def test_count(data, opts):
"""
Make a test for all genes iteratively.
@args data: Store all input data and results
@type data: Class object
@args opts: Input argument to the main TE function
@type opts: Instance
"""
print 'Start the statistical test.'
num = len(data.geneIDs)
pval = np.empty((num, 1))
pval.fill(np.nan)
explanatory0 = cm.create_matrix(data, model='H0')
explanatory1 = cm.create_matrix(data, model='H1')
librarySizes = np.hstack([data.libSizesRibo, data.libSizesRna])
lenSampleRibo = data.idxRibo.size
lenSampleRna = data.idxRna.size
for i in range(num):
sys.stdout.flush()
if i % 50 == 0:
print '\r%i genes finished...' % i ,
if i+1 == num:
print '\r%i genes finished.' % num
if opts.dispDiff and np.isnan(data.dispAdjRibo[i]):
continue
if not opts.dispDiff and np.isnan(data.dispAdj[i]):
continue
response = np.hstack([data.countRibo[i, :], data.countRna[i, :]])
if opts.dispDiff:
disp = np.hstack([np.repeat(data.dispAdjRibo[i], lenSampleRibo), np.repeat(data.dispAdjRna[i], lenSampleRna)])
else:
disp = data.dispAdj[i]
modNB0 = sm.GLM(response, explanatory0, family=sm.families.NegativeBinomial(alpha=disp), offset=np.log(librarySizes))
modNB1 = sm.GLM(response, explanatory1, family=sm.families.NegativeBinomial(alpha=disp), offset=np.log(librarySizes))
result0 = modNB0.fit()
result1 = modNB1.fit()
pval[i] = 1 - chi2.cdf(result0.deviance - result1.deviance, explanatory1.shape[1] - explanatory0.shape[1])
data.pval = pval
return data
def estimate_disp(data, opts):
""" Create explanatory matrix and estimate dispersion.
Temporarily save data in ./TmpData.pkl file.
@args data: Store all input data and results
@type data: Class object
@args opts: Input arguments to the main TE function
@type opts: Instance
"""
explanatory = cm.create_matrix(data, model='H1')
data.matrix = explanatory
outpath = opts.resPath
pklFile = outpath + 'TmpData.pkl'
if opts.dispDiff:
data = rd.disper_raw(data, opts)
else:
data = rd.disper_raw_scalar(data, opts)
with open(pklFile, 'wb') as FileOut:
pickle.dump(data, FileOut, pickle.HIGHEST_PROTOCOL)
print '*' * 25
data = fd.disper_fit(data, opts)
with open(pklFile, 'wb') as FileOut:
pickle.dump(data, FileOut, pickle.HIGHEST_PROTOCOL)
print '*' * 25
if opts.dispDiff:
data = ad.disper_adj(data, opts)
else:
data = ad.disper_adj_scalar(data, opts)
with open(pklFile, 'wb') as FileOut:
pickle.dump(data, FileOut, pickle.HIGHEST_PROTOCOL)
return data
def estimate_disp(data, opts):
""" Create explanatory matrix and estimate dispersion.
Temporarily save data in ./TmpData.pkl file.
@args data: Store all input data and results
@type data: Class object
@args opts: Input arguments to the main TE function
@type opts: Instance
"""
explanatory = cm.create_matrix(data, model='H1')
data.matrix = explanatory
outpath = opts.resPath
pklFile = outpath + 'TmpData.pkl'
if opts.dispDiff:
data = rd.disper_raw(data, opts)
else:
data = rd.disper_raw_scalar(data, opts)
with open(pklFile, 'wb') as FileOut:
pickle.dump(data, FileOut, pickle.HIGHEST_PROTOCOL)
print '*'*25
data = fd.disper_fit(data, opts)
with open(pklFile, 'wb') as FileOut:
pickle.dump(data, FileOut, pickle.HIGHEST_PROTOCOL)
print '*'*25
if opts.dispDiff:
data = ad.disper_adj(data, opts)
else:
data = ad.disper_adj_scalar(data, opts)
with open(pklFile, 'wb') as FileOut:
pickle.dump(data, FileOut, pickle.HIGHEST_PROTOCOL)
return data
def test_count(data, opts):
"""
Make a test for all genes iteratively.
@args data: Store all input data and results
@type data: Class object
@args opts: Input argument to the main TE function
@type opts: Instance
"""
print 'Start the statistical test.'
num = len(data.geneIDs)
pval = np.empty((num, 1))
pval.fill(np.nan)
explanatory0 = cm.create_matrix(data, model='H0')
explanatory1 = cm.create_matrix(data, model='H1')
librarySizes = np.hstack([data.libSizesRibo, data.libSizesRna])
lenSampleRibo = data.idxRibo.size
lenSampleRna = data.idxRna.size
errorCnt = 0
for i in range(num):
if opts.dispDiff and np.isnan(data.dispAdjRibo[i]):
continue
if not opts.dispDiff and np.isnan(data.dispAdj[i]):
continue
response = np.hstack([data.countRibo[i, :], data.countRna[i, :]])
if opts.dispDiff:
disp = np.hstack([
np.repeat(data.dispAdjRibo[i], lenSampleRibo),
np.repeat(data.dispAdjRna[i], lenSampleRna)
])
else:
disp = data.dispAdj[i]
try:
modNB0 = sm.GLM(response,
explanatory0,
family=sm.families.NegativeBinomial(alpha=disp),
offset=np.log(librarySizes))
modNB1 = sm.GLM(response,
explanatory1,
family=sm.families.NegativeBinomial(alpha=disp),
offset=np.log(librarySizes))
result0 = modNB0.fit()
result1 = modNB1.fit()
except sm.tools.sm_exceptions.PerfectSeparationError:
errorCnt += 1
else:
if not opts.dispDiff:
pval[i] = 1 - chi2.cdf(
result0.deviance - result1.deviance,
explanatory1.shape[1] - explanatory0.shape[1])
elif opts.dispDiff:
pval[i] = 1 - chi2.cdf(
result0.deviance - result1.deviance,
(explanatory1.shape[1] - explanatory0.shape[1]) / 2.5)
else:
pass
data.pval = pval
sys.stdout.write(
'Warning: Failed to do test: %i genes. P value set to \'nan\'.\n' %
errorCnt)
return data
def test_count(data, opts):
"""
Make a test for all genes iteratively.
@args data: Store all input data and results
@type data: Class object
@args opts: Input argument to the main TE function
@type opts: Instance
"""
print 'Start the statistical test.'
num = len(data.geneIDs)
pval = np.empty((num, 1))
pval.fill(np.nan)
explanatory0 = cm.create_matrix(data, model='H0')
explanatory1 = cm.create_matrix(data, model='H1')
librarySizes = np.hstack([data.libSizesRibo, data.libSizesRna])
lenSampleRibo = data.idxRibo.size
lenSampleRna = data.idxRna.size
for i in range(num):
sys.stdout.flush()
if i % 50 == 0:
print '\r%i genes finished...' % i,
if i + 1 == num:
print '\r%i genes finished.' % num
if opts.dispDiff and np.isnan(data.dispAdjRibo[i]):
continue
if not opts.dispDiff and np.isnan(data.dispAdj[i]):
continue
response = np.hstack([data.countRibo[i, :], data.countRna[i, :]])
if opts.dispDiff:
disp = np.hstack([
np.repeat(data.dispAdjRibo[i], lenSampleRibo),
np.repeat(data.dispAdjRna[i], lenSampleRna)
])
else:
disp = data.dispAdj[i]
modNB0 = sm.GLM(response,
explanatory0,
family=sm.families.NegativeBinomial(alpha=disp),
offset=np.log(librarySizes))
modNB1 = sm.GLM(response,
explanatory1,
family=sm.families.NegativeBinomial(alpha=disp),
offset=np.log(librarySizes))
result0 = modNB0.fit()
result1 = modNB1.fit()
pval[i] = 1 - chi2.cdf(result0.deviance - result1.deviance,
explanatory1.shape[1] - explanatory0.shape[1])
data.pval = pval
return data