def analyze_response_strength(rec, source, remove_artifacts=False, deconvolve=True, lpf=True, bsub=True, lowpass=1000):
"""Perform a standardized strength analysis on a record selected by response_query or baseline_query.
1. Determine timing of presynaptic stimulus pulse edges and spike
2. Measure peak deflection on raw trace
3. Apply deconvolution / artifact removal / lpf
4. Measure peak deflection on deconvolved trace
"""
data = Trace(rec.data, sample_rate=db.default_sample_rate)
if source == 'pulse_response':
# Find stimulus pulse edges for artifact removal
start = rec.pulse_start - rec.rec_start
pulse_times = [start, start + rec.pulse_dur]
if rec.spike_time is None:
# these pulses failed QC, but we analyze them anyway to make all data visible
spike_time = 11e-3
else:
spike_time = rec.spike_time - rec.rec_start
elif source == 'baseline':
# Fake stimulus information to ensure that background data receives
# the same filtering / windowing treatment
pulse_times = [10e-3, 12e-3]
spike_time = 11e-3
else:
raise ValueError("Invalid source %s" % source)
results = {}
results['raw_trace'] = data
results['pulse_times'] = pulse_times
results['spike_time'] = spike_time
# Measure crosstalk from pulse onset
p1 = data.time_slice(pulse_times[0]-200e-6, pulse_times[0]).median()
p2 = data.time_slice(pulse_times[0], pulse_times[0]+200e-6).median()
results['crosstalk'] = p2 - p1
# crosstalk artifacts in VC are removed before deconvolution
if rec.clamp_mode == 'vc' and remove_artifacts is True:
data = remove_crosstalk_artifacts(data, pulse_times)
remove_artifacts = False
# Measure deflection on raw data
results['pos_amp'], _ = measure_peak(data, '+', spike_time, pulse_times)
results['neg_amp'], _ = measure_peak(data, '-', spike_time, pulse_times)
# Deconvolution / artifact removal / filtering
if deconvolve:
tau = 15e-3 if rec.clamp_mode == 'ic' else 5e-3
else:
tau = None
dec_data = deconv_filter(data, pulse_times, tau=tau, lpf=lpf, remove_artifacts=remove_artifacts, bsub=bsub, lowpass=lowpass)
results['dec_trace'] = dec_data
# Measure deflection on deconvolved data
results['pos_dec_amp'], results['pos_dec_latency'] = measure_peak(dec_data, '+', spike_time, pulse_times)
results['neg_dec_amp'], results['neg_dec_latency'] = measure_peak(dec_data, '-', spike_time, pulse_times)
return results
def analyze_response_strength(rec, source, remove_artifacts=False, lpf=True, bsub=True, lowpass=1000):
"""Perform a standardized strength analysis on a record selected by response_query or baseline_query.
1. Determine timing of presynaptic stimulus pulse edges and spike
2. Measure peak deflection on raw trace
3. Apply deconvolution / artifact removal / lpf
4. Measure peak deflection on deconvolved trace
"""
data = Trace(rec.data, sample_rate=db.default_sample_rate)
if source == 'pulse_response':
# Find stimulus pulse edges for artifact removal
start = rec.pulse_start - rec.rec_start
pulse_times = [start, start + rec.pulse_dur]
if rec.spike_time is None:
# these pulses failed QC, but we analyze them anyway to make all data visible
spike_time = 11e-3
else:
spike_time = rec.spike_time - rec.rec_start
elif source == 'baseline':
# Fake stimulus information to ensure that background data receives
# the same filtering / windowing treatment
pulse_times = [10e-3, 12e-3]
spike_time = 11e-3
else:
raise ValueError("Invalid source %s" % source)
results = {}
results['raw_trace'] = data
results['pulse_times'] = pulse_times
results['spike_time'] = spike_time
# Measure crosstalk from pulse onset
p1 = data.time_slice(pulse_times[0]-200e-6, pulse_times[0]).median()
p2 = data.time_slice(pulse_times[0], pulse_times[0]+200e-6).median()
results['crosstalk'] = p2 - p1
# crosstalk artifacts in VC are removed before deconvolution
if rec.clamp_mode == 'vc' and remove_artifacts is True:
data = remove_crosstalk_artifacts(data, pulse_times)
remove_artifacts = False
# Measure deflection on raw data
results['pos_amp'], _ = measure_peak(data, '+', spike_time, pulse_times)
results['neg_amp'], _ = measure_peak(data, '-', spike_time, pulse_times)
# Deconvolution / artifact removal / filtering
tau = 15e-3 if rec.clamp_mode == 'ic' else 5e-3
dec_data = deconv_filter(data, pulse_times, tau=tau, lpf=lpf, remove_artifacts=remove_artifacts, bsub=bsub, lowpass=lowpass)
results['dec_trace'] = dec_data
# Measure deflection on deconvolved data
results['pos_dec_amp'], results['pos_dec_latency'] = measure_peak(dec_data, '+', spike_time, pulse_times)
results['neg_dec_amp'], results['neg_dec_latency'] = measure_peak(dec_data, '-', spike_time, pulse_times)
return results
def test_trace_timing():
# Make sure sample timing is handled exactly--need to avoid fp error here
a = np.random.normal(size=300)
sr = 50000
dt = 2e-5
t = np.arange(len(a)) * dt
# trace with no timing information
tr = Trace(a)
assert not tr.has_timing
assert not tr.has_time_values
with raises(TypeError):
tr.dt
with raises(TypeError):
tr.sample_rate
with raises(TypeError):
tr.time_values
view = tr[100:200]
assert not tr.has_timing
assert not tr.has_time_values
# invalid data
with raises(ValueError):
Trace(data=np.zeros((10, 10)))
# invalid timing information
with raises(TypeError):
Trace(data=a, dt=dt, time_values=t)
with raises(TypeError):
Trace(data=a, sample_rate=sr, time_values=t)
with raises(TypeError):
Trace(data=a, dt=dt, t0=0, time_values=t)
with raises(TypeError):
Trace(data=a, dt=dt, t0=0, sample_rate=sr)
with raises(ValueError):
Trace(data=a, time_values=t[:-1])
# trace with only dt
tr = Trace(a, dt=dt)
assert tr.dt == dt
assert np.allclose(tr.sample_rate, sr)
assert np.all(tr.time_values == t)
assert tr.has_timing
assert not tr.has_time_values
assert tr.regularly_sampled
# test view
view = tr[100:200]
assert view.t0 == tr.time_values[100]
assert view.time_values[0] == view.t0
assert view.dt == tr.dt
assert view._meta['sample_rate'] is None
assert not view.has_time_values
view = tr.time_slice(100*dt, 200*dt)
assert view.t0 == tr.time_values[100]
assert view.time_values[0] == view.t0
assert view.dt == tr.dt
assert view._meta['sample_rate'] is None
assert not view.has_time_values
# test nested view
view2 = view.time_slice(view.t0 + 20*dt, view.t0 + 50*dt)
assert view2.t0 == view.time_values[20] == tr.time_values[120]
# trace with only sample_rate
tr = Trace(a, sample_rate=sr)
assert tr.dt == dt
assert tr.sample_rate == sr
assert np.all(tr.time_values == t)
assert tr.has_timing
assert not tr.has_time_values
assert tr.regularly_sampled
# test view
view = tr[100:200]
assert view.t0 == tr.time_values[100]
assert view.time_values[0] == view.t0
assert view.sample_rate == tr.sample_rate
assert view._meta['dt'] is None
assert not view.has_time_values
# trace with only regularly-sampled time_values
tr = Trace(a, time_values=t)
assert tr.dt == dt
assert np.allclose(tr.sample_rate, sr)
assert np.all(tr.time_values == t)
assert tr.has_timing
assert tr.has_time_values
assert tr.regularly_sampled
# test view
view = tr[100:200]
assert view.t0 == tr.time_values[100]
assert view.time_values[0] == view.t0
assert view._meta['dt'] is None
assert view._meta['sample_rate'] is None
assert view.has_time_values
assert view.regularly_sampled
# trace with irregularly-sampled time values
t1 = np.cumsum(np.random.normal(loc=1, scale=0.02, size=a.shape))
tr = Trace(a, time_values=t1)
assert tr.dt == t1[1] - t1[0]
assert np.all(tr.time_values == t1)
assert tr.has_timing
assert tr.has_time_values
assert not tr.regularly_sampled
# test view
view = tr[100:200]
assert view.t0 == tr.time_values[100]
assert view.time_values[0] == view.t0
assert view._meta['dt'] is None
assert view._meta['sample_rate'] is None
assert view.has_time_values
assert not view.regularly_sampled