def coda(df, window, level):
"""
CODA processing from Windig, Phalp, & Payne 1996 Anal Chem
"""
# pull out the data
d = df.values
# smooth the data and standardize it
smooth_data = movingaverage(d, df.index, window)[0]
stand_data = (smooth_data - smooth_data.mean()) / smooth_data.std()
# scale the data to have unit length
scale_data = d / np.sqrt(np.sum(d ** 2, axis=0))
# calculate the "mass chromatographic quality" (MCQ) index
mcq = np.sum(stand_data * scale_data, axis=0) / np.sqrt(d.shape[0] - 1)
# filter out ions with an mcq below level
good_ions = [i for i, q in zip(df.columns, mcq) if q >= level]
return good_ions
def simple_peak_find(s, init_slope=500, start_slope=500, end_slope=200,
min_peak_height=50, max_peak_width=1.5):
"""
Given a Series, return a list of tuples indicating when
peaks start and stop and what their baseline is.
[(t_start, t_end, hints) ...]
"""
point_gap = 10
def slid_win(itr, size=2):
"""Returns a sliding window of size 'size' along itr."""
itr, buf = iter(itr), []
for _ in range(size):
buf += [next(itr)]
for l in itr:
yield buf
buf = buf[1:] + [l]
yield buf
# TODO: check these smoothing defaults
y, t = s.values, s.index.astype(float)
smooth_y = movingaverage(y, 9)
dxdt = np.gradient(smooth_y) / np.gradient(t)
# dxdt = -savitzkygolay(ts, 5, 3, deriv=1).y / np.gradient(t)
init_slopes = np.arange(len(dxdt))[dxdt > init_slope]
if len(init_slopes) == 0:
return []
# get the first points of any "runs" as a peak start
# runs can have a gap of up to 10 points in them
peak_sts = [init_slopes[0]]
peak_sts += [j for i, j in slid_win(init_slopes, 2) if j - i > 10]
peak_sts.sort()
en_slopes = np.arange(len(dxdt))[dxdt < -end_slope]
if len(en_slopes) == 0:
return []
# filter out any lone points farther than 10 away from their neighbors
en_slopes = [en_slopes[0]]
en_slopes += [i[1] for i in slid_win(en_slopes, 3)
if i[1] - i[0] < point_gap or i[2] - i[1] < point_gap]
en_slopes += [en_slopes[-1]]
# get the last points of any "runs" as a peak end
peak_ens = [j for i, j in slid_win(en_slopes[::-1], 2)
if i - j > point_gap] + [en_slopes[-1]]
peak_ens.sort()
# avals = np.arange(len(t))[np.abs(t - 0.675) < 0.25]
# print([i for i in en_slopes if i in avals])
# print([(t[i], i) for i in peak_ens if i in avals])
peak_list = []
pk2 = 0
for pk in peak_sts:
# don't allow overlapping peaks
if pk < pk2:
continue
# track backwards to find the true start
while dxdt[pk] > start_slope and pk > 0:
pk -= 1
# now find where the peak ends
dist_to_end = np.array(peak_ens) - pk
pos_end = pk + dist_to_end[dist_to_end > 0]
for pk2 in pos_end:
if (y[pk2] - y[pk]) / (t[pk2] - t[pk]) > start_slope:
# if the baseline beneath the peak is too large, let's
# keep going to the next dip
peak_list.append({'t0': t[pk], 't1': t[pk2]})
pk = pk2
elif t[pk2] - t[pk] > max_peak_width:
# make sure that peak is short enough
pk2 = pk + np.abs(t[pk:] - t[pk] - max_peak_width).argmin()
break
else:
break
else:
# if no end point is found, the end point
# is the end of the timeseries
pk2 = len(t) - 1
if pk == pk2:
continue
pk_hgt = max(y[pk:pk2]) - min(y[pk:pk2])
if pk_hgt < min_peak_height:
continue
peak_list.append({'t0': t[pk], 't1': t[pk2]})
return peak_list
def simple_peak_find(s,
init_slope=500,
start_slope=500,
end_slope=200,
min_peak_height=50,
max_peak_width=1.5):
"""
Given a Series, return a list of tuples indicating when
peaks start and stop and what their baseline is.
[(t_start, t_end, hints) ...]
"""
point_gap = 10
def slid_win(itr, size=2):
"""Returns a sliding window of size 'size' along itr."""
itr, buf = iter(itr), []
for _ in range(size):
buf += [next(itr)]
for l in itr:
yield buf
buf = buf[1:] + [l]
yield buf
# TODO: check these smoothing defaults
y, t = s.values, s.index.astype(float)
smooth_y = movingaverage(y, 9)
dxdt = np.gradient(smooth_y) / np.gradient(t)
# dxdt = -savitzkygolay(ts, 5, 3, deriv=1).y / np.gradient(t)
init_slopes = np.arange(len(dxdt))[dxdt > init_slope]
if len(init_slopes) == 0:
return []
# get the first points of any "runs" as a peak start
# runs can have a gap of up to 10 points in them
peak_sts = [init_slopes[0]]
peak_sts += [j for i, j in slid_win(init_slopes, 2) if j - i > 10]
peak_sts.sort()
en_slopes = np.arange(len(dxdt))[dxdt < -end_slope]
if len(en_slopes) == 0:
return []
# filter out any lone points farther than 10 away from their neighbors
en_slopes = [en_slopes[0]]
en_slopes += [
i[1] for i in slid_win(en_slopes, 3)
if i[1] - i[0] < point_gap or i[2] - i[1] < point_gap
]
en_slopes += [en_slopes[-1]]
# get the last points of any "runs" as a peak end
peak_ens = [
j for i, j in slid_win(en_slopes[::-1], 2) if i - j > point_gap
] + [en_slopes[-1]]
peak_ens.sort()
# avals = np.arange(len(t))[np.abs(t - 0.675) < 0.25]
# print([i for i in en_slopes if i in avals])
# print([(t[i], i) for i in peak_ens if i in avals])
peak_list = []
pk2 = 0
for pk in peak_sts:
# don't allow overlapping peaks
if pk < pk2:
continue
# track backwards to find the true start
while dxdt[pk] > start_slope and pk > 0:
pk -= 1
# now find where the peak ends
dist_to_end = np.array(peak_ens) - pk
pos_end = pk + dist_to_end[dist_to_end > 0]
for pk2 in pos_end:
if (y[pk2] - y[pk]) / (t[pk2] - t[pk]) > start_slope:
# if the baseline beneath the peak is too large, let's
# keep going to the next dip
peak_list.append({'t0': t[pk], 't1': t[pk2]})
pk = pk2
elif t[pk2] - t[pk] > max_peak_width:
# make sure that peak is short enough
pk2 = pk + np.abs(t[pk:] - t[pk] - max_peak_width).argmin()
break
else:
break
else:
# if no end point is found, the end point
# is the end of the timeseries
pk2 = len(t) - 1
if pk == pk2:
continue
pk_hgt = max(y[pk:pk2]) - min(y[pk:pk2])
if pk_hgt < min_peak_height:
continue
peak_list.append({'t0': t[pk], 't1': t[pk2]})
return peak_list
