def calculate_features(self, t_ms, v, i, feature_names):
delay = self.stim.delay * 1e-3
duration = self.stim.dur * 1e-3
t = t_ms * 1e-3
swp = EphysSweepFeatureExtractor(t,
v,
i,
start=delay,
end=delay + duration)
swp.process_spikes()
swp.sweep_feature("v_baseline") # Pre-compute the baseline
swp.process_new_spike_feature("slow_trough_norm_t",
sf.slow_trough_norm_t,
affected_by_clipping=True)
swp.process_new_spike_feature("slow_trough_delta_v",
sf.slow_trough_delta_voltage_feature,
affected_by_clipping=True)
sweep_keys = swp.sweep_feature_keys()
spike_keys = swp.spike_feature_keys()
out_features = []
for k in feature_names:
if k in sweep_keys:
out_features.append(swp.sweep_feature(k))
elif k in spike_keys:
out_features.append(swp.spike_feature(k).mean())
else:
out_features.append(np.nan)
return np.array(out_features)
def expt_ap_dim(specimen_id):
data_set = expt_data_set(specimen_id)
long_square_sweeps = lims_utils.get_sweeps_of_type("C1LSCOARSE",
specimen_id,
passed_only=True)
fi_curve_data = dict(
[amp_and_spike_count(data_set, sweep) for sweep in long_square_sweeps])
sweeps_by_amp = {
amp_and_spike_count(data_set, sweep)[0]: sweep
for sweep in long_square_sweeps
}
fi_arr = np.array([(amp, fi_curve_data[amp])
for amp in sorted(fi_curve_data.keys())])
spiking_sweeps = np.flatnonzero(fi_arr[:, 1])
if len(spiking_sweeps) == 0:
return np.nan, np.nan
rheo_sweep = sweeps_by_amp[fi_arr[spiking_sweeps[0], 0]]
# print specimen_id, rheo_sweep
v, i, t = lims_utils.get_sweep_v_i_t_from_set(data_set, rheo_sweep)
swp_ext = EphysSweepFeatureExtractor(t, v, start=1.02, end=2.02)
swp_ext.process_spikes()
return (swp_ext.spike_feature("width")[0] * 1e3,
swp_ext.spike_feature("peak_v")[0] -
swp_ext.spike_feature("trough_v")[0])
def from_electrophysiology(
cell_id: int,
ephys: NwbDataSet,
duration=2.0) -> 'ProcessedAllenNeuronElectrophysiology':
current_list = []
voltage_list = []
time_list = []
stim_amp_list = []
n_spikes_list = []
spike_features_list = []
for sweep_number in ephys.get_sweep_numbers():
sweep_metadata = ephys.get_sweep_metadata(sweep_number)
if sweep_metadata['aibs_stimulus_name'] == 'Long Square':
sweep_data = ephys.get_sweep(sweep_number)
amp = sweep_metadata['aibs_stimulus_amplitude_pa']
index_range = sweep_data["index_range"]
sampling_rate = sweep_data["sampling_rate"]
current = sweep_data["stimulus"][
index_range[0]:index_range[1] + 1]
voltage = sweep_data["response"][
index_range[0]:index_range[1] + 1]
# truncate
max_frames = int(duration * sampling_rate)
assert max_frames < len(voltage)
current = current[:max_frames] * 1e12 # in pA
voltage = voltage[:max_frames] * 1e3 # in mV
# extract featrures
time = np.arange(0, max_frames,
dtype=np.float) / sampling_rate # in seconds
ext = EphysSweepFeatureExtractor(t=time, v=voltage, i=current)
ext.process_spikes()
spike_features = ext.spikes()
n_spikes = len(spike_features)
current_list.append(current)
voltage_list.append(voltage)
time_list.append(time)
stim_amp_list.append(amp)
n_spikes_list.append(n_spikes)
spike_features_list.append(spike_features)
return ProcessedAllenNeuronElectrophysiology(
cell_id=cell_id,
current_list=current_list,
voltage_list=voltage_list,
time_list=time_list,
stim_amp_list=stim_amp_list,
n_spikes_list=n_spikes_list,
spike_features_list=spike_features_list)
def find_core1_trace(data_set, all_sweeps):
sweep_type = "C1LSCOARSE"
_, sweeps, statuses = ephys_utils.get_sweeps_of_type(sweep_type, all_sweeps)
sweep_status = dict(zip(sweeps, statuses))
sweep_info = {}
for s in sweeps:
if sweep_status[s][-6:] == "failed":
continue
v, i, t = ephys_utils.get_sweep_v_i_t_from_set(data_set, s)
if np.all(v[-100:] == 0): # Check for early termination of sweep
continue
stim_start, stim_dur, stim_amp, start_idx, end_idx = ephys_utils.get_step_stim_characteristics(i, t)
swp = EphysSweepFeatureExtractor(t, v, i, start=stim_start, end=(stim_start + stim_dur))
swp.process_spikes()
isi_cv = swp.sweep_feature("isi_cv", allow_missing=True)
sweep_info[s] = {"amp": stim_amp,
"n_spikes": len(swp.spikes()),
"quality": is_trace_good_quality(v, i, t),
"isi_cv": isi_cv}
rheobase_amp = 1e12
for s in sweep_info:
if sweep_info[s]["amp"] < rheobase_amp and sweep_info[s]["n_spikes"] > 0:
rheobase_amp = sweep_info[s]["amp"]
sweep_to_use_amp = 1e12
sweep_to_use_isi_cv = 1e121
sweep_to_use = -1
for s in sweep_info:
if sweep_info[s]["quality"] and sweep_info[s]["amp"] >= 39.0 + rheobase_amp and sweep_info[s]["isi_cv"] < 1.2 * sweep_to_use_isi_cv:
use_new_sweep = False
if sweep_to_use_isi_cv > 0.3 and ((sweep_to_use_isi_cv - sweep_info[s]["isi_cv"]) / sweep_to_use_isi_cv) >= 0.2:
use_new_sweep = True
elif sweep_info[s]["amp"] < sweep_to_use_amp:
use_new_sweep = True
if use_new_sweep:
_fit_stage_1_log.info("now using sweep" + str(s))
sweep_to_use = s
sweep_to_use_amp = sweep_info[s]["amp"]
sweep_to_use_isi_cv = sweep_info[s]["isi_cv"]
if sweep_to_use == -1:
_fit_stage_1_log.warn("Could not find appropriate core 1 sweep!")
return []
else:
return [sweep_to_use]
def load_sweep(ds, sweep_num):
sweep = ds.get_sweep(sweep_num)
v = sweep['response'] * 1e3 # to mV
i = sweep['stimulus'] * 1e12 # to pA
t = np.arange(0, len(v)) * (1.0 / sweep['sampling_rate'])
sweep_ext = EphysSweepFeatureExtractor(t=t, v=v, i=i)
sweep_ext.process_spikes()
spike_idxs = sweep_ext.spike_feature("threshold_index")
peak_idxs = sweep_ext.spike_feature("peak_index")
trough_idxs = sweep_ext.spike_feature("trough_index")
hz_in = sweep['sampling_rate']
hz_out = 25000.
return resample_timeseries(v, i, t,
spike_idxs, peak_idxs, trough_idxs,
hz_in, hz_out)
def ve_ap_dim(specimen_id, ve_path):
data_set = NwbDataSet(ve_path)
expt_set = expt_data_set(specimen_id)
long_square_sweeps = lims_utils.get_sweeps_of_type("C1LSCOARSE",
specimen_id,
passed_only=True)
fi_curve_data = dict([
amp_and_spike_count(data_set, sweep, expt_set)
for sweep in long_square_sweeps
])
sweeps_by_amp = {
amp_and_spike_count(data_set, sweep, expt_set)[0]: sweep
for sweep in long_square_sweeps
}
fi_arr = np.array([(amp, fi_curve_data[amp])
for amp in sorted(fi_curve_data.keys())])
spiking_sweeps = np.flatnonzero(fi_arr[:, 1])
if len(spiking_sweeps) == 0:
return np.nan, np.nan
rheo_sweep = sweeps_by_amp[fi_arr[spiking_sweeps[0], 0]]
# print specimen_id, rheo_sweep
v, i, t = lims_utils.get_sweep_v_i_t_from_set(data_set, rheo_sweep)
swp_ext = EphysSweepFeatureExtractor(t,
v,
start=1.02,
end=2.02,
filter=None)
swp_ext.process_spikes()
if len(swp_ext.spike_feature("width")) == 0:
print "NO SPIKES FOR {:d} ON SWEEP {:d}".format(
specimen_id, sweeps_by_amp[fi_arr[spiking_sweeps[0], 0]])
print fi_arr
print sweeps_by_amp
return np.nan, np.nan
return_vals = (swp_ext.spike_feature("width")[0] * 1e3,
swp_ext.spike_feature("peak_v")[0] -
swp_ext.spike_feature("trough_v")[0])
return return_vals
def is_trace_good_quality(v, i, t):
stim_start, stim_dur, stim_amp, start_idx, end_idx = ephys_utils.get_step_stim_characteristics(i, t)
swp = EphysSweepFeatureExtractor(t, v, i, start=stim_start, end=(stim_start + stim_dur))
swp.process_spikes()
spikes = swp.spikes()
rate = swp.sweep_feature("avg_rate")
if rate < 5.0:
return False
time_to_end = stim_start + stim_dur - spikes[-1]["threshold_t"]
avg_end_isi = (((spikes[-1]["threshold_t"] - spikes[-2]["threshold_t"]) +
(spikes[-2]["threshold_t"] - spikes[-3]["threshold_t"])) / 2.0)
if time_to_end > 2 * avg_end_isi:
return False
isis = np.diff([spk["threshold_t"] for spk in spikes])
if check_for_pause(isis):
return False
return True
v = v.magnitude
#print(v)
tf = trace.t_stop
#print(tf)
ti = trace.t_start
#print(ti)
t = np.linspace(0,float(tf - ti), len(v))
#print(t)
#i = np.zeros(len(v))
#print(np.isnan(v)[0])
#print('##########')
print(trace.sampling_rate)
plt.plot(t,v,label = (iteration_number))
try:
Trace_with_features = EphysSweepFeatureExtractor(t=t, v=v, filter = float(trace.sampling_rate)/2500,min_peak=-30.0, dv_cutoff=20.0, max_interval=0.005, min_height=2.0, thresh_frac=0.05, baseline_interval=0.1, baseline_detect_thresh=0.3, id=None)
Trace_with_features.process_spikes()
print(Trace_with_features.filter)
plt.plot(Trace_with_features.spike_feature("peak_t"),Trace_with_features.spike_feature("peak_v"),'k+')
except:
pass
plt.title('recording type: '+str(trace.name).split("-")[0]+' '+str(len(analog_signals))+' '+'traces'+' '+'_compiled')
plt.ylabel('Amplitude of signal: '+str(trace[0]).split()[1])
plt.xlabel('time (mS)')
plt.legend()
plt.show()
file_to_read = nio.StimfitIO('/mnt/5D4B-FA71/Data/190822/trace_alone.dat')
segments = file_to_read.read_block().segments
analog_signal = segment.analogsignals
import os
from allensdk.ephys.ephys_extractor import EphysSweepFeatureExtractor
from allensdk.core.cell_types_cache import CellTypesCache
filename = 'Cell 6 of 171117.abf'
seg_no = 17 #0 is -100pA, 4 is 0pA, 20 is 400pA. Now extrapolate
stim_start = 81.4e-3 #in s
stim_stop = 581.4e-3 #in s
Actualstim_start = seg_no * 2 + stim_start
Actualstim_stop = seg_no * 2 + stim_stop
Inputcurr = seg_no * 25 - 100 #in pA
reader = AxonIO(filename='Deepanjali data/WT step input cells/' + filename)
Vtrace = reader.read_block().segments[seg_no].analogsignals[0]
i = np.zeros(int((Vtrace.t_stop - Vtrace.t_start) * Vtrace.sampling_rate))
i[int(stim_start * Vtrace.sampling_rate):int(stim_stop *
Vtrace.sampling_rate)] = Inputcurr
i = np.array(i)
v = np.array([float(V) for V in Vtrace])
t = np.linspace(0, float(Vtrace.t_stop - Vtrace.t_start),
int((Vtrace.t_stop - Vtrace.t_start) * Vtrace.sampling_rate))
t = np.array(t)
sweep_ext = EphysSweepFeatureExtractor(t=t, v=v, i=i, filter=5)
sweep_ext.process_spikes()
sweep_ext.spike_feature_keys() #Lists all the features that can be extracted
sweep_ext.spike_feature("width") #Extracts AP width
def calculate_feature_errors(self, t_ms, v, i, feature_names, targets):
# Special case checks and penalty values
minimum_num_spikes = 2
missing_penalty_value = 20.0
max_fail_penalty = 250.0
min_fail_penalty = 75.0
overkill_reduction = 0.75
variance_factor = 0.1
fail_trace = False
delay = self.stim.delay * 1e-3
duration = self.stim.dur * 1e-3
t = t_ms * 1e-3
# penalize for failing to return to rest
start_index = np.flatnonzero(t >= delay)[0]
if np.abs(v[-1] - v[:start_index].mean()) > 2.0:
fail_trace = True
else:
pre_swp = EphysSweepFeatureExtractor(t, v, i, start=0, end=delay)
pre_swp.process_spikes()
if pre_swp.sweep_feature("avg_rate") > 0:
fail_trace = True
target_features_dict = {
f["name"]: {
"mean": f["mean"],
"stdev": f["stdev"]
}
for f in targets
}
if not fail_trace:
swp = EphysSweepFeatureExtractor(t,
v,
i,
start=delay,
end=delay + duration)
swp.process_spikes()
swp.sweep_feature("v_baseline") # Pre-compute the baseline
swp.process_new_spike_feature("slow_trough_norm_t",
sf.slow_trough_norm_t,
affected_by_clipping=True)
swp.process_new_spike_feature("slow_trough_delta_v",
sf.slow_trough_delta_voltage_feature,
affected_by_clipping=True)
sweep_keys = swp.sweep_feature_keys()
spike_keys = swp.spike_feature_keys()
if len(swp.spike_feature(
"threshold_t")) < minimum_num_spikes: # Enough spikes?
fail_trace = True
else:
avg_per_spike_peak_error = np.mean([
abs(peak_v - target_features_dict["peak_v"]["mean"])
for peak_v in swp.spike_feature("peak_v")
])
avg_overall_error = abs(
target_features_dict["peak_v"]["mean"] -
swp.spike_feature("peak_v").mean())
if avg_per_spike_peak_error > 3. * avg_overall_error: # Weird bi-modality of spikes; 3.0 is arbitrary
fail_trace = True
if fail_trace:
variance_start = np.flatnonzero(t >= delay - 0.1)[0]
variance_end = np.flatnonzero(t >= (delay + duration) / 2.0)[0]
trace_variance = v[variance_start:variance_end].var()
error_value = max(
max_fail_penalty - trace_variance * variance_factor,
min_fail_penalty)
errs = np.ones(len(feature_names)) * error_value
else:
errs = []
for k in feature_names:
if k in sweep_keys:
model_mean = swp.sweep_feature(k)
elif k in spike_keys:
model_mean = swp.spike_feature(k).mean()
else:
_log.debug("Could not find feature %s", k)
errs.append(missing_penalty_value)
continue
if np.isnan(model_mean):
errs.append(missing_penalty_value)
else:
target_mean = target_features_dict[k]['mean']
target_stdev = target_features_dict[k]['stdev']
errs.append(
np.abs((model_mean - target_mean) / target_stdev))
errs = np.array(errs)
return errs
def singlefeature(seg_nol,reader,filename,stim_start,stim_end):
'''This function is used to extract features from a neo.io.axonio.AxonIO object and extracts features from it.
seg_nol takes in a list of segment numbers. reader is the objecct. filename is the name of the neuron. stim_start and stim_end are when the current injection was started and ended.
Outputs a pandas dataframe with all the features'''
features = {}
for seg_no in seg_nol:
Vtrace = reader.read_block().segments[seg_no].analogsignals[0]
#Extra
features['Cell name'] = filename
features['Sampling rate'] = Vtrace.sampling_rate
features['stim_start'] = stim_start
features['stim_end'] = stim_end
Inputcurr = seg_no*25e-12 - 100e-12 #in A
i = np.zeros(int((Vtrace.t_stop - Vtrace.t_start)*Vtrace.sampling_rate))
i[int(stim_start*Vtrace.sampling_rate):int(stim_end*Vtrace.sampling_rate)] = Inputcurr*1e12
i = np.array(i) #in pA
v = np.array([float(V) for V in Vtrace]) #in mV
t = np.linspace(0,float(Vtrace.t_stop - Vtrace.t_start), int((Vtrace.t_stop - Vtrace.t_start)*Vtrace.sampling_rate))
t = np.array(t) # in s
# plt.plot(t,v, label = f'{filename}')
# plt.legend()
# plt.show()
sweep_ext = EphysSweepFeatureExtractor(t=t, v=v, i=i, filter=float(Vtrace.sampling_rate)/2500)
sweep_ext.process_spikes()
# E_rest
features[f'E_rest_{Inputcurr*1e12}'] = np.nanmean(v[t<=stim_start])*1e-3
# Input resistance #steady-state V is average of last 100ms of the current clamp duration
Vtracep25 = reader.read_block().segments[5].analogsignals[0]
Vtracep25 = np.array([float(V) for V in Vtracep25]) #in mV
Vtracen25 = reader.read_block().segments[3].analogsignals[0]
Vtracen25 = np.array([float(V) for V in Vtracen25]) #in mV
str_ind = (np.abs(t-stim_end+100e-3)).argmin()
stp_ind = (np.abs(t-stim_end)).argmin()
features[f'Rinput'] = (-1*np.nanmean(Vtracen25[str_ind:stp_ind]) + np.nanmean(Vtracep25[str_ind:stp_ind]))*1e-3/2/25e-12
# Total capacitance
Vtracep25_choppped = Vtracep25[:stp_ind]
Vtracen25_choppped = Vtracen25[:stp_ind]
vp63 = (np.nanmean(Vtracep25[str_ind:stp_ind]) - features[f'E_rest_{Inputcurr*1e12}'])*0.63 + features[f'E_rest_{Inputcurr*1e12}']
vn63 = (np.nanmean(Vtracen25[str_ind:stp_ind]) - features[f'E_rest_{Inputcurr*1e12}'])*0.63 + features[f'E_rest_{Inputcurr*1e12}']
tau = (t[(np.abs(Vtracep25_choppped-vp63)).argmin()] - stim_start + t[(np.abs(Vtracep25_choppped-vp63)).argmin()] - stim_start)/2
tauinv = (t[len(Vtracep25_choppped)-(np.abs(Vtracep25_choppped[stp_ind::-1]-vp63)).argmin()] - stim_start + t[len(Vtracep25_choppped)-(np.abs(Vtracep25_choppped[::-1]-vp63)).argmin()] - stim_start)/2
features[f'Cm'] = (tau+tauinv)/features[f'Rinput']/2
# AP1_amp
features[f'AP1_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("peak_v")[0]*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#APp_amp
features[f'APp_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("peak_v")[-2]*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#AP1_width
features[f'AP1_width_{Inputcurr*1e12}'] = sweep_ext.spike_feature("width")[0]
#APp_width
features[f'APp_width_{Inputcurr*1e12}'] = sweep_ext.spike_feature("width")[-2]
#AP1_thresh
features[f'AP1_thresh_{Inputcurr*1e12}'] = sweep_ext.spike_feature("threshold_v")[0]*1e-3
#APp_thresh
features[f'APp_thresh_{Inputcurr*1e12}'] = sweep_ext.spike_feature("threshold_v")[-2]*1e-3
#AP1_lat
features[f'AP1_lat_{Inputcurr*1e12}'] = sweep_ext.spike_feature("threshold_t")[0] - stim_start
#ISI1
features[f'ISI1_{Inputcurr*1e12}'] = sweep_ext.spike_feature("peak_t")[1] - sweep_ext.spike_feature("peak_t")[0]
#ISIl
features[f'ISIl_{Inputcurr*1e12}'] = sweep_ext.spike_feature("peak_t")[-1] - sweep_ext.spike_feature("peak_t")[-2]
#ISIavg
pt = sweep_ext.spike_feature("peak_t")
features[f'ISIavg_{Inputcurr*1e12}'] = np.nanmean([s-f for s,f in zip(pt[1:],pt[:-1])])
#freq
features[f'freq_{Inputcurr*1e12}'] = len(sweep_ext.spike_feature("peak_t"))/(stim_end - stim_start)
#Adptn_id = 1-ISI1/ISIl
features[f'Adptn_id_{Inputcurr*1e12}'] = 1 - features[f'ISI1_{Inputcurr*1e12}']/features[f'ISIl_{Inputcurr*1e12}']
#fAHP_AP1_amp
features[f'fAHP_AP1_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("fast_trough_v")[0]*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#fAHP_APp_amp
features[f'fAHP_APp_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("fast_trough_v")[-2]*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#mAHP_AP1_amp
features[f'mAHP_AP1_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("slow_trough_v")[0]*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#mAHP_APp_amp
features[f'mAHP_APp_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("slow_trough_v")[-2]*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#mAHP_AP1_dur
features[f'mAHP_AP1_dur_{Inputcurr*1e12}'] = (sweep_ext.spike_feature("slow_trough_t")[0] - sweep_ext.spike_feature("peak_t")[0])/features[f'ISI1_{Inputcurr*1e12}']
#mAHP_APp_dur = mAHP of second last spike (penultimate)
features[f'mAHP_APp_dur_{Inputcurr*1e12}'] = (sweep_ext.spike_feature("slow_trough_t")[-2] - sweep_ext.spike_feature("peak_t")[-2])/features[f'ISIl_{Inputcurr*1e12}']
#ADP_AP1_amp
features[f'ADP_AP1_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("adp_v")[0]*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#ADP_APp_amp
features[f'ADP_APp_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("adp_v")[-2]*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#mAHP_stimend_amp = within 50ms
features[f'mAHP_stimend_amp_{Inputcurr*1e12}'] = np.min(v[int((stim_end)*Vtrace.sampling_rate):int((stim_end + 50e-3)*Vtrace.sampling_rate)])*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
#sAHP_stimend_amp = within 200ms
features[f'sAHP_stimend_amp_{Inputcurr*1e12}'] = np.min(v[int((stim_end)*Vtrace.sampling_rate):int((stim_end + 200e-3)*Vtrace.sampling_rate)])*1e-3 - features[f'E_rest_{Inputcurr*1e12}']
return features
def features(Vtrace,
stim_start=preStimTime,
stim_end=preStimTime + injectTime,
truntime=runtime,
Inputcurr=Injectcurr):
features_df = feature_range_df.copy()
v = np.array(Vtrace) #in mV
t = np.linspace(0, truntime, len(Vtrace)) #in s
start_idx = (np.abs(t - stim_start)).argmin()
end_idx = (np.abs(t - stim_end)).argmin()
i = np.zeros(len(t))
i[start_idx:end_idx] = Inputcurr * 1e12 #in pA
try:
sweep_ext = EphysSweepFeatureExtractor(t=t,
v=v,
i=i,
start=stim_start,
end=stim_end,
filter=9.9)
sweep_ext.process_spikes()
except:
sweep_ext = EphysSweepFeatureExtractor(t=t,
v=v,
i=i,
start=stim_start,
end=stim_end,
filter=4.9)
sweep_ext.process_spikes()
features_df['raw'] = ''
# E_rest
if stim_start > 0.5:
features_df.loc[f'E_rest_{Inputcurr*1e12}', 'raw'] = np.nanmean(
v[int(0.5 * len(v) / truntime):int(stim_start * len(v) /
truntime)])
else:
features_df.loc[f'E_rest_{Inputcurr*1e12}',
'raw'] = np.nanmean(v[-int(0.5 * len(v) / truntime):])
# AP1_amp
features_df.loc[f'AP1_amp_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("peak_v")[
0] - features_df.loc[f'E_rest_{Inputcurr*1e12}', 'raw']
#APp_amp
features_df.loc[
f'APp_amp_{Inputcurr*1e12}', 'raw'] = sweep_ext.spike_feature(
"peak_v")[-2] - features_df.loc[f'E_rest_{Inputcurr*1e12}', 'raw']
#AP1_width
features_df.loc[f'AP1_width_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("width")[0]
#APp_width
features_df.loc[f'APp_width_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("width")[-2]
#AP1_thresh
features_df.loc[f'AP1_thresh_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("threshold_v")[0]
#APp_thresh
features_df.loc[f'APp_thresh_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("threshold_v")[-2]
#AP1_lat
features_df.loc[
f'AP1_lat_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("threshold_t")[0] - stim_start
#ISI1
features_df.loc[f'ISI1_{Inputcurr*1e12}', 'raw'] = sweep_ext.spike_feature(
"peak_t")[1] - sweep_ext.spike_feature("peak_t")[0]
#ISIl
features_df.loc[f'ISIl_{Inputcurr*1e12}', 'raw'] = sweep_ext.spike_feature(
"peak_t")[-1] - sweep_ext.spike_feature("peak_t")[-2]
# #ISIavg
pt = sweep_ext.spike_feature("peak_t")
features_df.loc[f'ISIavg_{Inputcurr*1e12}', 'raw'] = np.nanmean(
[s - f for s, f in zip(pt[1:], pt[:-1])])
# features_df.loc[f'ISIavg_{Inputcurr*1e12}','raw'] = 'skip'
#freq
features_df.loc[f'freq_{Inputcurr*1e12}', 'raw'] = len(
sweep_ext.spike_feature("peak_t")) / (stim_end - stim_start)
#Adptn_id = 1-ISI1/ISIl
features_df.loc[f'Adptn_id_{Inputcurr*1e12}', 'raw'] = 1 - features_df.loc[
f'ISI1_{Inputcurr*1e12}',
'raw'] / features_df.loc[f'ISIl_{Inputcurr*1e12}', 'raw']
#fAHP_AP1_amp
features_df.loc[f'fAHP_AP1_amp_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("fast_trough_v")[
0] - features_df.loc[f'E_rest_{Inputcurr*1e12}', 'raw']
#fAHP_APp_amp
features_df.loc[f'fAHP_APp_amp_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("fast_trough_v")[
-2] - features_df.loc[f'E_rest_{Inputcurr*1e12}',
'raw']
# #mAHP_AP1_amp
# features_df.loc[f'mAHP_AP1_amp_{Inputcurr*1e12}','raw'] = sweep_ext.spike_feature("slow_trough_v")[0] - features_df.loc[f'E_rest_{Inputcurr*1e12}','raw']
features_df.loc[f'mAHP_AP1_amp_{Inputcurr*1e12}', 'raw'] = 'skip'
#mAHP_APp_amp
features_df.loc[f'mAHP_APp_amp_{Inputcurr*1e12}',
'raw'] = sweep_ext.spike_feature("slow_trough_v")[
-2] - features_df.loc[f'E_rest_{Inputcurr*1e12}',
'raw']
# #mAHP_AP1_dur
# features_df.loc[f'mAHP_AP1_dur_{Inputcurr*1e12}','raw'] = (sweep_ext.spike_feature("slow_trough_t")[0] - sweep_ext.spike_feature("peak_t")[0])/features_df.loc[f'ISI1_{Inputcurr*1e12}','raw']
features_df.loc[f'mAHP_AP1_dur_{Inputcurr*1e12}', 'raw'] = 'skip'
#mAHP_APp_dur = mAHP of second last spike (penultimate)
features_df.loc[f'mAHP_APp_dur_{Inputcurr*1e12}',
'raw'] = (sweep_ext.spike_feature("slow_trough_t")[-2] -
sweep_ext.spike_feature("peak_t")[-2]
) / features_df.loc[f'ISIl_{Inputcurr*1e12}',
'raw']
# #ADP_AP1_amp
# features_df.loc[f'ADP_AP1_amp_{Inputcurr*1e12}','raw'] = sweep_ext.spike_feature("adp_v")[0] - features_df.loc[f'E_rest_{Inputcurr*1e12}','raw']
features_df.loc[f'ADP_AP1_amp_{Inputcurr*1e12}', 'raw'] = 'skip'
# #ADP_APp_amp
# features_df.loc[f'ADP_APp_amp_{Inputcurr*1e12}','raw'] = sweep_ext.spike_feature("adp_v")[-2] - features_df.loc[f'E_rest_{Inputcurr*1e12}','raw']
features_df.loc[f'ADP_APp_amp_{Inputcurr*1e12}', 'raw'] = 'skip'
#mAHP_stimend_amp = within 50ms
end50_idx = (np.abs(t - stim_end - 50e-3)).argmin()
features_df.loc[f'mAHP_stimend_amp_{Inputcurr*1e12}', 'raw'] = np.min(
v[end_idx:end50_idx]) - features_df.loc[f'E_rest_{Inputcurr*1e12}',
'raw']
#sAHP_stimend_amp = within 200ms
end200_idx = (np.abs(t - stim_end - 200e-3)).argmin()
features_df.loc[f'sAHP_stimend_amp_{Inputcurr*1e12}', 'raw'] = np.min(
v[end_idx:end200_idx]) - features_df.loc[f'E_rest_{Inputcurr*1e12}',
'raw']
features_df = features_df.replace('', np.nan)
features_df['rescaled'] = ''
for i in features_df.index:
if features_df.loc[i, 'raw'] == 'skip':
features_df.loc[i, 'rescaled'] = 0
elif np.isnan(features_df.loc[i, 'raw']):
features_df.loc[
i,
'rescaled'] = 10 #Penalty for not having the feature altogether
else:
Max = features_df.loc[i, 'Max']
Min = features_df.loc[i, 'Min']
features_df.loc[i, 'rescaled'] = 2 / (Max - Min) * (
features_df.loc[i, 'raw'] - Min) - 1
features_df = features_df.replace('', np.nan)
features_df['cost'] = ''
for i in features_df.index:
if features_df.loc[i, 'rescaled'] != '' and ~np.isnan(
features_df.loc[i, 'rescaled']):
if features_df.loc[i, 'rescaled'] > 1 or features_df.loc[
i, 'rescaled'] < -1:
features_df.loc[i, 'cost'] = np.abs(
features_df.loc[i, 'rescaled']) * features_df.loc[i,
'weight']
else:
features_df.loc[i, 'cost'] = 0
features_df = features_df.replace('', np.nan)
return features_df
def get_features(filename, nameit):
#filename is the hoc file of the simulation with
#vectors already present as recv and rect
neuron.xopen(filename)
features_pd = pd.DataFrame()
stim_start = 1000e-3
stim_end = 1500e-3
features = {}
#Extra
features['Cell name'] = nameit
features['stim_start'] = stim_start
features['stim_end'] = stim_end
Inputcurr = 150e-12 #in A
v = np.array(h.recv) #in mV
t = np.array(h.rect)*1e-3 #in s
start_idx = (np.abs(t - stim_start)).argmin()
end_idx = (np.abs(t - stim_end)).argmin()
i = np.zeros(len(t))
i[start_idx:end_idx] = Inputcurr*1e12 #in pA
sweep_ext = EphysSweepFeatureExtractor(t=t, v=v, i=i)
sweep_ext.process_spikes()
# E_rest
features[f'E_rest_{Inputcurr*1e12}'] = np.nanmean(v[i==0])
# AP1_amp
features[f'AP1_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("peak_v")[0] - features[f'E_rest_{Inputcurr*1e12}']
#APp_amp
features[f'APp_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("peak_v")[-2] - features[f'E_rest_{Inputcurr*1e12}']
#AP1_width
features[f'AP1_width_{Inputcurr*1e12}'] = sweep_ext.spike_feature("width")[0]
#APp_width
features[f'APp_width_{Inputcurr*1e12}'] = sweep_ext.spike_feature("width")[-2]
#AP1_thresh
features[f'AP1_thresh_{Inputcurr*1e12}'] = sweep_ext.spike_feature("threshold_v")[0]
#APp_thresh
features[f'APp_thresh_{Inputcurr*1e12}'] = sweep_ext.spike_feature("threshold_v")[-2]
#AP1_lat
features[f'AP1_lat_{Inputcurr*1e12}'] = sweep_ext.spike_feature("threshold_t")[0] - stim_start
#ISI1
features[f'ISI1_{Inputcurr*1e12}'] = sweep_ext.spike_feature("peak_t")[1] - sweep_ext.spike_feature("peak_t")[0]
#ISIl
features[f'ISIl_{Inputcurr*1e12}'] = sweep_ext.spike_feature("peak_t")[-1] - sweep_ext.spike_feature("peak_t")[-2]
#ISIavg
pt = sweep_ext.spike_feature("peak_t")
features[f'ISIavg_{Inputcurr*1e12}'] = np.nanmean([s-f for s,f in zip(pt[1:],pt[:-1])])
#freq
features[f'freq_{Inputcurr*1e12}'] = len(sweep_ext.spike_feature("peak_t"))/(stim_end - stim_start)
#Adptn_id = 1-ISI1/ISIl
features[f'Adptn_id_{Inputcurr*1e12}'] = 1 - features[f'ISI1_{Inputcurr*1e12}']/features[f'ISIl_{Inputcurr*1e12}']
#fAHP_AP1_amp
features[f'fAHP_AP1_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("fast_trough_v")[0] - features[f'E_rest_{Inputcurr*1e12}']
#fAHP_APp_amp
features[f'fAHP_APp_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("fast_trough_v")[-2] - features[f'E_rest_{Inputcurr*1e12}']
#mAHP_AP1_amp
features[f'mAHP_AP1_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("slow_trough_v")[0] - features[f'E_rest_{Inputcurr*1e12}']
#mAHP_APp_amp
features[f'mAHP_APp_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("slow_trough_v")[-2] - features[f'E_rest_{Inputcurr*1e12}']
#mAHP_AP1_dur
features[f'mAHP_AP1_dur_{Inputcurr*1e12}'] = (sweep_ext.spike_feature("slow_trough_t")[0] - sweep_ext.spike_feature("peak_t")[0])/features[f'ISI1_{Inputcurr*1e12}']
#mAHP_APp_dur = mAHP of second last spike (penultimate)
features[f'mAHP_APp_dur_{Inputcurr*1e12}'] = (sweep_ext.spike_feature("slow_trough_t")[-2] - sweep_ext.spike_feature("peak_t")[-2])/features[f'ISIl_{Inputcurr*1e12}']
#ADP_AP1_amp
features[f'ADP_AP1_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("adp_v")[0] - features[f'E_rest_{Inputcurr*1e12}']
#ADP_APp_amp
features[f'ADP_APp_amp_{Inputcurr*1e12}'] = sweep_ext.spike_feature("adp_v")[-2] - features[f'E_rest_{Inputcurr*1e12}']
#mAHP_stimend_amp = within 50ms
end50_idx = (np.abs(t - stim_end - 50e-3)).argmin()
print(end50_idx)
features[f'mAHP_stimend_amp_{Inputcurr*1e12}'] = np.min(v[end_idx:end50_idx]) - features[f'E_rest_{Inputcurr*1e12}']
#sAHP_stimend_amp = within 200ms
end200_idx = (np.abs(t - stim_end - 200e-3)).argmin()
features[f'sAHP_stimend_amp_{Inputcurr*1e12}'] = np.min(v[end_idx:end200_idx]) - features[f'E_rest_{Inputcurr*1e12}']
return [features, t, v]
def calculate_feature_errors(self, t_ms, v, i):
# Special case checks and penalty values
minimum_num_spikes = 2
missing_penalty_value = 20.0
max_fail_penalty = 250.0
min_fail_penalty = 75.0
overkill_reduction = 0.75
variance_factor = 0.1
fail_trace = False
delay = self.stim.delay * 1e-3
duration = self.stim.dur * 1e-3
t = t_ms * 1e-3
feature_names = self.description.data['features']
# penalize for failing to return to rest
start_index = np.flatnonzero(t >= delay)[0]
if np.abs(v[-1] - v[:start_index].mean()) > 2.0:
fail_trace = True
else:
swp = EphysSweepFeatureExtractor(t,
v,
i,
start=0,
end=delay,
filter=None)
swp.process_spikes()
if swp.sweep_feature("avg_rate") > 0:
fail_trace = True
target_features = self.description.data['target_features']
target_features_dict = {
f["name"]: {
"mean": f["mean"],
"stdev": f["stdev"]
}
for f in target_features
}
if not fail_trace:
swp = EphysSweepFeatureExtractor(t,
v,
i,
start=delay,
end=(delay + duration),
filter=None)
swp.process_spikes()
if len(swp.spikes()) < minimum_num_spikes: # Enough spikes?
fail_trace = True
else:
avg_per_spike_peak_error = np.mean([
abs(spk["peak_v"] - target_features_dict["peak_v"]["mean"])
for spk in swp.spikes()
])
avg_overall_error = abs(
target_features_dict["peak_v"]["mean"] -
swp.spike_feature("peak_v").mean())
if avg_per_spike_peak_error > 3.0 * avg_overall_error: # Weird bi-modality of spikes; 3.0 is arbitrary
fail_trace = True
if fail_trace:
variance_start = np.flatnonzero(t >= delay - 0.1)[0]
variance_end = np.flatnonzero(t >= (delay + duration) / 2.0)[0]
trace_variance = v[variance_start:variance_end].var()
error_value = max(
max_fail_penalty - trace_variance * variance_factor,
min_fail_penalty)
errs = np.ones(len(feature_names)) * error_value
else:
errs = []
# Calculate additional features not done by swp.process_spikes()
baseline_v = swp.sweep_feature("v_baseline")
other_features = {}
threshold_t = swp.spike_feature("threshold_t")
fast_trough_t = swp.spike_feature("fast_trough_t")
slow_trough_t = swp.spike_feature("slow_trough_t")
delta_t = slow_trough_t - fast_trough_t
delta_t[np.isnan(delta_t)] = 0.
other_features["slow_trough_delta_time"] = np.mean(
delta_t[:-1] / np.diff(threshold_t))
fast_trough_v = swp.spike_feature("fast_trough_v")
slow_trough_v = swp.spike_feature("slow_trough_v")
delta_v = fast_trough_v - slow_trough_v
delta_v[np.isnan(delta_v)] = 0.
other_features["slow_trough_delta_v"] = delta_v.mean()
for f in feature_names:
target_mean = target_features_dict[f]['mean']
target_stdev = target_features_dict[f]['stdev']
if target_stdev == 0:
print("Feature with 0 stdev: ", f)
if f in swp.spike_feature_keys():
model_mean = swp.spike_feature(f).mean()
elif f in swp.sweep_feature_keys():
model_mean = swp.sweep_feature(f)
elif f in other_features:
model_mean = other_features[f]
else:
model_mean = np.nan
if np.isnan(model_mean):
errs.append(missing_penalty_value)
else:
errs.append(
np.abs((model_mean - target_mean) / target_stdev))
errs = np.array(errs)
return errs