def pvalues(self, pBackgroundModel, pDataList, pIndexReferencePoint):
# cdf
p_value_list = []
for i, data_element in enumerate(pDataList):
relative_distance = i - pIndexReferencePoint
# check if relative distance is available.
# if not, use either min or max key for the distribution
if relative_distance not in pBackgroundModel:
if relative_distance < 0:
relative_distance = min(pBackgroundModel.keys())
else:
relative_distance = max(pBackgroundModel.keys())
if data_element == 0.0:
p_value_list.append(0.0)
else:
p_value_list.append(
cnb.cdf(data_element,
pBackgroundModel[relative_distance][0],
pBackgroundModel[relative_distance][1]))
# for reason I do not understand atm the values needs to be inverted again, it seems it is not enough to do this in try/catch region
p_value_list = np.array(p_value_list, dtype=np.float64)
p_value_list = 1 - p_value_list
# remove possible occuring nan with a p-value of 1
mask = np.isnan(p_value_list)
mask_inf = np.isinf(p_value_list)
p_value_list = np.array(p_value_list)
mask = np.logical_or(mask, mask_inf)
p_value_list[mask] = 1.0
return p_value_list
def compute_new_p_values(pData, pBackgroundModel, pPValue, pMergedLinesDict,
pPeakInteractionsThreshold, pViewpointObj):
accepted = {}
accepted_lines = []
if isinstance(pPValue, float):
for key in pData:
if key in pBackgroundModel:
log.debug('Recompute p-values. Old: {}'.format(pData[key][-3]))
pData[key][-3] = 1 - cnb.cdf(float(pData[key][-1]),
float(pBackgroundModel[key][0]),
float(pBackgroundModel[key][1]))
log.debug('new {}\n\n'.format(pData[key][-3]))
if pData[key][-3] <= pPValue:
if float(pData[key][-1]) >= pPeakInteractionsThreshold:
accepted[key] = pData[key]
target_content = pMergedLinesDict[key][0][:3]
target_content[2] = pMergedLinesDict[key][-1][2]
accepted_lines.append(target_content)
else:
log.debug('key not in background')
elif isinstance(pPValue, dict):
for key in pData:
if key in pBackgroundModel:
log.debug('Recompute p-values. Old: {}'.format(pData[key][-3]))
pData[key][-3] = 1 - cnb.cdf(float(pData[key][-1]),
float(pBackgroundModel[key][0]),
float(pBackgroundModel[key][1]))
log.debug('new {}\n\n'.format(pData[key][-3]))
if pData[key][-3] <= pPValue[key]:
if float(pData[key][-1]) >= pPeakInteractionsThreshold:
accepted[key] = pData[key]
target_content = pMergedLinesDict[key][0][:3]
target_content[2] = pMergedLinesDict[key][-1][2]
accepted_lines.append(target_content)
else:
log.debug('key not in background')
return accepted, accepted_lines
def compute_p_values_mask(pGenomicDistanceDistributionsObsExp, pGenomicDistanceDistributionsKeyList,
pPValuePreselection, pGenomicDistanceDistributionPosition, pResolution,
pMinimumInteractionsThreshold, pObsExpThreshold, pQueue):
try:
true_values = []
if len(pGenomicDistanceDistributionsKeyList) == 0:
pQueue.put(true_values)
return
float_dict = isinstance(pPValuePreselection, float)
for i, key in enumerate(pGenomicDistanceDistributionsKeyList):
data_obs_exp = np.array(pGenomicDistanceDistributionsObsExp[key])
# do not fit and not compute any p-value if all values on this distance are small than the pMinimumInteractionsThreshold
mask = data_obs_exp >= pObsExpThreshold
nbinom_parameters = fit_nbinom.fit(data_obs_exp)
p_value = 1 - cnb.cdf(data_obs_exp[mask], nbinom_parameters['size'], nbinom_parameters['prob'])
if float_dict:
mask_distance = p_value <= pPValuePreselection
else:
key_genomic = int(key * pResolution)
mask_distance = p_value <= pPValuePreselection[key_genomic]
j = 0
for k, value in enumerate(mask):
if value:
if mask_distance[j]:
true_values.append(pGenomicDistanceDistributionPosition[key][k])
j += 1
except Exception as exp:
pQueue.put('Fail: ' + str(exp) + traceback.format_exc())
return
pQueue.put(true_values)
return