def generate_charts(ga: pyeasyga.GeneticAlgorithm):
all_vals = []
best_vals = []
start = time_ns()
ga.create_first_generation()
all_vals.append([x.fitness for x in ga.current_generation])
best_vals.append(ga.current_generation[0].fitness)
for _ in range(1, ga.generations):
if ga.current_generation[0].fitness == 0:
break
ga.create_next_generation()
all_vals.append([x.fitness for x in ga.current_generation])
best_vals.append(ga.current_generation[0].fitness)
print(f'\nin {(time_ns() - start) / 1000000}ms')
avg = [np.average(x) for x in all_vals]
fig, ax = plt.subplots()
ax.plot(best_vals, label="Max")
ax.plot(avg, label="Avg")
plt.xlabel("Generations")
plt.ylabel("Value")
legend = ax.legend(loc='lower right')
plt.show()
def generate(target_params,
insert_aa_seq,
population_size=100,
mutation_probability=0.3,
max_gens_since_improvement=50,
genetic_code=11,
verbose=False):
# back translate to an initial seq
insert = ""
for aa in insert_aa_seq:
try:
insert += Bio.Data.CodonTable.unambiguous_dna_by_id[
genetic_code].back_table[aa]
except:
if aa == "*":
insert += Bio.Data.CodonTable.unambiguous_dna_by_id[
genetic_code].back_table[None]
# create the genetic algorithm instance
ga = GeneticAlgorithm(dna_to_vector(insert),
crossover_probability=0,
maximise_fitness=False,
population_size=population_size,
mutation_probability=mutation_probability)
# get the target values of k
k = list(target_params.keys())
# generate the target vector from the input dict
target = np.array([])
for _k in sorted([x for x in k if x != "codons"]):
target = np.concatenate((target, [
x[1] for x in sorted(target_params[_k].items(), key=lambda x: x[0])
]))
if "codons" in k:
target = np.concatenate((target, [
x[1] for x in sorted(target_params["codons"].items(),
key=lambda x: x[0])
]))
def vector(seq):
output = k_mer_frequencies(seq, [x for x in k if x != "codons"],
include_missing=True,
vector=True)
if "codons" in k:
output = np.concatenate((output, [
x[1]
for x in sorted(codon_frequencies(seq, genetic_code).items(),
key=lambda x: x[0])
]))
return output
def fitness(individual, data):
individual = vector_to_dna(individual)
# fitness = np.linalg.norm(target - vector(individual))
fitness = jensen_shannon_divergence([
dit.ScalarDistribution(target),
dit.ScalarDistribution(vector(individual))
])
return fitness
ga.fitness_function = fitness
synonymous_codons = _synonymous_codons(genetic_codes[genetic_code])
def mutate(individual):
while True:
# choose a random codon
codon_idx = np.random.randint(len(individual) / 6) * 6
# figure out which codon it is
codon = vector_to_dna(individual[codon_idx:codon_idx + 6])
# ensure that mutations actually change the sequence
if len(synonymous_codons[codon]) != 1:
break
# choose a new one at random for the AA
new_codon = dna_to_vector(
np.random.choice(
[x for x in synonymous_codons[codon] if x != codon]))
# replace it in the individual
individual[codon_idx:codon_idx + 6] = new_codon
return individual
ga.mutate_function = mutate
def create_individual(seed_data):
individual = vector_to_dna(seed_data)
new = ""
for codon in [
individual[i:i + 3] for i in range(0, len(individual), 3)
]:
if len(synonymous_codons[codon]) == 1:
new += codon
continue
new += np.random.choice(
[x for x in synonymous_codons[codon] if x != codon])
return dna_to_vector(new)
ga.create_individual = create_individual
# set up for GA run
ga.create_first_generation()
gens_since_improvement = 0
best_indv_fitness = ga.best_individual()[0]
counter = 1
# run the GA
while gens_since_improvement < max_gens_since_improvement:
ga.create_next_generation()
if ga.best_individual()[0] < best_indv_fitness:
best_indv_fitness = ga.best_individual()[0]
gens_since_improvement = 0
else:
gens_since_improvement += 1
if verbose:
print(
"Gen: %s\tSince Improvement: %s/%s\tFitness: %s".expandtabs(15)
% (counter, gens_since_improvement, max_gens_since_improvement,
ga.best_individual()[0]),
end="\r")
counter += 1
if verbose: print()
best_seq = vector_to_dna(ga.best_individual()[1])
best_freqs = vector(best_seq)
return best_seq
def generate(target_params, insert_aa_seq, population_size=100, mutation_probability=0.3, crossover_probability=0.8, max_gens_since_improvement=50, genetic_code=11, verbose=False):
'''Generate a sequence matching :math:`k`-mer usage.
Args:
target_params (dict): The parameters to optimize towards. Should be of the format {:math:`k_n`: {:math:`k_{n1}`: 0.2, :math:`k_{n2}`: 0.3,...}...}
insert_aa_seq (str): The amino acid sequence for the optimized sequence.
population_size (int, optional): The size of the population for the genetic algorithm. Defaults to 100.
mutation_probability (float, optional): The likelihood of changing each member of each generation. Defaults to 0.3.
crossover_probability (float, optional): The likelihood of each member of the population undergoing crossover. Defaults to 0.8.
max_gens_since_improvement (int, optional): The number of generations of no improvement after which to stop optimization. Defaults to 50.
genetic_code (int, optional): The genetic code to use. Defaults to 11, the standard genetic code.
verbose (bool, optional): Whether to print the generation number, generations since improvement, and fitness. Defaults to false.
Returns:
str: The generated sequence.
'''
# back translate to an initial seq
insert = ""
for aa in insert_aa_seq:
try:
insert += Bio.Data.CodonTable.unambiguous_dna_by_id[genetic_code].back_table[aa]
except:
if aa == "*":
insert += Bio.Data.CodonTable.unambiguous_dna_by_id[genetic_code].back_table[None]
# create the genetic algorithm instance
ga = GeneticAlgorithm(dna_to_vector(insert),
crossover_probability=crossover_probability,
maximise_fitness=False,
population_size=population_size,
mutation_probability=mutation_probability)
# get the target values of k
k = list(target_params.keys())
# generate the target vector from the input dict
target = np.array([])
for _k in sorted([x for x in k if x != "codons"]):
target = np.concatenate((target, [x[1] for x in sorted(target_params[_k].items(), key=lambda x: x[0])]))
if "codons" in k:
target = np.concatenate((target, [x[1] for x in sorted(target_params["codons"].items(), key=lambda x: x[0])]))
def vector(seq):
output = k_mer_frequencies(seq, [x for x in k if x != "codons"], include_missing=True, vector=True)
if "codons" in k:
output = np.concatenate((output, [x[1] for x in sorted(codon_frequencies(seq, genetic_code).items(), key=lambda x: x[0])]))
return output
def fitness(individual, data):
individual = vector_to_dna(individual)
# fitness = np.linalg.norm(target - vector(individual))
fitness = jensen_shannon_divergence([dit.ScalarDistribution(target), dit.ScalarDistribution(vector(individual))])
return fitness
ga.fitness_function = fitness
synonymous_codons = _synonymous_codons(genetic_codes[genetic_code])
def mutate(individual):
while True:
# choose a random codon
codon_idx = np.random.randint(len(individual) / 6) * 6
# figure out which codon it is
codon = vector_to_dna(individual[codon_idx:codon_idx+6])
# ensure that mutations actually change the sequence
if len(synonymous_codons[codon]) != 1:
break
# choose a new one at random for the AA
new_codon = dna_to_vector(np.random.choice([x for x in synonymous_codons[codon] if x != codon]))
# replace it in the individual
individual[codon_idx:codon_idx+6] = new_codon
return individual
ga.mutate_function = mutate
def crossover(parent_1, parent_2):
parent_1, parent_2 = list(parent_1), list(parent_2)
index = random.randrange(1, len(parent_1) / 6) * 6
child_1 = parent_1[:index] + parent_2[index:]
child_2 = parent_2[:index] + parent_1[index:]
return child_1, child_2
ga.crossover_function = crossover
def create_individual(seed_data):
individual = vector_to_dna(seed_data)
new = ""
for codon in [individual[i:i+3] for i in range(0, len(individual), 3)]:
if len(synonymous_codons[codon]) == 1:
new += codon
continue
new += np.random.choice([x for x in synonymous_codons[codon] if x != codon])
return dna_to_vector(new)
ga.create_individual = create_individual
# set up for GA run
ga.create_first_generation()
gens_since_improvement = 0
best_indv_fitness = ga.best_individual()[0]
counter = 1
# run the GA
try:
while gens_since_improvement < max_gens_since_improvement:
ga.create_next_generation()
if ga.best_individual()[0] < best_indv_fitness:
best_indv_fitness = ga.best_individual()[0]
gens_since_improvement = 0
else:
gens_since_improvement += 1
if verbose:
print("Gen: %s\tSince Improvement: %s/%s\tFitness: %s".expandtabs(15) % (counter, gens_since_improvement, max_gens_since_improvement, ga.best_individual()[0]), end="\r")
counter += 1
except KeyboardInterrupt:
print("\nStopping early...")
if verbose: print()
best_seq = vector_to_dna(ga.best_individual()[1])
best_freqs = vector(best_seq)
assert Seq(best_seq).translate(genetic_code) == Seq(insert).translate(genetic_code)
return best_seq