def predict_for_deepphos(train_file_name,
sites,
predictFrame='general',
hierarchy=None,
kinase=None):
'''
:param train_file_name: input of your prdict file
it must be a .csv file and theinput format is proteinName, postion,sites, shortseq
:param sites: the sites predict: site = 'S','T' OR 'Y'
:param predictFrame: 'general' or 'kinase'
:param hierarchy: if predictFrame is kinse: you must input the hierarchy:
group,family,subfamily,kinase to choose corresponding model
:param kinase: kinase name
:return:
a file with the score
'''
win1 = 51
win2 = 33
win3 = 15
from methods.dataprocess_predict import getMatrixInput
[X_test1, y_test, ids, position] = getMatrixInput(train_file_name, sites,
win1)
[X_test2, _, _, _] = getMatrixInput(train_file_name, sites, win2)
[X_test3, _, _, _] = getMatrixInput(train_file_name, sites, win3)
# print X_test1.shape
# print len(position)
from methods.model_n import model_net
model = model_net(X_test1, X_test2, X_test3, y_test, nb_epoch=0)
#load model weight
if predictFrame == 'general':
outputfile = 'general_{:s}'.format(site)
if site == ('S', 'T'):
model_weight = './models/model_general_S,T.h5'
if site == 'Y':
model_weight = './models/model_general_Y.h5'
if predictFrame == 'kinase':
outputfile = 'kinase_{:s}_{:s}'.format(hierarchy, kinase)
model_weight = './models/model_{:s}_{:s}.h5'.format(hierarchy, kinase)
# print model_weight
model.load_weights(model_weight)
predictions_t = model.predict([X_test1, X_test2, X_test3])
results_ST = np.column_stack((ids, position, predictions_t[:, 1]))
result = pd.DataFrame(results_ST)
result.to_csv(outputfile + "prediction_phosphorylation.txt",
index=False,
header=None,
sep='\t',
quoting=csv.QUOTE_NONNUMERIC)
def train_for_deepphos(train_file_name,
site,
predictFrame='general',
background_weight=None):
'''
:param train_file_name: input of your train file
it must be a .csv file and theinput format is label,proteinName, postion,sites, shortsequence,
:param site: the sites predict: site = 'S','T' OR 'Y'
:param predictFrame: 'general' or 'kinase'
:param background_weight: the model you want load to pretrain new model
usually used in kinase training
:return:
'''
win1 = 51
win2 = 33
win3 = 15
from methods.dataprocess_train import getMatrixLabel
X_train1, y_train = getMatrixLabel(train_file_name, sites, win1)
X_train2, _ = getMatrixLabel(train_file_name, sites, win2)
X_train3, _ = getMatrixLabel(train_file_name, sites, win3)
modelname = "general_{:s}".format(site)
if predictFrame == 'general':
modelname = "general_model_{:s}".format(site)
if predictFrame == 'kinase':
modelname = "kinase_model_{:s}".format(site)
from methods.model_n import model_net
model = model_net(X_train1,
X_train2,
X_train3,
y_train,
weights=background_weight)
model.save_weights(modelname + '.h5', overwrite=True)
def predict_for_deepphos_from_json(input, organism):
'''
:param train_file_name: input of your prdict file
it must be a .csv file and theinput format is proteinName, postion,sites, shortseq
:param sites: the sites predict: site = 'S','T' OR 'Y'
:param predictFrame: 'general' or 'kinase'
:param hierarchy: if predictFrame is kinse: you must input the hierarchy:
group,family,subfamily,kinase to choose corresponding model
:param kinase: kinase name
:return:
a file with the score
'''
print("----------------------------")
print("in predict_for_deepphos")
win1 = 51
win2 = 33
win3 = 15
from methods.dataprocess_predict import getMatrixInputFromJson
print('running X_test1 ----------------------------')
[X_test1, y_test, ids, position, full_names,
names] = getMatrixInputFromJson(input, organism, win1)
print('running X_test2 ----------------------------')
[X_test2, _, _, _, _, _] = getMatrixInputFromJson(input, organism, win2)
print('running X_test3 ----------------------------')
[X_test3, _, _, _, _, _] = getMatrixInputFromJson(input, organism, win3)
print('----------------------------')
result_json = {}
# print X_test1.shape
# print len(position)
from methods.model_n import model_net
model = model_net(X_test1, X_test2, X_test3, y_test, nb_epoch=0)
#load model weight
# if predictFrame == 'general':
#
# if site == ('S','T'):
# outputfile = 'general_S_T'
# model_weight = './models/model_general_S,T.h5'
# if site == 'Y':
# outputfile = 'general_Y'
# model_weight = './models/model_general_Y.h5'
#for kin in ['family_CDK']:
for kin in [
'family_CDK', 'family_CK2', 'family_MAPK', 'family_PKC',
'family_Src', 'group_AGC', 'group_Atypical', 'group_CAMK',
'group_CMGC', 'group_TK', 'kinase_CDC2', 'kinase_CK2a1',
'kinase_PKACa', 'kinase_PKCa', 'kinase_SRC', 'subfamily_CDC2',
'subfamily_CDK2', 'subfamily_ERK1', 'subfamily_PKCa'
]:
# if predictFrame == 'kinase':
# outputfile = 'kinase_{:s}_{:s}'.format(hierarchy, kinase)
# model_weight = './models/model_{:s}_{:s}.h5'.format(hierarchy, kinase)
print(kin)
outputfile = 'kinase_{:s}'.format(kin)
model_weight = './models/model_{:s}.h5'.format(kin)
print("-----------------------------------------------------")
print(model_weight)
try:
model.load_weights(model_weight)
predictions_t = model.predict([X_test1, X_test2, X_test3])
results_ST = np.column_stack(
(ids, names, full_names, position, predictions_t[:, 1]))
result = pd.DataFrame(
results_ST,
columns=['target', 'name', 'full_name', 'position', 'score'])
print(result.to_json(orient='records'))
result_json[kin] = result.to_json(orient='records')
#print(result_json)
# result.to_csv(outputfile + "_api_prediction_phosphorylation.txt", index=False, header=None, sep='\t',
# quoting=csv.QUOTE_NONNUMERIC)
except:
print("there was an error")
result_json[kin] = json.dumps([{
'target': "",
'full_name': "",
'position': "",
'score': ""
}])
print(result_json)
return json.dumps(result_json)