def sitagliptin_replacement() -> GoalDirectedBenchmark:
# Find a molecule dissimilar to sitagliptin, but with the same properties
smiles = "Fc1cc(c(F)cc1F)CC(N)CC(=O)N3Cc2nnc(n2CC3)C(F)(F)F"
sitagliptin = Chem.MolFromSmiles(smiles)
target_logp = logP(sitagliptin)
target_tpsa = tpsa(sitagliptin)
similarity = TanimotoScoringFunction(smiles,
fp_type="ECFP4",
score_modifier=GaussianModifier(
mu=0, sigma=0.1))
lp = RdkitScoringFunction(descriptor=logP,
score_modifier=GaussianModifier(mu=target_logp,
sigma=0.2))
tp = RdkitScoringFunction(descriptor=tpsa,
score_modifier=GaussianModifier(mu=target_tpsa,
sigma=5))
isomers = IsomerScoringFunction("C16H15F6N5O")
specification = uniform_specification(1, 10, 100)
return GoalDirectedBenchmark(
name="Sitagliptin MPO",
objective=GeometricMeanScoringFunction([similarity, lp, tp, isomers]),
contribution_specification=specification,
)
def smarts_with_other_target(smarts: str, other_molecule: str) -> ScoringFunction:
smarts_scoring_function = SMARTSScoringFunction(target=smarts)
other_mol = Chem.MolFromSmiles(other_molecule)
target_logp = logP(other_mol)
target_tpsa = tpsa(other_mol)
target_bertz = bertz(other_mol)
lp = RdkitScoringFunction(descriptor=logP,
score_modifier=GaussianModifier(mu=target_logp, sigma=0.2))
tp = RdkitScoringFunction(descriptor=tpsa,
score_modifier=GaussianModifier(mu=target_tpsa, sigma=5))
bz = RdkitScoringFunction(descriptor=bertz,
score_modifier=GaussianModifier(mu=target_bertz, sigma=30))
return GeometricMeanScoringFunction([smarts_scoring_function, lp, tp, bz])
def weird_physchem() -> GoalDirectedBenchmark:
min_bertz = RdkitScoringFunction(descriptor=bertz,
score_modifier=MaxGaussianModifier(
mu=1500, sigma=200))
mol_under_400 = RdkitScoringFunction(descriptor=mol_weight,
score_modifier=MinGaussianModifier(
mu=400, sigma=40))
aroma = RdkitScoringFunction(descriptor=num_aromatic_rings,
score_modifier=MinGaussianModifier(mu=3,
sigma=1))
fluorine = RdkitScoringFunction(descriptor=AtomCounter('F'),
score_modifier=GaussianModifier(mu=6,
sigma=1.0))
opt_weird = ArithmeticMeanScoringFunction(
[min_bertz, mol_under_400, aroma, fluorine])
specification = uniform_specification(1, 10, 100)
return GoalDirectedBenchmark(name='Physchem MPO',
objective=opt_weird,
contribution_specification=specification)
def start_pop_ranolazine() -> GoalDirectedBenchmark:
ranolazine = 'COc1ccccc1OCC(O)CN2CCN(CC(=O)Nc3c(C)cccc3C)CC2'
modifier = ClippedScoreModifier(upper_x=0.7)
similar_to_ranolazine = TanimotoScoringFunction(ranolazine,
fp_type='AP',
score_modifier=modifier)
logP_under_4 = RdkitScoringFunction(descriptor=logP,
score_modifier=MaxGaussianModifier(
mu=7, sigma=1))
aroma = RdkitScoringFunction(descriptor=num_aromatic_rings,
score_modifier=MinGaussianModifier(mu=1,
sigma=1))
fluorine = RdkitScoringFunction(descriptor=AtomCounter('F'),
score_modifier=GaussianModifier(mu=1,
sigma=1.0))
optimize_ranolazine = ArithmeticMeanScoringFunction(
[similar_to_ranolazine, logP_under_4, fluorine, aroma])
specification = uniform_specification(1, 10, 100)
return GoalDirectedBenchmark(name='Ranolazine MPO',
objective=optimize_ranolazine,
contribution_specification=specification,
starting_population=[ranolazine])
def test_isomer_scoring_function_penalizes_incorrect_number_atoms():
c11h24_arithmetic = IsomerScoringFunction('C12H24',
mean_function='arithmetic')
c11h24_geometric = IsomerScoringFunction('C12H24',
mean_function='geometric')
# all those smiles fit the formula C11H24O
smiles1 = 'CCCCCCCCOCCC'
smiles2 = 'CC(CCOC)CCCCCC'
smiles3 = 'COCCCC(CC(C)CC)C'
# the penalty corresponds to a deviation of 1.0 from the gaussian modifier in the number of C atoms
c_score = GaussianModifier(mu=0, sigma=1)(1.0)
n_atoms_score = 1.0
h_score = 1.0
expected_score_arithmetic = (n_atoms_score + c_score + h_score) / 3.0
expected_score_geometric = (n_atoms_score * c_score * h_score)**(1 / 3)
assert c11h24_arithmetic.score(smiles1) == pytest.approx(
expected_score_arithmetic)
assert c11h24_arithmetic.score(smiles2) == pytest.approx(
expected_score_arithmetic)
assert c11h24_arithmetic.score(smiles3) == pytest.approx(
expected_score_arithmetic)
assert c11h24_geometric.score(smiles1) == pytest.approx(
expected_score_geometric)
assert c11h24_geometric.score(smiles2) == pytest.approx(
expected_score_geometric)
assert c11h24_geometric.score(smiles3) == pytest.approx(
expected_score_geometric)
def ranolazine_mpo() -> GoalDirectedBenchmark:
"""
Make start_pop_ranolazine more polar and add a fluorine
"""
ranolazine = "COc1ccccc1OCC(O)CN2CCN(CC(=O)Nc3c(C)cccc3C)CC2"
modifier = ClippedScoreModifier(upper_x=0.7)
similar_to_ranolazine = TanimotoScoringFunction(ranolazine,
fp_type="AP",
score_modifier=modifier)
logP_under_4 = RdkitScoringFunction(descriptor=logP,
score_modifier=MaxGaussianModifier(
mu=7, sigma=1))
tpsa_f = RdkitScoringFunction(descriptor=tpsa,
score_modifier=MaxGaussianModifier(mu=95,
sigma=20))
fluorine = RdkitScoringFunction(descriptor=AtomCounter("F"),
score_modifier=GaussianModifier(mu=1,
sigma=1.0))
optimize_ranolazine = GeometricMeanScoringFunction(
[similar_to_ranolazine, logP_under_4, fluorine, tpsa_f])
specification = uniform_specification(1, 10, 100)
return GoalDirectedBenchmark(
name="Ranolazine MPO",
objective=optimize_ranolazine,
contribution_specification=specification,
starting_population=[ranolazine],
)
def pioglitazone_mpo() -> GoalDirectedBenchmark:
# pioglitazone with same mw but less rotatable bonds
smiles = 'O=C1NC(=O)SC1Cc3ccc(OCCc2ncc(cc2)CC)cc3'
pioglitazone = Chem.MolFromSmiles(smiles)
target_molw = mol_weight(pioglitazone)
similarity = TanimotoScoringFunction(smiles, fp_type='ECFP4',
score_modifier=GaussianModifier(mu=0, sigma=0.1))
mw = RdkitScoringFunction(descriptor=mol_weight,
score_modifier=GaussianModifier(mu=target_molw, sigma=10))
rb = RdkitScoringFunction(descriptor=num_rotatable_bonds,
score_modifier=GaussianModifier(mu=2, sigma=0.5))
specification = uniform_specification(1, 10, 100)
return GoalDirectedBenchmark(name='Pioglitazone MPO',
objective=GeometricMeanScoringFunction([similarity, mw, rb]),
contribution_specification=specification)
def test_gaussian_function():
mu = -1.223
sigma = 0.334
f = GaussianModifier(mu=mu, sigma=sigma)
assert f(mu) == 1.0
assert f(scalar_value) == gaussian(scalar_value, mu, sigma)
assert np.allclose(f(value_array), gaussian(value_array, mu, sigma))
def tpsa_benchmark(target: float) -> GoalDirectedBenchmark:
benchmark_name = f'TPSA (target: {target})'
objective = RdkitScoringFunction(descriptor=tpsa,
score_modifier=GaussianModifier(mu=target, sigma=20.0))
specification = uniform_specification(1, 10, 100)
return GoalDirectedBenchmark(name=benchmark_name,
objective=objective,
contribution_specification=specification)
def __init__(self, molecular_formula: str) -> None:
"""
Args:
molecular_formula: target molecular formula
"""
super().__init__()
element_occurrences = parse_molecular_formula(molecular_formula)
total_number_atoms = sum(element_tuple[1] for element_tuple in element_occurrences)
# scoring functions for each element
self.functions = [RdkitScoringFunction(descriptor=AtomCounter(element),
score_modifier=GaussianModifier(mu=n_atoms, sigma=1.0))
for element, n_atoms in element_occurrences]
# scoring functions for the total number of atoms
self.functions.append(RdkitScoringFunction(descriptor=num_atoms,
score_modifier=GaussianModifier(mu=total_number_atoms, sigma=2.0)))
def amlodipine_rings() -> GoalDirectedBenchmark:
# amlodipine with 3 rings
amlodipine = TanimotoScoringFunction(r'Clc1ccccc1C2C(=C(/N/C(=C2/C(=O)OCC)COCCN)C)\C(=O)OC', fp_type='ECFP4')
rings = RdkitScoringFunction(descriptor=num_rings,
score_modifier=GaussianModifier(mu=3, sigma=0.5))
specification = uniform_specification(1, 10, 100)
return GoalDirectedBenchmark(name='Amlodipine MPO',
objective=GeometricMeanScoringFunction([amlodipine, rings]),
contribution_specification=specification)
def perindopril_rings() -> GoalDirectedBenchmark:
# perindopril with two aromatic rings
perindopril = TanimotoScoringFunction('O=C(OCC)C(NC(C(=O)N1C(C(=O)O)CC2CCCCC12)C)CCC',
fp_type='ECFP4')
arom_rings = RdkitScoringFunction(descriptor=num_aromatic_rings,
score_modifier=GaussianModifier(mu=2, sigma=0.5))
specification = uniform_specification(1, 10, 100)
return GoalDirectedBenchmark(name='Perindopril MPO',
objective=GeometricMeanScoringFunction([perindopril, arom_rings]),
contribution_specification=specification)
def determine_scoring_functions(
molecular_formula: str) -> List[RdkitScoringFunction]:
element_occurrences = parse_molecular_formula(molecular_formula)
total_number_atoms = sum(element_tuple[1]
for element_tuple in element_occurrences)
# scoring functions for each element
functions = [
RdkitScoringFunction(descriptor=AtomCounter(element),
score_modifier=GaussianModifier(mu=n_atoms,
sigma=1.0))
for element, n_atoms in element_occurrences
]
# scoring functions for the total number of atoms
functions.append(
RdkitScoringFunction(descriptor=num_atoms,
score_modifier=GaussianModifier(
mu=total_number_atoms, sigma=2.0)))
return functions
def test_isomer_scoring_function_penalizes_incorrect_number_atoms():
c11h24 = IsomerScoringFunction('C12H24')
# all those smiles fit the formula C11H24O
smiles1 = 'CCCCCCCCOCCC'
smiles2 = 'CC(CCOC)CCCCCC'
smiles3 = 'COCCCC(CC(C)CC)C'
# the penalty corresponds to a deviation of 1.0 from the gaussian modifier in the number of C atoms
penalty_tot_num_atoms = 1.0 - GaussianModifier(mu=0, sigma=1)(1.0)
expected_score = 1.0 - penalty_tot_num_atoms / 3.0
assert c11h24.score(smiles1) == pytest.approx(expected_score)
assert c11h24.score(smiles2) == pytest.approx(expected_score)
assert c11h24.score(smiles3) == pytest.approx(expected_score)
