def test_builder_returns_false_when_smiles_already_present():
uut = OptimizedMolecules.Builder()
uut.append('C', col1=1)
result = uut.append('C', col2=2)
assert not result
def test_to_csv_call_with_without_columns_and_columns_raises(tmp_path):
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0, col2=1)
uut = builder.build()
with pytest.raises(ValueError):
uut.to_csv('dummy_path', columns=[], without_columns=[])
def test_top_fraction_raises_with_fraction_greater_than_one():
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0)
uut = builder.build()
with pytest.raises(ValueError):
uut.get_first_fraction(1.2, by_column='col1')
def test_first_n_raises_with_zero_n():
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0)
uut = builder.build()
with pytest.raises(ValueError):
uut.get_first_n(0, by_column='col1')
def test_first_fraction_raises_with_negative_fraction():
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0)
uut = builder.build()
with pytest.raises(ValueError):
uut.get_first_fraction(-1, by_column='col1')
def test_most_similar_tanimoto_multiple_elements_happy_path():
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0)
builder.append('Cl', col1=0)
builder.append('F', col1=0)
uut = builder.build()
expected = pd.DataFrame.from_dict(
{
'C': {
'tanimoto_similarity': 1.0,
'most_similar_smiles': 'C'
},
'F': {
'tanimoto_similarity': 1.0,
'most_similar_smiles': 'F'
},
'Cl': {
'tanimoto_similarity': 0.0,
'most_similar_smiles': 'C'
},
},
orient='index')
result = uut.most_similar_tanimoto(['C', 'F'])
assert result.equals(expected)
def test_builder_in_operator_returns_true_when_smiles_already_present():
uut = OptimizedMolecules.Builder()
uut.append('C', col2=2)
result = 'C' in uut
assert result
def test_top_fraction_raises_with_non_existing_column():
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0)
uut = builder.build()
with pytest.raises(ValueError):
uut.get_first_fraction(1, by_column='no_such_column')
def test_rmse_raises_when_col2_contains_nans():
builder = OptimizedMolecules.Builder()
builder.append('C', col1=1)
builder.append('H', col1=1, col2=2)
uut = builder.build()
with pytest.raises(ValueError):
uut.rmse('col1', 'col2')
def test_rmse_raises_when_col2_not_in_molecules():
builder = OptimizedMolecules.Builder()
builder.append('C', col=1)
builder.append('H', col=2)
uut = builder.build()
with pytest.raises(ValueError):
uut.rmse('col', 'no_such_column')
def test_builder_does_not_calculate_validity_when_total_samples_is_zero():
uut = OptimizedMolecules.Builder()
uut.append('C', col1=1)
uut.append('H', col2=2)
result = uut.build()
assert 'validity' not in result.metrics
def test_builder_happy_path():
uut = OptimizedMolecules.Builder()
uut.append('C', col1=1, col2=2)
uut.append('H', col1=3, col2=4)
expected = pd.DataFrame.from_dict({'C': {'col1': 1, 'col2': 2}, 'H': {'col1': 3, 'col2': 4}}, orient='index')
result = uut.build()
assert result.molecules.equals(expected)
def test_rmse_no_error_happy_path():
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0, col2=0)
builder.append('H', col1=1, col2=1)
uut = builder.build()
result = uut.rmse('col1', 'col2')
assert result == 0
def test_builder_calculates_validity_when_total_samples_positive():
uut = OptimizedMolecules.Builder()
uut.append('C', col1=1)
uut.append('H', col2=2)
uut.total_samples += 10
result = uut.build()
assert result.metrics['validity'] == 0.2
def test_builder_happy_path_with_missing_data():
uut = OptimizedMolecules.Builder()
uut.append('C', col1=1)
uut.append('H', col2=2)
expected = pd.DataFrame.from_dict({'C': {'col1': 1}, 'H': {'col2': 2}}, orient='index')
result = uut.build()
assert result.molecules.equals(expected)
def test_top_n_happy_path_descending():
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0)
builder.append('H', col1=1)
uut = builder.build()
expected = pd.DataFrame.from_dict({'H': {'col1': 1}}, orient='index')
result = uut.get_first_fraction(.5, by_column='col1', sort_ascending=False)
assert result.equals(expected)
def test_to_csv_happy_path(tmp_path):
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0, col2=1)
builder.append('H', col1=2.5, col2=1)
uut = builder.build()
file = tmp_path / 'tmp_file.csv'
expected = textwrap.dedent('''\
SMILES,col1,col2
C,0.0,1
H,2.5,1
''')
uut.to_csv(file)
assert file.read_text() == expected
def descent_steps(self, smiles_docking_score_fn, size):
assert smiles_docking_score_fn is not None
assert size is not None
logger.info('Descent steps start')
batch_size = 32
min_valid_steps = 5
results = []
while len(results) < size:
logger.info(f'Valid descent steps results: {len(results)} / {size}')
one_hots = self.train[np.random.choice(self.train.shape[0], batch_size, replace=False), :]
latents = self.cvae.encode(one_hots)
latent_changes = [latents]
for _ in range(self.descent_iterations):
latents += self.mlp.gradient(latents) * self.descent_lr
latent_changes.append(np.copy(latents))
decoded_series = [self.cvae.hot_to_smiles(self.cvae.decode(delta), strip=True, numpy=True)
for delta in latent_changes]
smiles_changes = [
[canonicalize(smi) for smi in smi_list]
for smi_list in decoded_series
]
for i in range(batch_size):
descent_results = OptimizedMolecules.Builder()
for j in range(self.descent_iterations + 1):
smi = smiles_changes[j][i]
if smi is not None and is_valid(smi):
try:
docking_score = smiles_docking_score_fn(smi, n_cpu=self.docking_n_cpu)
descent_results.append(
smi,
latent_vector=latent_changes[j][i],
step=j,
**docking_score
)
except (ValueError, RuntimeError, TypeError):
logger.error('Docking failed for %s', smi)
if descent_results.size > min_valid_steps:
results.append(descent_results.build())
return results
def test_to_csv_with_without_columns(tmp_path):
builder = OptimizedMolecules.Builder()
builder.append('C', col1=0, col2=1)
builder.append('H', col1=2.5, col2=1)
uut = builder.build()
file = tmp_path / 'tmp_file.csv'
expected = textwrap.dedent('''\
SMILES,col1
C,0.0
H,2.5
''')
uut.to_csv(file, without_columns=['col2'])
assert file.read_text() == expected
def generate_optimized_molecules(self, number_molecules, smiles_docking_score_fn):
results_builder = OptimizedMolecules.Builder()
while results_builder.size < number_molecules:
logger.info(f'Generated {results_builder.size} / {number_molecules}')
latents = np.random.normal(size=(self.batch_size, self.latent))
before = [self.mlp.latent_score(latents[i].reshape(1, -1))
for i in range(self.batch_size)]
for _ in range(self.descent_iterations):
latents += self.mlp.gradient(latents) * self.descent_lr
try:
smiles = [canonicalize(smi) for smi in
self.cvae.hot_to_smiles(self.cvae.decode(latents), strip=True, numpy=True)]
except (RuntimeError, ValueError):
logger.error('Decoding failed')
continue
for i, smi in enumerate(smiles):
try:
if smi is not None and is_valid(smi) and lipinski_filter(smi):
if smi not in results_builder:
latent_score = self.mlp.latent_score(latents[i].reshape(1, -1))
logger.info(f'Optimized from {before[i]} to {latent_score}')
output_path = os.path.join(
self.output_dir,
f'{results_builder.size}.mol2') if self.output_dir is not None else None
docking_score = smiles_docking_score_fn(smi, output_path=output_path,
n_cpu=self.docking_n_cpu)
results_builder.append(
smi,
latent_vector=latents[i],
predicted_score=latent_score,
**docking_score
)
else:
logger.info('Generated SMILES %s already present in OptimizedMoleculesBuilder', smi)
except Exception:
logger.error('Docking failed for ' + smi)
if results_builder.size >= number_molecules:
logger.info('Generating finished')
break
results_builder.total_samples += self.batch_size
return results_builder.build()
def random_gauss(self, smiles_docking_score_fn, size):
assert smiles_docking_score_fn is not None
assert size is not None
logger.info('Random gauss sampling start')
results_builder = OptimizedMolecules.Builder()
while results_builder.size < size:
logger.info(f'Random sampled {results_builder.size} / {size}')
latents = np.random.normal(size=(self.batch_size, self.latent))
smiles = [canonicalize(smi) for smi in
self.cvae.hot_to_smiles(self.cvae.decode(latents), strip=True, numpy=True)]
for i, smi in enumerate(smiles):
try:
if smi is not None and is_valid(smi) and lipinski_filter(smi):
if smi not in results_builder:
output_path = os.path.join(
self.output_dir,
f'{results_builder.size}.gauss.mol2') if self.output_dir is not None else None
docking_score = smiles_docking_score_fn(smi, output_path=output_path,
n_cpu=self.docking_n_cpu)
results_builder.append(
smi,
latent_vector=latents[i],
predicted_score=self.mlp.latent_score(latents[i].reshape(1, -1)),
**docking_score
)
else:
logger.info('Generated SMILES %s already present in OptimizedMoleculesBuilder', smi)
except Exception:
logger.error('Docking failed for ' + smi)
if results_builder.size >= size:
logger.info('Random sampling finished')
break
results_builder.total_samples += self.batch_size
logger.info('Random gauss sampling finished')
return results_builder.build()
def test_builder_should_raise_when_smiles_is_not_str():
uut = OptimizedMolecules.Builder()
with pytest.raises(TypeError):
uut.append(34)
def test_builder_should_raise_on_empty_smiles():
uut = OptimizedMolecules.Builder()
with pytest.raises(ValueError):
uut.append('')
def test_builder_in_operator_returns_false_when_smiles_not_already_present():
uut = OptimizedMolecules.Builder()
result = 'C' in uut
assert not result
def test_to_csv_call_with_without_columns_and_columns_raises(tmp_path):
uut = OptimizedMolecules.Builder().build()
with pytest.raises(ValueError):
uut.to_csv('dummy_path', columns=[], without_columns=[])
