def test_isomeric_canonicalisation():
endiandric_acid = r'OC(=O)[C@H]5C2\C=C/C3[C@@H]5CC4[C@H](C\C=C\C=C\c1ccccc1)[C@@H]2[C@@H]34'
with_stereocenters = canonicalize(endiandric_acid, include_stereocenters=True)
without_stereocenters = canonicalize(endiandric_acid, include_stereocenters=False)
expected_with_stereocenters = 'O=C(O)[C@H]1C2C=CC3[C@@H]1CC1[C@H](C/C=C/C=C/c4ccccc4)[C@@H]2[C@@H]31'
expected_without_stereocenters = 'O=C(O)C1C2C=CC3C1CC1C(CC=CC=Cc4ccccc4)C2C31'
assert with_stereocenters == expected_with_stereocenters
assert without_stereocenters == expected_without_stereocenters
def process_smis(smis,
scoring_function,
pool,
canonicalization,
duplicate_removal,
scoring_parallelization,
max_smi_len=100):
if canonicalization:
smis = pool(
delayed(lambda smi: canonicalize(smi, include_stereocenters=False))
(smi) for smi in smis)
smis = list(
filter(lambda smi: (smi is not None) and (len(smi) < max_smi_len),
smis))
if duplicate_removal:
smis = list(set(smis))
if scoring_function is None:
return smis
if scoring_parallelization:
scores = pool(delayed(scoring_function)(smi) for smi in smis)
else:
scores = [scoring_function(smi) for smi in smis]
smis, scores = filter_by_score(smis, scores, -1e-8)
return smis, scores
def vec2smiles(self, vec, rem_bos, rem_eos):
string = self.vec2string(vec, rem_bos, rem_eos)
if string is None:
return None
smiles = canonicalize(string)
if smiles is None or len(smiles) == 0:
return None
if len(smiles) > self.max_smiles_length:
return None
return smiles
def test_list_canonicalization_removes_none():
m1 = 'CCC(OCOCO)CC(=O)NCC'
m2 = 'this.is.not.a.molecule'
m3 = 'c1ccccc1'
m4 = 'CC(OCON=N)CC'
molecules = [m1, m2, m3, m4]
canonicalized_molecules = canonicalize_list(molecules)
valid_molecules = [m1, m3, m4]
expected = [canonicalize(smiles) for smiles in valid_molecules]
assert canonicalized_molecules == expected
def canonicalize_and_score_smiles(self, smis, scoring_function, pool):
smis = pool(delayed(lambda smi: canonicalize(smi, include_stereocenters=False))(smi) for smi in smis)
smis = list(filter(lambda smi: (smi is not None) and self.char_dict.allowed(smi), smis))
scores = pool(delayed(scoring_function.score)(smi) for smi in smis)
#scores = [0.0 for smi in smis]
filtered_smis_and_scores = list(
filter(
lambda smi_and_score: smi_and_score[1] > scoring_function.scoring_function.corrupt_score,
zip(smis, scores),
)
)
smis, scores = map(list, zip(*filtered_smis_and_scores)) if len(filtered_smis_and_scores) > 0 else ([], [])
return smis, scores
def sample_unique_molecules(model: DistributionMatchingGenerator,
number_molecules: int,
max_tries=10) -> List[str]:
"""
Sample from the given generator until the desired number of unique (distinct) molecules
has been sampled (i.e., ignore duplicate molecules).
Args:
model: model to sample from
number_molecules: number of unique (distinct) molecules to generate
max_tries: determines the maximum number N of samples to draw, N = number_molecules * max_tries
Returns:
A list of number_molecules unique molecules, in canonalized form.
If this was not possible with the given max_tries, the list may be shorter.
The generation order is kept.
"""
max_samples = max_tries * number_molecules
number_already_sampled = 0
unique_list: List[str] = []
unique_set: Set[str] = set()
while len(unique_list
) < number_molecules and number_already_sampled < max_samples:
remaining_to_sample = number_molecules - len(unique_list)
samples = model.generate(remaining_to_sample)
number_already_sampled += remaining_to_sample
for smiles in samples:
canonical_smiles = canonicalize(smiles)
if canonical_smiles is not None and canonical_smiles not in unique_set:
unique_set.add(canonical_smiles)
unique_list.append(canonical_smiles)
# this should always be True
assert len(unique_set) == len(unique_list)
return unique_list
def mutate(p_gene, scoring_function):
c_gene = mutation(p_gene)
c_smiles = canonicalize(cfg_util.decode(gene_to_cfg(c_gene)))
c_score = scoring_function.score(c_smiles)
return Molecule(c_score, c_smiles, c_gene)
def _canonicalize(self):
if self._canonical_smiles is not None:
return
canonical = [canonicalize(mol) for mol in self._smiles]
self._canonical_smiles = [s for s in canonical if s is not None]