def sc_hard_fexofenadine() -> GoalDirectedBenchmark:
specification = uniform_specification(1, 10, 100)
benchmark_object = hard_fexofenadine()
sa_biased = ScoringFunctionSAWrapper(benchmark_object.objective,
SCScoreModifier())
return GoalDirectedBenchmark(name='SC_fexofenadine',
objective=sa_biased,
contribution_specification=specification)
def sa_ranolazine() -> GoalDirectedBenchmark:
specification = uniform_specification(1, 10, 100)
benchmark_object = start_pop_ranolazine()
sa_biased = ScoringFunctionSAWrapper(benchmark_object.objective,
SAScoreModifier())
return GoalDirectedBenchmark(name='SA_ranolazine',
objective=sa_biased,
contribution_specification=specification)
def sa_hard_osimertinib() -> GoalDirectedBenchmark:
specification = uniform_specification(1, 10, 100)
benchmark_object = hard_osimertinib()
sa_biased = ScoringFunctionSAWrapper(benchmark_object.objective,
SAScoreModifier())
return GoalDirectedBenchmark(name='SA_osimertinib',
objective=sa_biased,
contribution_specification=specification)
def sa_qed_benchmark() -> GoalDirectedBenchmark:
specification = uniform_specification(1, 10, 100)
sa_qed = ScoringFunctionSAWrapper(RdkitScoringFunction(descriptor=qed), SAScoreModifier())
return GoalDirectedBenchmark(name='SA_QED',
objective=sa_qed,
contribution_specification=specification)
def generate_optimized_molecules(
self,
scoring_function: ScoringFunction,
number_molecules: int,
starting_population: Optional[List[str]] = None) -> List[str]:
# sa_scoring_function = ScoringFunctionSAWrapper(scoring_function, SAScoreModifier(mu=3.2356, sigma=1.0156))
# sa_scoring_function = ScoringFunctionSAWrapper(scoring_function, SCScoreModifier(mu=2.9308, sigma=0.1803))
# sa_scoring_function = ScoringFunctionSAWrapper(scoring_function, SmilesModifier(self.sigma, self.mu))
# sa_scoring_function = ScoringFunctionSAWrapper(scoring_function, SAScoreModifier(mu=self.mu, sigma=self.sigma))
# sa_scoring_function = ScoringFunctionSAWrapper(scoring_function, SCScoreModifier(mu=self.mu, sigma=self.sigma))
sa_scoring_function = ScoringFunctionSAWrapper(
scoring_function, SmilesModifier(self.sigma, self.mu))
if number_molecules > self.population_size:
self.population_size = number_molecules
print(
f'Benchmark requested more molecules than expected: new population is {number_molecules}'
)
# fetch initial population?
if starting_population is None:
print('selecting initial population...')
if self.random_start:
starting_population = np.random.choice(self.all_smiles,
self.population_size)
else:
starting_population = self.top_k(self.all_smiles,
sa_scoring_function,
self.population_size)
# select initial population
population_smiles = heapq.nlargest(self.population_size,
starting_population,
key=sa_scoring_function.score)
population_mol = [Chem.MolFromSmiles(s) for s in population_smiles]
population_scores = self.pool(
delayed(score_mol)(m, sa_scoring_function.score)
for m in population_mol)
# evolution: go go go!!
t0 = time()
patience = 0
for generation in range(self.generations):
# new_population
mating_pool = make_mating_pool(population_mol, population_scores,
self.offspring_size)
offspring_mol = self.pool(
delayed(reproduce)(mating_pool, self.mutation_rate)
for _ in range(self.population_size))
# add new_population
population_mol += offspring_mol
population_mol = sanitize(population_mol)
# stats
gen_time = time() - t0
mol_sec = self.population_size / gen_time
t0 = time()
old_scores = population_scores
population_scores = self.pool(
delayed(score_mol)(m, sa_scoring_function.score)
for m in population_mol)
population_tuples = list(zip(population_scores, population_mol))
population_tuples = sorted(population_tuples,
key=lambda x: x[0],
reverse=True)[:self.population_size]
population_mol = [t[1] for t in population_tuples]
population_scores = [t[0] for t in population_tuples]
# early stopping
if population_scores == old_scores:
patience += 1
print(f'Failed to progress: {patience}')
if patience >= self.patience:
print(f'No more patience, bailing...')
break
else:
patience = 0
print(f'{generation} | '
f'max: {np.max(population_scores):.3f} | '
f'avg: {np.mean(population_scores):.3f} | '
f'min: {np.min(population_scores):.3f} | '
f'std: {np.std(population_scores):.3f} | '
f'sum: {np.sum(population_scores):.3f} | '
f'{gen_time:.2f} sec/gen | '
f'{mol_sec:.2f} mol/sec')
# finally
return [Chem.MolToSmiles(m) for m in population_mol][:number_molecules]
