def compute_pval_oneclass(X, null_dist, Y=None, single=False, B=9, c=5):
mictools.utils.check_data(X, Y=Y)
bins = np.linspace(0, 1, NULL_HIST_RES + 1)
# observed values/distribution
names = ['Var1', 'Var2']
Xa = X.values
if Y is None:
_, tic = minepy.pstats(Xa, alpha=B, c=c, est="mic_e")
index = pd.MultiIndex.from_tuples(list(
itertools.combinations(X.index, 2)),
names=names)
else:
Ya = Y.values
if single:
_, tic = mictools.utils.sstats(Xa, Ya, alpha=B, c=c, est="mic_e")
index = pd.MultiIndex.from_arrays([X.index, Y.index], names=names)
else:
_, tic = minepy.cstats(Xa, Ya, alpha=B, c=c, est="mic_e")
index = pd.MultiIndex.from_product([X.index, Y.index], names=names)
tic = tic.flatten()
observed_hist = np.histogram(tic, bins)[0].astype(np.int64)
# right-tailed area
observed_hist_cum = np.cumsum(observed_hist[::-1])[::-1]
# p-values
null_hist_cum = null_dist["NullCountCum"].values
pval = (np.interp(tic, bins[:-1], null_hist_cum) + 1) / \
(null_hist_cum[0] + 1)
pval = pd.Series(pval, index=index)
# observed
obs = pd.Series(tic, index=index)
# distribution
index = pd.MultiIndex.from_arrays([bins[:-1], bins[1:]],
names=('BinStart', 'BinEnd'))
obs_dist = pd.DataFrame(
{
"ObsCount": observed_hist,
"ObsCountCum": observed_hist_cum
},
index=index,
columns=["ObsCount", "ObsCountCum"])
return obs_dist, obs, pval
def main(params, inputs, outputs):
### 输入数据 ###
X = inputs.x
Y = inputs.y
### 输入参数 ###
type = params.type
### 计算X变量间的最大信息系数MIC和总信息系数TIC ###
if type == 'X之间':
mic, tic = pstats(X, alpha=9, c=5, est="mic_e")
### 计算X与Y间的最大信息系数MIC和总信息系数TIC ###
if type == 'X与Y之间':
mic, tic = cstats(X, Y, alpha=9, c=5, est="mic_e")
### 输出结果 ###
pickle.dump(mic, open(outputs.mic, "wb"))
pickle.dump(tic, open(outputs.tic, "wb"))
import numpy as np from minepy import pstats, cstats import time np.random.seed(0) # build the X matrix, 8 variables, 320 samples X = np.random.rand(8, 320) # build the Y matrix, 4 variables, 320 samples Y = np.random.rand(4, 320) # compute pairwise statistics MIC_e and normalized TIC_e between samples in X, # B=9, c=5 mic_p, tic_p = pstats(X, alpha=9, c=5, est="mic_e") # compute statistics between each pair of samples in X and Y mic_c, tic_c = cstats(X, Y, alpha=9, c=5, est="mic_e") print "normalized TIC_e (X):" print tic_p print "MIC_e (X vs. Y):" print mic_c
# encoding=utf-8
import numpy as np
from minepy import pstats, cstats
import time
np.random.seed(0)
# build the X matrix, 8 variables, 320 samples
X = np.random.rand(8, 320)
# build the Y matrix, 4 variables, 320 samples
Y = np.random.rand(4, 320)
# compute pairwise statistics MIC_e and normalized TIC_e between samples in X,
# B=9, c=5
mic_p, tic_p = pstats(X, alpha=9, c=5, est="mic_e")
# compute statistics between each pair of samples in X and Y
mic_c, tic_c = cstats(X, Y, alpha=9, c=5, est="mic_e")
print("normalized TIC_e (X):")
print(tic_p)
print("MIC_e (X vs. Y):")
print(mic_c)
def score_interaction(bait, prey, bait_monomer_sec_id, prey_monomer_sec_id):
def longest_intersection(arr):
# Compute longest continuous stretch
n = len(arr)
s = set()
ans = 0
for ele in arr:
s.add(ele)
for i in range(n):
if (arr[i] - 1) not in s:
j = arr[i]
while (j in s):
j += 1
ans = max(ans, j - arr[i])
return ans
def normalized_xcorr(a, b):
# Normalize matrices
a = (a - np.mean(a, axis=1, keepdims=True)) / (np.std(
a, axis=1, keepdims=True))
b = (b - np.mean(b, axis=1, keepdims=True)) / (np.std(
b, axis=1, keepdims=True))
nxcorr = [] # normalized cross-correlation
lxcorr = [] # cross-correlation lag
if np.array_equal(a, b):
# Compare all rows of a against all rows of a, including itself (auto-correlation)
for i in range(0, len(a)):
for j in range(i, len(a)):
nxcorr.append(
np.correlate(a[i], a[j], 'valid')[0] /
len(a[i])) # Normalize by length
lxcorr.append(np.argmax(np.correlate(a[i], a[j],
'same'))) # Peak
else:
# Compare all rows of a against all rows of b
for i in range(0, len(a)):
for j in range(0, len(b)):
nxcorr.append(
np.correlate(a[i], b[j], 'valid')[0] /
len(a[i])) # Normalize by length
lxcorr.append(np.argmax(np.correlate(a[i], b[j],
'same'))) # Peak
return np.array(nxcorr), np.array(lxcorr)
def sec_xcorr(bm, pm):
# Compute SEC xcorr scores
bnx, blx = normalized_xcorr(bm, bm)
pnx, plx = normalized_xcorr(pm, pm)
bpnx, bplx = normalized_xcorr(bm, pm)
xcorr_shape = np.mean(bpnx)
xcorr_apex = np.mean(bplx)
xcorr_shift = max(
[abs(xcorr_apex - np.mean(blx)),
abs(xcorr_apex - np.mean(plx))])
return xcorr_shape, xcorr_shift, xcorr_apex
def mass_similarity(bm, pm):
# Sum bait and prey peptides
bpabundance = np.sum(bm, axis=1, keepdims=True).mean()
ppabundance = np.sum(pm, axis=1, keepdims=True).mean()
# Compute abundance ratio of bait and prey protein
abundance_ratio = bpabundance / ppabundance
if abundance_ratio > 1:
abundance_ratio = 1 / abundance_ratio
return abundance_ratio
# Compute bait and prey overlap
overlap = (np.nansum(bait, axis=0) > 0) | (np.nansum(prey, axis=0) > 0)
total_overlap = np.count_nonzero(overlap)
# Compute bait and prey intersection
intersection = (np.nansum(bait, axis=0) > 0) & (np.nansum(prey, axis=0) >
0)
total_intersection = np.count_nonzero(intersection)
if total_intersection > 0:
longest_intersection = longest_intersection(intersection.nonzero()[0])
# Require at least three consecutive overlapping data points
if longest_intersection > 2:
# Prepare total bait and prey profiles & Replace nan with 0
total_bait = np.nan_to_num(bait)
total_prey = np.nan_to_num(prey)
# Remove non-overlapping segments
bait[:, ~intersection] = np.nan
prey[:, ~intersection] = np.nan
# Remove completely empty peptides
bait = bait[(np.nansum(bait, axis=1) > 0), :]
prey = prey[(np.nansum(prey, axis=1) > 0), :]
# Replace nan with 0
bait = np.nan_to_num(bait)
prey = np.nan_to_num(prey)
# Require at least one remaining peptide for bait and prey
if (bait.shape[0] > 0) and (prey.shape[0] > 0):
# Compute cross-correlation scores
xcorr_shape, xcorr_shift, xcorr_apex = sec_xcorr(bait, prey)
# Compute MIC/TIC scores
mic_stat, tic_stat = cstats(bait[:, intersection],
prey[:, intersection],
est="mic_e")
mic = mic_stat.mean(axis=0).mean(
) # Axis 0: summary for prey peptides / Axis 1: summary for bait peptides
tic = tic_stat.mean(axis=0).mean(
) # Axis 0: summary for prey peptides / Axis 1: summary for bait peptides
# Compute mass similarity score
abundance_ratio = mass_similarity(bait, prey)
# Compute total mass similarity score
total_abundance_ratio = mass_similarity(total_bait, total_prey)
# Compute relative intersection score
relative_overlap = total_intersection / total_overlap
# Compute delta monomer score
delta_monomer = np.abs(bait_monomer_sec_id -
prey_monomer_sec_id)
# Compute apex monomer score
apex_monomer = np.min(
np.array(bait_monomer_sec_id - xcorr_apex,
prey_monomer_sec_id - xcorr_apex))
return ({
'var_xcorr_shape': xcorr_shape,
'var_xcorr_shift': xcorr_shift,
'var_abundance_ratio': abundance_ratio,
'var_total_abundance_ratio': total_abundance_ratio,
'var_mic': mic,
'var_tic': tic,
'var_sec_overlap': relative_overlap,
'var_sec_intersection': longest_intersection,
'var_delta_monomer': delta_monomer,
'var_apex_monomer': apex_monomer
})