein'
from specify_cells import *
from plot_results import *
import scipy.optimize as optimize
import random
"""
This simulation uses scipy.optimize to iterate through AMPA_KIN mechanism parameters to fit target EPSP kinetics.
"""
morph_filename = 'EB2-late-bifurcation.swc'
mech_filename = '043016 Type A - km2_NMDA_KIN5_Pr'
def synaptic_kinetics_error(x, plot=0):
"""
:param x: list of parameters
:param plot: int or bool: method can be called manually to compare actual to target and fit waveforms
:return: float: Error
"""
for i, syn in enumerate(stim_syn_list):
syn.target(syn_type).kon = x[0]
syn.target(syn_type).koff = x[1]
syn.target(syn_type).CC = x[2]
syn.target(syn_type).CO = x[3]
syn.target(syn_type).Beta = x[4]
syn.target(syn_type).Alpha = x[5]
sim.run(v_init)
t = np.array(sim.tvec)
vm = np.array(sim.rec_list[0]['vec'])
interp_t = np.arange(0, duration, 0.001)
interp_vm = np.interp(interp_t, t, vm)
left, right = time2index(interp_t, equilibrate-3., equilibrate-1.)
baseline = np.average(interp_vm[left:right])
interp_vm -= baseline
start, end = time2index(interp_t, equilibrate, duration)
y = interp_vm[start:end]
interp_t = interp_t[start:end]
interp_t -= interp_t[0]
amp = np.max(y)
t_peak = np.where(y == amp)[0][0]
y /= amp
rise_10 = np.where(y[0:t_peak] >= 0.1)[0][0]
rise_90 = np.where(y[0:t_peak] >= 0.9)[0][0]
rise_tau = interp_t[rise_90] - interp_t[rise_10]
decay_90 = np.where(y[t_peak:] <= 0.9)[0][0]
decay_10 = np.where(y[t_peak:] <= 0.1)[0]
if decay_10.any():
decay_tau = interp_t[decay_10[0]] - interp_t[decay_90]
else:
decay_tau = 1000. # large error if trace has not decayed to 10% in 1 second
Ro = np.array(sim.rec_list[3]['vec'])
Rc_max = np.max(np.array(sim.rec_list[1]['vec'])+np.array(sim.rec_list[2]['vec'])+Ro)
"""
if 4. * decay_tau > duration - equilibrate:
steady_state = Ro[-1]
else:
t_steady = time2index(t, equilibrate, equilibrate + 4. * decay_tau)[1]
steady_state = Ro[t_steady]
if steady_state < target_val['steady_state']:
steady_state = target_val['steady_state'] # don't penalize decay to less than target
rise_tau = optimize.curve_fit(model_exp_rise, interp_t[1:t_peak], y[1:t_peak], p0=target_val['rise_tau'])[0]
decay_tau = optimize.curve_fit(model_exp_decay, interp_t[t_peak+1:]-interp_t[t_peak], y[t_peak+1:],
p0=target_val['decay_tau'])[0]
"""
result = {'rise_tau': rise_tau, 'decay_tau': decay_tau, 'Rc_max': Rc_max} # , 'steady_state': steady_state}
Err = 0.
for target in result:
Err += ((target_val[target] - result[target])/target_range[target])**2.
print('kon: %.3f, koff: %.3f, CC: %.3f, CO: %.3f, Beta: %.3f, Alpha: %.3f, Error: %.4E, Rise: %.3f, Decay: %.3f, '
'Rc_max: %.3f' % (x[0], x[1], x[2], x[3], x[4], x[5], Err, rise_tau, decay_tau, Rc_max))
if plot:
#fit_rise = model_exp_rise(interp_t[:t_peak], rise_tau)
#fit_decay = model_exp_decay(interp_t[:-t_peak], decay_tau)
#target_rise = model_exp_rise(interp_t[:t_peak], target_val['rise_tau'])
#target_decay = model_exp_decay(interp_t[:-t_peak], target_val['decay_tau'])
plt.plot(interp_t, y, label="actual", color='b')
#plt.plot(interp_t[:t_peak], fit_rise, label="fit", color='r')
#plt.plot(interp_t[:-t_peak]+interp_t[t_peak], fit_decay, color='r')
#plt.plot(interp_t[:t_peak], target_rise, label="target", color='g')
#plt.plot(interp_t[:-t_peak]+interp_t[t_peak], target_decay, color='g')
plt.legend(loc='best')
plt.show()
plt.close()
else:
return Err
equilibrate = 250. # time to steady-state
duration = 1250.
v_init = -67.
num_syns = 1
spike_times = h.Vector([equilibrate])
cell = CA1_Pyr(morph_filename, mech_filename, full_spines=True)
syn_type = 'AMPA_KIN'
sim = QuickSim(duration)
# look for a trunk bifurcation
trunk_bifurcation = [trunk for trunk in cell.trunk if len(trunk.children) > 1 and trunk.children[0].type == 'trunk' and
trunk.children[1].type == 'trunk']
# get where the thickest trunk branch gives rise to the tuft
if trunk_bifurcation: # follow the thicker trunk
trunk = max(trunk_bifurcation[0].children[:2], key=lambda node: node.sec(0.).diam)
trunk = (node for node in cell.trunk if cell.node_in_subtree(trunk, node) and 'tuft' in (child.type for child in
node.children)).next()
else:
trunk = (node for node in cell.trunk if 'tuft' in (child.type for child in node.children)).next()
tuft = (child for child in trunk.children if child.type == 'tuft').next()
trunk = trunk_bifurcation[0]
#sim.append_rec(cell, trunk, loc=1., description='trunk vm')
spine_list = []
spine_list.extend(trunk.spines)
for spine in spine_list:
syn = Synapse(cell, spine, [syn_type], stochastic=0)
random.seed(0)
stim_syn_list = [spine_list[i].synapses[0] for i in random.sample(range(len(spine_list)), num_syns)]
for i, syn in enumerate(stim_syn_list):
syn.source.play(spike_times)
sim.append_rec(cell, syn.node, object=syn.target(syn_type), param='_ref_g')
sim.append_rec(cell, syn.node, object=syn.target(syn_type), param='_ref_Rb')
sim.append_rec(cell, syn.node, object=syn.target(syn_type), param='_ref_Rc')
sim.append_rec(cell, syn.node, object=syn.target(syn_type), param='_ref_Ro')
#the target values and acceptable ranges
target_val = {'rise_tau': .1, 'decay_tau': 7., 'Rc_max': 0.9} # extrapolating from Chen...Murphy and Harnett...Magee
target_range = {'rise_tau': 0.01, 'decay_tau': 0.1, 'Rc_max': 0.01}
#the initial guess and bounds
#x = [kon, koff, CC, CO, Beta, Alpha)
#x0 = [10., .05, 25., 31., 2.5, 1.5]
#x0 = [60., 10., 60., 5., 100., 60.]
x0 = [139.87, 4.05, 54.54, 10.85, 102.37, 111.66] # following basinhopping and stalled simplex
xmin = [10., .1, 1., 1., 1., 1.]
xmax = [500., 10., 100., 50., 200., 200.]
#x1 = [139.87, 4.05, 54.54, 10.85, 102.37, 111.66] # following basinhopping and stalled simplex
x1 = [12.88, 6.47, 69.97, 6.16, 100.63, 173.04]
# rewrite the bounds in the way required by optimize.minimize
xbounds = [(low, high) for low, high in zip(xmin, xmax)]
blocksize = 0.5 # defines the fraction of the xrange that will be explored at each step
# basinhopping starts with this value and reduces it by 10% every 'interval' iterations
mytakestep = MyTakeStep(blocksize, xmin, xmax)
minimizer_kwargs = dict(method=null_minimizer)
"""
result = optimize.basinhopping(synaptic_kinetics_error, x0, niter= 720, niter_success=100, disp=True, interval=20,
minimizer_kwargs=minimizer_kwargs, take_step=mytakestep)
#synaptic_kinetics_error(result.x, plot=1)
"""
polished_result = optimize.minimize(synaptic_kinetics_error, x1, method='Nelder-Mead', options={'ftol': 1e-3,
'xtol': 1e-3, 'disp': True})
synaptic_kinetics_error(polished_result.x, plot=1)
#synaptic_kinetics_error(x1, plot=1)