def filter_with_threshold(self, reactant_smiles, target, threshold):
"""Filters reactions based on a score threshold.
Args:
reactant_smiles (str): SMILES string of reactants.
target (str): SMILES string of target product.
threshold (float): Value scores must be above to pass filter.
Returns:
2-tuple of (np.ndarray of np.ndarray of np.bool, float): Whether the
reaction passed the filer and the score of the reaction.
"""
[pfp, rfp] = create_rxn_Morgan2FP_separately(reactant_smiles,
target,
rxnfpsize=2048,
pfpsize=2048,
useFeatures=False)
pfp = np.asarray(pfp, dtype='float32')
rfp = np.asarray(rfp, dtype='float32')
rxnfp = pfp - rfp
score = self.model.predict(
[pfp.reshape(1, 2048),
rxnfp.reshape(1, 2048)])
filter_flag = (score > threshold)
return filter_flag, float(score)
def transform_input(self,
reactant_smiles,
target,
rxnfpsize=2048,
pfpsize=2048,
useFeatures=False):
"""Transforms the input for the API model from SMILES strings to product and reaction fingerprints
Args:
reactant_smiles (str): SMILES string of the reactants
target (str): SMILES string of the target
rxnfpsize (int): Length of desired fingerprint for the reaction. Must agree with model input shape
pfpsize (int): Length of desired fingerprint for the product. Must agree with model input shape
useFeatures (bool): Flag to use features or not when generating fingerprint. Should agree with how model was trained
Returns:
list of dict: Input fingerprints, formatted for a call to the tensorflow API model
"""
pfp, rfp = create_rxn_Morgan2FP_separately(reactant_smiles,
target,
rxnfpsize=rxnfpsize,
pfpsize=pfpsize,
useFeatures=useFeatures)
pfp = np.asarray(pfp, dtype='float32')
rfp = np.asarray(rfp, dtype='float32')
rxnfp = pfp - rfp
return [{'input_1': pfp.tolist(), 'input_2': rxnfp.tolist()}]
def get_n_conditions(self,
rxn,
n=10,
singleSlvt=True,
with_smiles=True,
return_scores=False):
"""
Reaction condition recommendations for a rxn (SMILES) from top n NN
Returns the top n parseable conditions.
"""
# print('started neuralnet')
self.singleSlvt = singleSlvt
self.with_smiles = with_smiles
# print('rxn to recommend context for : {}'.format(rxn), contextRecommender_loc)
try:
rsmi = rxn.split('>>')[0]
psmi = rxn.split('>>')[1]
rct_mol = Chem.MolFromSmiles(rsmi)
prd_mol = Chem.MolFromSmiles(psmi)
[atom.ClearProp('molAtomMapNumber') for \
atom in rct_mol.GetAtoms() if atom.HasProp('molAtomMapNumber')]
[atom.ClearProp('molAtomMapNumber') for \
atom in prd_mol.GetAtoms() if atom.HasProp('molAtomMapNumber')]
rsmi = Chem.MolToSmiles(rct_mol, isomericSmiles=True)
psmi = Chem.MolToSmiles(prd_mol, isomericSmiles=True)
[pfp,
rfp] = fp.create_rxn_Morgan2FP_separately(rsmi,
psmi,
rxnfpsize=self.fp_size,
pfpsize=self.fp_size,
useFeatures=False,
calculate_rfp=True,
useChirality=True)
pfp = pfp.reshape(1, self.fp_size)
rfp = rfp.reshape(1, self.fp_size)
rxnfp = pfp - rfp
c1_input = []
r1_input = []
r2_input = []
s1_input = []
s2_input = []
inputs = [
pfp, rxnfp, c1_input, r1_input, r2_input, s1_input, s2_input
]
(top_combos,
top_combo_scores) = self.predict_top_combos(inputs=inputs)
if return_scores:
return (top_combos[:n], top_combo_scores[:n])
else:
return top_combos[:n]
except Exception as e:
# MyLogger.print_and_log('Failed for reaction {} because {}. Returning None.'.format(
# rxn, e), contextRecommender_loc, level=2)
print(('Failed for reaction {} because {}. Returning None.'.format(
rxn, e), contextRecommender_loc))
return [[]]
def evaluate(self, reactant_smiles, target, **kwargs):
# Strip chirality
# rmol = Chem.MolFromSmiles(reactant_smiles)
# pmol = Chem.MolFromSmiles(target)
# reactant_smiles = Chem.MolToSmiles(rmol, False)
# target = Chem.MolToSmiles(pmol, False)
[pfp, rfp] = create_rxn_Morgan2FP_separately(reactant_smiles,
target,
rxnfpsize=2048,
pfpsize=2048,
useFeatures=False)
pfp = np.asarray(pfp, dtype='float32')
rfp = np.asarray(rfp, dtype='float32')
rxnfp = pfp - rfp
score = self.model.predict(
[pfp.reshape(1, 2048),
rxnfp.reshape(1, 2048)])
outcome = {'smiles': target, 'template_ids': [], 'num_examples': 0}
all_outcomes = []
all_outcomes.append([{
'rank': 1.0,
'outcome': outcome,
'score': float(score[0][0]),
'prob': float(score[0][0]),
}])
return all_outcomes
def path_condition(self, n, path):
"""Reaction condition recommendation for a reaction path with multiple reactions
path: a list of reaction SMILES for each step
return: a list of reaction context with n options for each step
"""
rsmi_list = []
psmi_list = []
contexts = []
for rxn in path:
try:
rsmi = rxn.split('>>')[0]
psmi = rxn.split('>>')[1]
rct_mol = Chem.MolFromSmiles(rsmi)
prd_mol = Chem.MolFromSmiles(psmi)
[atom.ClearProp('molAtomMapNumber')for \
atom in rct_mol.GetAtoms() if atom.HasProp('molAtomMapNumber')]
[atom.ClearProp('molAtomMapNumber')for \
atom in prd_mol.GetAtoms() if atom.HasProp('molAtomMapNumber')]
rsmi = Chem.MolToSmiles(rct_mol, isomericSmiles=True)
psmi = Chem.MolToSmiles(prd_mol, isomericSmiles=True)
[pfp, rfp
] = fp.create_rxn_Morgan2FP_separately(rsmi,
psmi,
rxnfpsize=self.fp_size,
pfpsize=self.fp_size,
useFeatures=False,
calculate_rfp=True,
useChirality=True)
pfp = pfp.reshape(1, self.fp_size)
rfp = rfp.reshape(1, self.fp_size)
rxnfp = pfp - rfp
c1_input = []
r1_input = []
r2_input = []
s1_input = []
s2_input = []
inputs = [
pfp, rxnfp, c1_input, r1_input, r2_input, s1_input,
s2_input
]
top_combos = self.predict_top_combos(inputs=inputs,
c1_rank_thres=1,
s1_rank_thres=3,
s2_rank_thres=1,
r1_rank_thres=4,
r2_rank_thres=1)
contexts.append(top_combos[:n])
except Exception as e:
MyLogger.print_and_log(
'Failed for reaction {} because {}. Returning None.'.
format(rxn, e),
contextRecommender_loc,
level=2)
return contexts
def filter_with_threshold(self, reactant_smiles, target, threshold):
[pfp, rfp] = create_rxn_Morgan2FP_separately(reactant_smiles,
target,
rxnfpsize=2048,
pfpsize=2048,
useFeatures=False)
pfp = np.asarray(pfp, dtype='float32')
rfp = np.asarray(rfp, dtype='float32')
rxnfp = pfp - rfp
score = self.model.predict(
[pfp.reshape(1, 2048),
rxnfp.reshape(1, 2048)])
filter_flag = (score > threshold)
return filter_flag, float(score)
def evaluate(self, reactant_smiles, target, **kwargs):
"""Evaluates likelihood of given reaction.
Args:
reactant_smiles (str): SMILES string of reactants.
target (str): SMILES string of target product.
**kwargs: Unused.
Returns:
A list of reaction outcomes.
"""
# Strip chirality
# rmol = Chem.MolFromSmiles(reactant_smiles)
# pmol = Chem.MolFromSmiles(target)
# reactant_smiles = Chem.MolToSmiles(rmol, False)
# target = Chem.MolToSmiles(pmol, False)
[pfp, rfp] = create_rxn_Morgan2FP_separately(reactant_smiles,
target,
rxnfpsize=2048,
pfpsize=2048,
useFeatures=False)
pfp = np.asarray(pfp, dtype='float32')
rfp = np.asarray(rfp, dtype='float32')
rxnfp = pfp - rfp
score = self.model.predict(
[pfp.reshape(1, 2048),
rxnfp.reshape(1, 2048)])
outcome = {'smiles': target, 'template_ids': [], 'num_examples': 0}
all_outcomes = []
all_outcomes.append([{
'rank': 1.0,
'outcome': outcome,
'score': float(score[0][0]),
'prob': float(score[0][0]),
}])
return all_outcomes
