def run(self):
logger.info('MakeModel {}nM: Init'.format(self.value))
data = pd.read_csv(self.input().path)
# get unique molecules and its smiles
mols = data[['molregno', 'canonical_smiles']]
mols = mols.drop_duplicates('molregno')
mols = mols.set_index('molregno')
mols = mols.sort_index()
# group targets by molregno
targets = data[['molregno', 'target_chembl_id']]
#targets = targets.sort_index(by='molregno')
targets = targets.sort_values(by='molregno')
targets = targets.groupby('molregno').apply(
lambda x: ','.join(x.target_chembl_id))
targets = targets.apply(lambda x: x.split(','))
targets = pd.DataFrame(targets, columns=['targets'])
# merge it
mols['ROMol'] = mols.apply(
lambda x: Chem.MolFromSmiles(x['canonical_smiles']), axis=1)
dataset = pd.merge(mols, targets, left_index=True, right_index=True)
dataset = dataset.ix[dataset['ROMol'].notnull()]
# generate fingerprints
dataset['FP'] = dataset.apply(lambda row: computeFP(row['ROMol']),
axis=1)
dataset = dataset.ix[dataset['FP'].notnull()]
logger.info('MakeModel {}nM: Data ready'.format(self.value))
# generate models training data
X = [f.fp for f in dataset['FP']]
mlb = MultiLabelBinarizer()
y = mlb.fit_transform(dataset['targets'])
# train the model
morgan_bnb = OneVsRestClassifier(MultinomialNB())
morgan_bnb.fit(X, y)
morgan_bnb.targets = mlb.classes_
# save the model
joblib.dump(morgan_bnb, self.output().path)
logger.info('MakeModel {}nM: Done'.format(self.value))
print 'fps done' from sklearn.naive_bayes import BernoulliNB, MultinomialNB from sklearn.multiclass import OneVsRestClassifier from sklearn.externals import joblib from sklearn.preprocessing import MultiLabelBinarizer X = [f.fp for f in dataset['FP']] yy = [c for c in dataset['targets']] ##print dataset['targets'].head() mlb = MultiLabelBinarizer() y = mlb.fit_transform(yy) ## this is for newer versions of sklearn morgan_bnb = OneVsRestClassifier(MultinomialNB()) print 'model building' morgan_bnb.fit(X, y) morgan_bnb.targets = mlb.classes_ print morgan_bnb.multilabel_ print morgan_bnb.targets joblib.dump(morgan_bnb, '../chembl_22/models/1uM/mNB_1uM_all.pkl') print 'done!'
