def output(self, conformer):
"""Writes a Conformer.
Args:
conformer: dataset_pb2.Conformer
"""
matches = topology_from_geom.bond_topologies_from_geom(
bond_lengths=self._geometry_data.bond_lengths,
conformer_id=conformer.conformer_id,
fate=conformer.fate,
bond_topology=conformer.bond_topologies[0],
geometry=conformer.optimized_geometry,
matching_parameters=self._matching_parameters)
if not matches.bond_topology:
logging.error('No bond topology matched for %s',
conformer.conformer_id)
else:
del conformer.bond_topologies[:]
conformer.bond_topologies.extend(matches.bond_topology)
for bt in conformer.bond_topologies:
try:
bt.bond_topology_id = self._geometry_data.smiles_id_dict[
bt.smiles]
except KeyError:
logging.error(
'Did not find bond topology id for smiles %s',
bt.smiles)
self._wrapped_outputter.output(conformer)
def _add_alternative_bond_topologies(self, conformer, smiles_id_dict):
beam.metrics.Metrics.counter(_METRICS_NAMESPACE,
'attempted_topology_matches').inc()
matching_parameters = smu_molecule.MatchingParameters()
matching_parameters.must_match_all_bonds = True
matching_parameters.smiles_with_h = False
matching_parameters.smiles_with_labels = False
matching_parameters.neutral_forms_during_bond_matching = True
matching_parameters.consider_not_bonded = True
matching_parameters.ring_atom_count_cannot_decrease = False
matches = topology_from_geom.bond_topologies_from_geom(
bond_lengths=self._cached_bond_lengths,
conformer_id=conformer.conformer_id,
fate=conformer.fate,
bond_topology=conformer.bond_topologies[0],
geometry=conformer.optimized_geometry,
matching_parameters=matching_parameters)
if not matches.bond_topology:
beam.metrics.Metrics.counter(_METRICS_NAMESPACE,
'no_topology_matches').inc()
return
del conformer.bond_topologies[:]
conformer.bond_topologies.extend(matches.bond_topology)
for bt in conformer.bond_topologies:
try:
bt.bond_topology_id = smiles_id_dict[bt.smiles]
except KeyError:
beam.metrics.Metrics.counter(
_METRICS_NAMESPACE, 'topology_match_smiles_failure').inc()
def test_scores(self):
carbon = dataset_pb2.BondTopology.AtomType.ATOM_C
single_bond = dataset_pb2.BondTopology.BondType.BOND_SINGLE
double_bond = dataset_pb2.BondTopology.BondType.BOND_DOUBLE
# For testing, turn off the need for complete matching.
smu_molecule.default_must_match_all_bonds = False
all_distributions = bond_length_distribution.AllAtomPairLengthDistributions(
)
x, y = triangular_distribution(1.0, 1.4, 2.0)
df = pd.DataFrame({"length": x, "count": y})
bldc1c = bond_length_distribution.EmpiricalLengthDistribution(df, 0.0)
all_distributions.add(carbon, carbon, single_bond, bldc1c)
x, y = triangular_distribution(1.0, 1.5, 2.0)
df = pd.DataFrame({"length": x, "count": y})
bldc2c = bond_length_distribution.EmpiricalLengthDistribution(df, 0.0)
all_distributions.add(carbon, carbon, double_bond, bldc2c)
bond_topology = text_format.Parse(
"""
atoms: ATOM_C
atoms: ATOM_C
bonds: {
atom_a: 0
atom_b: 1
bond_type: BOND_SINGLE
}
""", dataset_pb2.BondTopology())
geometry = text_format.Parse(
"""
atom_positions {
x: 0.0
y: 0.0
z: 0.0
},
atom_positions {
x: 0.0
y: 0.0
z: 0.0
}
""", dataset_pb2.Geometry())
geometry.atom_positions[1].x = 1.4 / smu_utils_lib.BOHR_TO_ANGSTROMS
matching_parameters = smu_molecule.MatchingParameters()
matching_parameters.must_match_all_bonds = False
result = topology_from_geom.bond_topologies_from_geom(
all_distributions, bond_topology, geometry, matching_parameters)
self.assertIsNotNone(result)
self.assertEqual(len(result.bond_topology), 2)
self.assertEqual(len(result.bond_topology[0].bonds), 1)
self.assertEqual(len(result.bond_topology[1].bonds), 1)
self.assertGreater(result.bond_topology[0].score,
result.bond_topology[1].score)
self.assertEqual(result.bond_topology[0].bonds[0].bond_type,
single_bond)
self.assertEqual(result.bond_topology[1].bonds[0].bond_type,
double_bond)
def output(self, molecule):
"""Writes a Molecule.
Args:
molecule: dataset_pb2.Molecule
"""
matches = topology_from_geom.bond_topologies_from_geom(
molecule,
bond_lengths=self._geometry_data.bond_lengths,
matching_parameters=self._matching_parameters)
if not matches.bond_topology:
logging.error('No bond topology matched for %s',
molecule.molecule_id)
else:
del molecule.bond_topologies[:]
molecule.bond_topologies.extend(matches.bond_topology)
for bt in molecule.bond_topologies:
bt.source = dataset_pb2.BondTopology.SOURCE_CUSTOM
try:
bt.bond_topology_id = self._db.find_bond_topology_id_for_smiles(
bt.smiles)
except KeyError:
logging.error(
'Did not find bond topology id for smiles %s',
bt.smiles)
self._wrapped_outputter.output(molecule)
def test_multi_topology_detection(self):
"""Tests that we can find multiple versions of the same topology."""
single = dataset_pb2.BondTopology.BondType.BOND_SINGLE
double = dataset_pb2.BondTopology.BondType.BOND_DOUBLE
all_dist = bond_length_distribution.AllAtomPairLengthDistributions()
all_dist.add(dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N, single,
triangular_distribution(1.0, 1.5, 2.0))
all_dist.add(dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N, double,
triangular_distribution(1.0, 1.4, 2.0))
# This molecule is a flat aromatic square of nitrogens. The single and
# double bonds can be rotated such that it's the same topology but
# individual bonds have switched single/double.
# We set it so the bond lengths favor one of the two arrangements
molecule = dataset_pb2.Molecule(molecule_id=123)
molecule.properties.errors.fate = dataset_pb2.Properties.FATE_SUCCESS
molecule.bond_topologies.add(bond_topology_id=123, smiles='N1=NN=N1')
molecule.bond_topologies[0].atoms.extend([
dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N,
])
molecule.bond_topologies[0].bonds.extend([
dataset_pb2.BondTopology.Bond(atom_a=0, atom_b=1, bond_type=single),
dataset_pb2.BondTopology.Bond(atom_a=1, atom_b=2, bond_type=double),
dataset_pb2.BondTopology.Bond(atom_a=2, atom_b=3, bond_type=single),
dataset_pb2.BondTopology.Bond(atom_a=3, atom_b=0, bond_type=double),
])
dist15a = 1.5 / smu_utils_lib.BOHR_TO_ANGSTROMS
dist14a = 1.4 / smu_utils_lib.BOHR_TO_ANGSTROMS
molecule.optimized_geometry.atom_positions.extend([
dataset_pb2.Geometry.AtomPos(x=0, y=0, z=0),
dataset_pb2.Geometry.AtomPos(x=0, y=dist15a, z=0),
dataset_pb2.Geometry.AtomPos(x=dist14a, y=dist15a, z=0),
dataset_pb2.Geometry.AtomPos(x=dist14a, y=0, z=0),
])
matching_parameters = topology_molecule.MatchingParameters()
result = topology_from_geom.bond_topologies_from_geom(
molecule, all_dist, matching_parameters)
self.assertLen(result.bond_topology, 2)
first = result.bond_topology[0]
self.assertEqual(smu_utils_lib.get_bond_type(first, 0, 1), single)
self.assertEqual(smu_utils_lib.get_bond_type(first, 1, 2), double)
self.assertEqual(smu_utils_lib.get_bond_type(first, 2, 3), single)
self.assertEqual(smu_utils_lib.get_bond_type(first, 3, 0), double)
second = result.bond_topology[1]
self.assertEqual(smu_utils_lib.get_bond_type(second, 0, 1), double)
self.assertEqual(smu_utils_lib.get_bond_type(second, 1, 2), single)
self.assertEqual(smu_utils_lib.get_bond_type(second, 2, 3), double)
self.assertEqual(smu_utils_lib.get_bond_type(second, 3, 0), single)
def find_by_topology(
self,
smiles,
bond_lengths,
matching_parameters=topology_molecule.MatchingParameters()):
"""Find all molecules which have a detected bond topology.
Note that this *redoes* the detection. If you want the default detected
versions, you can just query by SMILES string. This is only useful if you
adjust the distance thresholds for what a matching bond is.
To adjust those, you probably want to use
AllAtomPairLengthDistributions.add_from_string_spec
Args:
smiles: smiles string for the target bond topology
bond_lengths: AllAtomPairLengthDistributions
matching_parameters: controls the algorithm for matching topologies.
Generally should not need to be modified.
Yields:
dataset_pb2.Molecule
"""
query_bt = smu_utils_lib.rdkit_molecule_to_bond_topology(
smu_utils_lib.smiles_to_rdkit_molecule(smiles))
expanded_stoich = smu_utils_lib.expanded_stoichiometry_from_topology(
query_bt)
cnt_matched_molecule = 0
cnt_molecule = 0
logging.info('Starting query for %s with stoich %s', smiles,
expanded_stoich)
for molecule in self.find_by_expanded_stoichiometry_list(
[expanded_stoich]):
if not smu_utils_lib.molecule_eligible_for_topology_detection(
molecule):
continue
cnt_molecule += 1
matches = topology_from_geom.bond_topologies_from_geom(
molecule,
bond_lengths=bond_lengths,
matching_parameters=matching_parameters)
if smiles in [bt.smiles for bt in matches.bond_topology]:
cnt_matched_molecule += 1
del molecule.bond_topologies[:]
molecule.bond_topologies.extend(matches.bond_topology)
for bt in molecule.bond_topologies:
try:
bt.source = dataset_pb2.BondTopology.SOURCE_CUSTOM
bt.bond_topology_id = self.find_bond_topology_id_for_smiles(
bt.smiles)
except KeyError:
logging.error(
'Did not find bond topology id for smiles %s',
bt.smiles)
yield molecule
logging.info('Topology query for %s matched %d / %d', smiles,
cnt_matched_molecule, cnt_molecule)
def topology_query(db, smiles):
"""Find all conformers which have a detected bond topology.
Note that this *redoes* the detection. If you want to use the default detected
versions, you can just query by SMILES string. This is only useful if you
adjust the distance thresholds for what a matching bond is.
Args:
db: smu_sqlite.SMUSQLite
smiles: smiles string for the target bond topology
Yields:
dataset_pb2.Conformer
"""
mol = Chem.MolFromSmiles(smiles, sanitize=False)
Chem.SanitizeMol(mol, Chem.rdmolops.SanitizeFlags.SANITIZE_ADJUSTHS)
mol = Chem.AddHs(mol)
query_bt = utilities.molecule_to_bond_topology(mol)
expanded_stoich = smu_utils_lib.expanded_stoichiometry_from_topology(
query_bt)
matching_parameters = _get_geometry_matching_parameters()
geometry_data = GeometryData.get_singleton()
cnt_matched_conformer = 0
cnt_conformer = 0
logging.info('Starting query for %s with stoich %s', smiles,
expanded_stoich)
for conformer in db.find_by_expanded_stoichiometry(expanded_stoich):
if not smu_utils_lib.conformer_eligible_for_topology_detection(
conformer):
continue
cnt_conformer += 1
matches = topology_from_geom.bond_topologies_from_geom(
bond_lengths=geometry_data.bond_lengths,
conformer_id=conformer.conformer_id,
fate=conformer.fate,
bond_topology=conformer.bond_topologies[0],
geometry=conformer.optimized_geometry,
matching_parameters=matching_parameters)
if smiles in [bt.smiles for bt in matches.bond_topology]:
cnt_matched_conformer += 1
del conformer.bond_topologies[:]
conformer.bond_topologies.extend(matches.bond_topology)
for bt in conformer.bond_topologies:
try:
bt.bond_topology_id = geometry_data.smiles_id_dict[
bt.smiles]
except KeyError:
logging.error(
'Did not find bond topology id for smiles %s',
bt.smiles)
yield conformer
logging.info('Topology query for %s matched %d / %d', smiles,
cnt_matched_conformer, cnt_conformer)
def process(self, molecule):
"""Called by Beam.
Returns a TopologyMatches for the plausible BondTopology's in `molecule`.
Args:
molecule:
Yields:
dataset_pb2.TopologyMatches
"""
# Adjust as needed...
# if molecule.properties.errors.fate != dataset_pb2.Properties.FATE_SUCCESS:
# return
matching_parameters = topology_molecule.MatchingParameters()
matching_parameters.neutral_forms_during_bond_matching = True
matching_parameters.must_match_all_bonds = True
matching_parameters.consider_not_bonded = True
matching_parameters.ring_atom_count_cannot_decrease = False
yield topology_from_geom.bond_topologies_from_geom(
molecule, self._bond_lengths, matching_parameters)
def process(self, conformer):
"""Called by Beam.
Returns a TopologyMatches for the plausible BondTopology's in `conformer`.
Args:
conformer:
Yields:
dataset_pb2.TopologyMatches
"""
# Adjust as needed...
# if conformer.fate != dataset_pb2.Conformer.FATE_SUCCESS:
# return
matching_parameters = smu_molecule.MatchingParameters()
matching_parameters.neutral_forms_during_bond_matching = True
matching_parameters.must_match_all_bonds = True
matching_parameters.consider_not_bonded = True
matching_parameters.ring_atom_count_cannot_decrease = False
yield topology_from_geom.bond_topologies_from_geom(
self._bond_lengths, conformer.conformer_id, conformer.fate,
conformer.bond_topologies[0], conformer.optimized_geometry,
matching_parameters)
def test_multi_topology_detection(self):
"""Tests that we can find multiple versions of the same topology."""
single = dataset_pb2.BondTopology.BondType.BOND_SINGLE
double = dataset_pb2.BondTopology.BondType.BOND_DOUBLE
all_dist = bond_length_distribution.AllAtomPairLengthDistributions()
for bond_type in [single, double]:
all_dist.add(
dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N, bond_type,
bond_length_distribution.FixedWindowLengthDistribution(
1.0, 2.0, None))
# This conformer is a flat aromatic square of nitrogens. The single and
# double bonds can be rotated such that it's the same topology but
# individual bonds have switched single/double.
conformer = dataset_pb2.Conformer()
conformer.bond_topologies.add(bond_topology_id=123, smiles="N1=NN=N1")
conformer.bond_topologies[0].atoms.extend([
dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N,
dataset_pb2.BondTopology.ATOM_N,
])
conformer.bond_topologies[0].bonds.extend([
dataset_pb2.BondTopology.Bond(atom_a=0, atom_b=1,
bond_type=single),
dataset_pb2.BondTopology.Bond(atom_a=1, atom_b=2,
bond_type=double),
dataset_pb2.BondTopology.Bond(atom_a=2, atom_b=3,
bond_type=single),
dataset_pb2.BondTopology.Bond(atom_a=3, atom_b=0,
bond_type=double),
])
dist15a = 1.5 / smu_utils_lib.BOHR_TO_ANGSTROMS
conformer.optimized_geometry.atom_positions.extend([
dataset_pb2.Geometry.AtomPos(x=0, y=0, z=0),
dataset_pb2.Geometry.AtomPos(x=0, y=dist15a, z=0),
dataset_pb2.Geometry.AtomPos(x=dist15a, y=dist15a, z=0),
dataset_pb2.Geometry.AtomPos(x=dist15a, y=0, z=0),
])
matching_parameters = smu_molecule.MatchingParameters()
result = topology_from_geom.bond_topologies_from_geom(
bond_lengths=all_dist,
conformer_id=123,
fate=dataset_pb2.Conformer.FATE_SUCCESS,
bond_topology=conformer.bond_topologies[0],
geometry=conformer.optimized_geometry,
matching_parameters=matching_parameters)
self.assertLen(result.bond_topology, 2)
# The returned order is arbitrary so we figure out which is is marked
# as the starting topology.
starting_idx = min([
i for i, bt, in enumerate(result.bond_topology)
if bt.is_starting_topology
])
other_idx = (starting_idx + 1) % 2
starting = result.bond_topology[starting_idx]
self.assertTrue(starting.is_starting_topology)
self.assertEqual(smu_utils_lib.get_bond_type(starting, 0, 1), single)
self.assertEqual(smu_utils_lib.get_bond_type(starting, 1, 2), double)
self.assertEqual(smu_utils_lib.get_bond_type(starting, 2, 3), single)
self.assertEqual(smu_utils_lib.get_bond_type(starting, 3, 0), double)
other = result.bond_topology[other_idx]
self.assertFalse(other.is_starting_topology)
self.assertEqual(smu_utils_lib.get_bond_type(other, 0, 1), double)
self.assertEqual(smu_utils_lib.get_bond_type(other, 1, 2), single)
self.assertEqual(smu_utils_lib.get_bond_type(other, 2, 3), double)
self.assertEqual(smu_utils_lib.get_bond_type(other, 3, 0), single)
def test_scores(self):
carbon = dataset_pb2.BondTopology.ATOM_C
single_bond = dataset_pb2.BondTopology.BondType.BOND_SINGLE
double_bond = dataset_pb2.BondTopology.BondType.BOND_DOUBLE
# For testing, turn off the need for complete matching.
topology_molecule.default_must_match_all_bonds = False
all_distributions = bond_length_distribution.AllAtomPairLengthDistributions(
)
bldc1c = triangular_distribution(1.0, 1.4, 2.0)
all_distributions.add(carbon, carbon, single_bond, bldc1c)
bldc2c = triangular_distribution(1.0, 1.5, 2.0)
all_distributions.add(carbon, carbon, double_bond, bldc2c)
molecule = dataset_pb2.Molecule()
molecule.bond_topologies.append(
text_format.Parse(
"""
atoms: ATOM_C
atoms: ATOM_C
bonds: {
atom_a: 0
atom_b: 1
bond_type: BOND_SINGLE
}
""", dataset_pb2.BondTopology()))
molecule.optimized_geometry.MergeFrom(
text_format.Parse(
"""
atom_positions {
x: 0.0
y: 0.0
z: 0.0
},
atom_positions {
x: 0.0
y: 0.0
z: 0.0
}
""", dataset_pb2.Geometry()))
molecule.optimized_geometry.atom_positions[1].x = (
1.4 / smu_utils_lib.BOHR_TO_ANGSTROMS)
matching_parameters = topology_molecule.MatchingParameters()
matching_parameters.must_match_all_bonds = False
molecule.properties.errors.fate = dataset_pb2.Properties.FATE_SUCCESS
molecule.molecule_id = 1001
result = topology_from_geom.bond_topologies_from_geom(
molecule, all_distributions, matching_parameters)
self.assertIsNotNone(result)
self.assertLen(result.bond_topology, 2)
self.assertLen(result.bond_topology[0].bonds, 1)
self.assertLen(result.bond_topology[1].bonds, 1)
self.assertEqual(result.bond_topology[0].bonds[0].bond_type, single_bond)
self.assertEqual(result.bond_topology[1].bonds[0].bond_type, double_bond)
self.assertGreater(result.bond_topology[0].topology_score,
result.bond_topology[1].topology_score)
self.assertAlmostEqual(
np.sum(np.exp([bt.topology_score for bt in result.bond_topology])), 1.0)
self.assertAlmostEqual(result.bond_topology[0].geometry_score,
np.log(bldc1c.pdf(1.4)))
self.assertAlmostEqual(result.bond_topology[1].geometry_score,
np.log(bldc2c.pdf(1.4)))