def test_PubChemDataSet_creation(self):
""" testing PubChemDataSet creation """
# some cids from PubChem AID 1
active_compounds = [11969872, 390525, 394646]
inactive_compounds = [1018, 4775, 219294]
ds = PubChemDataSet(1)
df = ds.get_compounds()
assert all(df.loc[active_compounds, 'Activity'] == 1)
assert all(df.loc[inactive_compounds, 'Activity'] == 0)
def test_PubChemDataSet_clean_load(self):
""" testing a clean loaded dataset """
df = PubChemDataSet(1).clean_load()
assert len((df.Activity[df.Activity == 1])) == len(
(df.Activity[df.Activity == 0]))
assert None not in [
Chem.MolToSmiles(mol) if mol else None for mol in df.rdkit
]
def build_models(aid, sub_directory):
best_scores = {}
aid_sub_directory = os.path.join(sub_directory, str(aid))
# if its already been modeled move on
if os.path.exists(aid_sub_directory):
return
else:
os.mkdir(aid_sub_directory)
try:
ds = PubChemDataSet(aid).clean_load()
ds.to_csv('{}/training.csv'.format(aid_sub_directory))
y = ds.Activity
X = PubChemDataSetDescriptors(ds).load_ECFP6()
print("=======building model for aid {0}=======".format(aid))
print("======={0} compounds: {1} active, {2} inactive=======".format(
y.shape[0], (y == 1).sum(), (y == 0).sum()))
except:
print("error on aid {0}".format(aid))
return
for name, clf in SKLearnModels.CLASSIFIERS:
pipe = Pipeline([(name, clf)])
print("=======5-fold CV on {0}=======".format(name))
parameters = SKLearnModels.PARAMETERS[name]
cv_search = GridSearchCV(pipe,
parameters,
cv=5,
scoring='accuracy',
n_jobs=1,
verbose=0)
cv_search.fit(X.values, y.values)
print("================================")
print("The best parameters for {0} are :\n{1}".format(
name, cv_search.best_params_))
print("The best score is {0}".format(cv_search.best_score_))
best_scores[name] = cv_search.best_score_
joblib.dump(cv_search.best_estimator_,
'{}/{}.pkl'.format(aid_sub_directory, name))
with open('{}/results.csv'.format(aid_sub_directory), 'w') as results_file:
for model, score in best_scores.items():
results_file.write(model + ',' + str(score) + '\n')
def test_PubChemDataSet_load(self):
""" testing PubChemDataSet load """
# some smiles from PubChem AID 1
smiles = ['COC1=CC(=C2C(=C1)OC(=CC2=O)C3(C=CC(=O)C=C3)O)O',
'C1=CC=C(C=C1)CCCC(=O)O']
df = PubChemDataSet(1).load()
for smi in smiles:
print(smi in df.SMILES.tolist())
assert smi in df.SMILES.tolist()
def setup(self):
""" sets up class with some dummy data """
self.clfs = [clf[0] for clf in SKLearnModels.CLASSIFIERS]
self.ds = PubChemDataSet(1224861).load()
mols = [Chem.MolFromSmiles(smi) for smi in self.ds.SMILES]
calc = MoleculeDescriptors.MolecularDescriptorCalculator(
[desc[0] for desc in Descriptors.descList])
self.X = pd.DataFrame(
[list(calc.CalcDescriptors(mol)) for mol in mols],
columns=list(calc.GetDescriptorNames()),
index=self.ds.index)
def setup(self):
self.ds = PubChemDataSet(1).clean_load()
self.y = self.ds.Activity
self.X = PubChemDataSetDescriptors(self.ds).load_rdkit()
#print(self.X.shape, self.y.shape)
self.y = self.y[self.X.notnull().all(1)]
self.X = self.X[self.X.notnull().all(1)]
#print(self.X.shape, self.y.shape)
self.y = self.y[~np.isinf(self.X.values).any(1)]
self.X = self.X[~np.isinf(self.X.values).any(1)]
def build_model(aid, model):
try:
ds = PubChemDataSet(aid).clean_load()
y = ds.Activity
X = PubChemDataSetDescriptors(ds).load_rdkit()
# TODO: put this into a cleaner step
# remove null values
y = y[X.notnull().all(1)]
X = X[X.notnull().all(1)]
# TODO: put this into a cleaner step
# remove null values
y = y[~np.isinf(X.values).any(1)]
X = X[~np.isinf(X.values).any(1)]
print("=======building model for aid {0}=======".format(aid))
print("======={0} compounds: {1} active, {2} inactive=======".format(
y.shape[0], (y == 1).sum(), (y == 0).sum()))
except:
raise Exception("error on aid {0}".format(aid))
pipe = Pipeline(
list(SKLearnModels.PREPROCESS) +
[(name, clf)
for name, clf in SKLearnModels.CLASSIFIERS if name == model])
print("=======5-fold CV on {0}=======".format(model))
parameters = SKLearnModels.PARAMETERS[model]
cv_search = GridSearchCV(pipe,
parameters,
cv=5,
scoring='accuracy',
n_jobs=-1,
verbose=0)
cv_search.fit(X.values, y.values)
print("================================")
print("The best parameters for {0} are :".format(model))
for param, val in cv_search.best_params_.items():
print("{}: {}".format(param.split('__')[1], val))
print("The best accuracy score is {0:.2f}%".format(cv_search.best_score_ *
100))
# Save to pickle
cv_search.best_estimator_
def main():
# assays from cluster 0
aids = [
'119', '103', '99', '133', '71', '145', '5', '33', '113', '43', '139',
'115', '55', '31', '67', '81', '143', '87', '109', '129', '39', '49',
'137', '65', '79', '93', '57', '15', '107', '59', '37', '101', '123',
'41', '45', '7', '83', '91', '53', '13', '21', '95', '105', '9', '131',
'125', '97', '29', '121', '3', '25', '141', '23', '77', '19', '1',
'47', '73', '35', '89', '85'
]
aids = list(map(int, aids))
import os
ds_test = pd.read_csv(os.getenv('QSAR_DATA') + 'reach_curated.csv')
ds_test.index = ds_test.ECNumber
best_models = {}
predictions = []
for aid in aids:
try:
ds = PubChemDataSet(aid).clean_load()
y = ds.Activity
X = PubChemDataSetDescriptors(ds).load_rdkit()
# TODO: put this into a cleaner step
# remove null values
y = y[X.notnull().all(1)]
X = X[X.notnull().all(1)]
# TODO: put this into a cleaner step
# remove null values
y = y[~np.isinf(X.values).any(1)]
X = X[~np.isinf(X.values).any(1)]
print("=======building model for aid {0}=======".format(aid))
print(
"======={0} compounds: {1} active, {2} inactive=======".format(
y.shape[0], (y == 1).sum(), (y == 0).sum()))
except:
print("error on aid {0}".format(aid))
continue
for name, clf in SKLearnModels.CLASSIFIERS:
pipe = Pipeline(list(SKLearnModels.PREPROCESS) + [(name, clf)])
print("=======5-fold CV on {0}=======".format(name))
parameters = SKLearnModels.PARAMETERS[name]
cv_search = GridSearchCV(pipe,
parameters,
cv=5,
scoring='accuracy',
n_jobs=-1,
verbose=0)
cv_search.fit(X.values, y.values)
print("================================")
print("The best parameters for {0} are :\n{1}".format(
name, cv_search.best_params_))
print("The best score is {0}".format(cv_search.best_score_))
best_models[aid] = cv_search.best_estimator_
X_test = PubChemDataSetDescriptors(ds_test).load_rdkit()
# save null or inf values
dropped_cmps = X_test[~(X_test.notnull().all(1)) |
(np.isinf(X_test.values).any(1))]
# remove null and inf values
X_test = X_test[X_test.notnull().all(1)]
X_test = X_test[~np.isinf(X_test.values).any(1)]
print("Making predictions on {0} compounds".format(X_test.shape[0]))
preds = pd.DataFrame(cv_search.predict(X_test),
index=X_test.index,
columns=[aid])
predictions.append(preds)
print(pd.concat(predictions, axis=1))
import os
filename = os.getenv('QSAR_DATA') + 'reach_predictions_cluster_0.csv'
pd.concat(predictions, axis=1).to_csv(filename)
m = pd.DataFrame(best_models, index=aids)
m.to_pickle(os.getenv('QSAR_DATA') + 'models.csv')
def main():
aids = [119, 79, 83, 7, 37, 99, 129, 59, 41]
profile = DS.profile_3.load()
best_models = {}
predictions = []
for aid in aids:
try:
ds = PubChemDataSet(aid).clean_load()
y = ds.Activity
X = PubChemDataSetDescriptors(ds).load_rdkit()
# TODO: put this into a cleaner step
# remove null values
y = y[X.notnull().all(1)]
X = X[X.notnull().all(1)]
# TODO: put this into a cleaner step
# remove null values
y = y[~np.isinf(X.values).any(1)]
X = X[~np.isinf(X.values).any(1)]
print("=======building model for aid {0}=======".format(aid))
print(
"======={0} compounds: {1} active, {2} inactive=======".format(
y.shape[0], (y == 1).sum(), (y == 0).sum()))
except:
print("error on aid {0}".format(aid))
continue
for name, clf in SKLearnModels.CLASSIFIERS:
pipe = Pipeline(list(SKLearnModels.PREPROCESS) + [(name, clf)])
print("=======5-fold CV on {0}=======".format(name))
parameters = SKLearnModels.PARAMETERS[name]
cv_search = GridSearchCV(pipe,
parameters,
cv=5,
scoring='accuracy',
n_jobs=-1,
verbose=0)
cv_search.fit(X.values, y.values)
print("================================")
print("The best parameters for {0} are :\n{1}".format(
name, cv_search.best_params_))
print("The best score is {0}".format(cv_search.best_score_))
best_models[aid] = cv_search.best_estimator_
ds_test = profile.get_subprofile([aid]).get_nulls().as_ds()
X_test = PubChemDataSetDescriptors(ds_test).load_rdkit()
# save null or inf values
dropped_cmps = X_test[~(X_test.notnull().all(1)) |
(np.isinf(X_test.values).any(1))]
# remove null and inf values
X_test = X_test[X_test.notnull().all(1)]
X_test = X_test[~np.isinf(X_test.values).any(1)]
print("Making predictions on {0} compounds".format(X_test.shape[0]))
preds = pd.DataFrame(cv_search.predict(X_test),
index=X_test.index,
columns=[aid])
predictions.append(preds)
print(pd.concat(predictions, axis=1))
import os
filename = os.getenv('QSAR_DATA') + 'missing_data_predictions.csv'
pd.concat(predictions, axis=1).to_csv(filename)
def setup(self):
""" PubChem AID 1224861 with
25 actives and 109 inactives
"""
self.ds = PubChemDataSet(1).clean_load()
def test_StructureChecker_cleaner(self):
""" testing structure checker """
ds = PubChemDataSet(1).load()
pipe = PubChemDataSetCleaner(steps=[StructureCleaner()])
ds = pipe.run(ds)
assert None not in ds.rdkit.values
