def predict_reference(weights_path,
gap_id,
fasta_path,
embedding_dim,
latent_dim,
min_seed_length,
beam_length,
base_path=None):
importer = SequenceImporter()
reverser = SequenceReverser()
sequences = importer.import_fasta([fasta_path])[0:2]
model = GapPredictModel(min_seed_length=min_seed_length,
stateful=False,
embedding_dim=embedding_dim,
latent_dim=latent_dim,
with_gpu=True)
model.load_weights(weights_path + 'my_model_weights.h5')
terminal_directory_character = UTILS.get_terminal_directory_character()
first_directory = "beam_search" + terminal_directory_character + "regenerate_seq" + terminal_directory_character
for i in range(len(sequences)):
sequence = sequences[i]
#TODO: messy since we assume that i will always be [0, 1]
if i == 0:
flank = "left_flank"
static_path = first_directory + flank + terminal_directory_character
elif i == 1:
flank = "right_flank"
static_path = first_directory + flank + terminal_directory_character
forward_seed = sequence[:min_seed_length]
prediction_length = len(sequence) - min_seed_length
reverse_complement_seed = reverser.reverse_complement(
sequence)[:min_seed_length]
directionality = "forward"
flank_id = gap_id + "_" + flank + "_" + directionality
predict(model,
forward_seed,
prediction_length,
beam_length,
flank_id,
base_path=base_path,
directory=static_path + directionality)
directionality = "reverse_complement"
flank_id = gap_id + "_" + flank + "_" + directionality
predict(model,
reverse_complement_seed,
prediction_length,
beam_length,
flank_id,
base_path=base_path,
directory=static_path + directionality)
def predict_arbitrary_length(weights_path,
gap_id,
fasta_path,
embedding_dim,
latent_dim,
length_to_predict,
base_path=None):
importer = SequenceImporter()
reverser = SequenceReverser()
sequences = importer.import_fasta([fasta_path])
left_flank = sequences[0]
right_flank = reverser.reverse_complement(sequences[1])
model = GapPredictModel(min_seed_length=None,
stateful=False,
batch_size=1,
embedding_dim=embedding_dim,
latent_dim=latent_dim,
with_gpu=True)
model.load_weights(weights_path + 'my_model_weights.h5')
forward_predict = predict_gaps(model,
left_flank,
length_to_predict,
base_path=base_path,
directory="predict_gap/forward")
reverse_predict = predict_gaps(model,
right_flank,
length_to_predict,
base_path=base_path,
directory="predict_gap/reverse_complement")
writer = DataWriter(root_directory=base_path)
postfix = "_LD_" + str(latent_dim)
writer.save_complements(forward_predict,
reverse_predict,
gap_id,
postfix=postfix,
fasta_ref=fasta_path)
def predict_arbitrary_length(weights_path,
gap_id,
fasta_path,
embedding_dim,
latent_dim,
length_to_predict,
beam_length,
base_path=None):
importer = SequenceImporter()
reverser = SequenceReverser()
sequences = importer.import_fasta([fasta_path])
left_flank = sequences[0]
right_flank = reverser.reverse_complement(sequences[1])
model = GapPredictModel(min_seed_length=None,
stateful=False,
embedding_dim=embedding_dim,
latent_dim=latent_dim,
with_gpu=True)
model.load_weights(weights_path + 'my_model_weights.h5')
terminal_directory_character = UTILS.get_terminal_directory_character()
directory = "beam_search" + terminal_directory_character + "predict_gap" + terminal_directory_character
flank_id = gap_id + "_" + "left_forward"
predict(model,
left_flank,
length_to_predict,
beam_length,
flank_id,
base_path=base_path,
directory=directory + "forward")
flank_id = gap_id + "_" + "right_reverse_complement"
predict(model,
right_flank,
length_to_predict,
beam_length,
flank_id,
base_path=base_path,
directory=directory + "reverse_complement")
def train_model(base_directory,
min_seed_length,
reference,
reads,
epochs,
batch_sizes,
rnn_dims,
embedding_dims,
replicates,
patience,
seed_range_upper,
log_samples=False):
include_reverse_complement = True
importer = SequenceImporter()
reverse_complementer = SequenceReverser()
reads = importer.import_fastq([reads], include_reverse_complement)
reference_sequence = importer.import_fasta([reference])
reference_sequences = [
reference_sequence[0],
reverse_complementer.reverse_complement(reference_sequence[0]),
reference_sequence[1],
reverse_complementer.reverse_complement(reference_sequence[1])
]
lengths = np.array(list(map(lambda x: len(x), reference_sequences)))
with_gpu = True
log_samples = log_samples
log_training = False
early_stopping = True
legend = ['Mean Sequence']
terminal_directory_character = dir_utils.get_terminal_directory_character()
for batch_size in batch_sizes:
for embedding_dim in embedding_dims:
for latent_dim in rnn_dims:
for i in range(replicates):
weights_path = dir_utils.clean_directory_string(
base_directory)
inner_directory = "BS_" + str(batch_size) + "_ED_" + str(embedding_dim) + "_LD_" + str(latent_dim) \
+ "_E_" + str(epochs) + "_R_" + str(i)
writer_path = weights_path + inner_directory + terminal_directory_character
dir_utils.mkdir(weights_path)
dir_utils.mkdir(writer_path)
model = GapPredictModel(
min_seed_length=min_seed_length,
stateful=False,
batch_size=batch_size,
epochs=epochs,
embedding_dim=embedding_dim,
latent_dim=latent_dim,
with_gpu=with_gpu,
log_samples=log_samples,
reference_sequences=reference_sequences,
log_training=log_training,
early_stopping=early_stopping,
patience=patience,
seed_range_upper=seed_range_upper,
base_path=base_directory)
start_time = time.time()
history = model.fit(reads)
model.save_weights(weights_path + 'my_model_weights.h5')
end_time = time.time()
print("Fitting took " + str(end_time - start_time) + "s")
accuracy, loss = model.validation_history()
if log_training:
training_accuracy = model.training_history()
else:
training_accuracy = history.history['acc']
training_loss = history.history['loss']
if early_stopping:
best_epoch = model.get_best_epoch()
_plot_training_validation(epochs,
accuracy,
loss,
training_accuracy,
training_loss,
lengths,
writer_path,
legend=legend,
best_epoch=best_epoch)
class TestSequenceImporter(TestCase):
def setUp(self):
self.importer = SequenceImporter()
self.converter = BaseQualityConverter()
self.reverser = SequenceReverser()
def test_import_fastq(self):
sequences = self.importer.import_fastq([
'../data/mini_read_for_test.fastq',
'../data/mini_read_for_test_2.fastq.gz'
])
expected_sequences = np.array([
"AATTGAGTCGTAGTATCCACACCAAGCCGGCGTTATCCGGTGAGGCGCAATGTTGCGGGGGCTTTATCCCTGGTGGCATTGGTTGCTGGAAAGAGAAA",
"AATTGAGTCGTAGTATCCACACCAAGCCGGCGTTATCCGGTGAGGCGCAATGTTGCGGGGGCTTTATCCCTGGTGGCATTGGTTGCTGGAAAGAGAAA",
"GTACAGCTCAGAAAGCGGAGTTGCGCCAAGATTGTTAACCAGCGCAATCACCCGATCGCCAGACTGGAGCGGTTGTTTGGTTTGTTGTTCTTCCTGCC",
"TTACCTTGTGGAGCGACATCCAGAGGCACTTCACCGCTTGCCAGCGGCTTACCATCCAGCGCCACCATCCAGTGCAGGAGCTCGTTATCGCTATGACG"
])
np.testing.assert_array_equal(expected_sequences, sequences)
def test_import_fasta(self):
sequences = self.importer.import_fasta([
'../data/mini_read_for_test.fasta',
'../data/mini_read_for_test_2.fasta.gz'
])
expected_sequences = np.array([
"CTGTTGAGCTATACTGTGGGAAGTCTGCTTGGCCCGTCATTTACCGCTATGCTAATGCAGAATTTCTCCGATAATTTATTGTTTATCATGATCGCCAGCGTATCGTTTATCTATTTGCTG",
"AATTTCTCCGATAATTTATTGTTTATCATGATCGCCAGCGTATCGTTTATCTATTTGCTGCTGTTGAGCTATACTGTGGGAAGTCTGCTTGGCCCGTCATTTACCGCTATGCTAATGCAG",
"ATGATCGCCAGCGTATCGTTTATCTATTTGCTGAATTTCTCCGATAATTTATTGTTTATCGCTTGGCCCGTCATTTACCGCTATGCTAATGCAGCTGTTGAGCTATACTGTGGGAAGTCT",
"ATGATCGCCAGCGTATCGTTTATCTATTTGCTGAATTTCTCCGATAATTTATTGTTTATCGCTTGGCCCGTCATTTACCGCTATGCTAATGCAGCTGTTGAGCTATACTGTGGGAAGTCT",
"CTGTTGAGCTATACTGTGGGAAGTCTGCTTGGCCCGTCATTTACCGCTATGCTAATGCAGAATTTCTCCGATAATTTATTGTTTATCATGATCGCCAGCGTATCGTTTATCTATTTGCTG",
"AATTTCTCCGATAATTTATTGTTTATCATGATCGCCAGCGTATCGTTTATCTATTTGCTGCTGTTGAGCTATACTGTGGGAAGTCTGCTTGGCCCGTCATTTACCGCTATGCTAATGCAG",
"ATGATCGCCAGCGTATCGTTTATCTATTTGCTGAATTTCTCCGATAATTTATTGTTTATCGCTTGGCCCGTCATTTACCGCTATGCTAATGCAGCTGTTGAGCTATACTGTGGGAAGTCT"
])
np.testing.assert_array_equal(expected_sequences, sequences)
def test_import_fastq_with_reverse_complement(self):
sequences = self.importer.import_fastq([
'../data/mini_read_for_test.fastq',
'../data/mini_read_for_test_2.fastq.gz'
], True)
expected_sequences = np.array([
"AATTGAGTCGTAGTATCCACACCAAGCCGGCGTTATCCGGTGAGGCGCAATGTTGCGGGGGCTTTATCCCTGGTGGCATTGGTTGCTGGAAAGAGAAA",
self.reverser.reverse_complement(
"AATTGAGTCGTAGTATCCACACCAAGCCGGCGTTATCCGGTGAGGCGCAATGTTGCGGGGGCTTTATCCCTGGTGGCATTGGTTGCTGGAAAGAGAAA"
),
"AATTGAGTCGTAGTATCCACACCAAGCCGGCGTTATCCGGTGAGGCGCAATGTTGCGGGGGCTTTATCCCTGGTGGCATTGGTTGCTGGAAAGAGAAA",
self.reverser.reverse_complement(
"AATTGAGTCGTAGTATCCACACCAAGCCGGCGTTATCCGGTGAGGCGCAATGTTGCGGGGGCTTTATCCCTGGTGGCATTGGTTGCTGGAAAGAGAAA"
),
"GTACAGCTCAGAAAGCGGAGTTGCGCCAAGATTGTTAACCAGCGCAATCACCCGATCGCCAGACTGGAGCGGTTGTTTGGTTTGTTGTTCTTCCTGCC",
self.reverser.reverse_complement(
"GTACAGCTCAGAAAGCGGAGTTGCGCCAAGATTGTTAACCAGCGCAATCACCCGATCGCCAGACTGGAGCGGTTGTTTGGTTTGTTGTTCTTCCTGCC"
),
"TTACCTTGTGGAGCGACATCCAGAGGCACTTCACCGCTTGCCAGCGGCTTACCATCCAGCGCCACCATCCAGTGCAGGAGCTCGTTATCGCTATGACG",
self.reverser.reverse_complement(
"TTACCTTGTGGAGCGACATCCAGAGGCACTTCACCGCTTGCCAGCGGCTTACCATCCAGCGCCACCATCCAGTGCAGGAGCTCGTTATCGCTATGACG"
)
])
np.testing.assert_array_equal(expected_sequences, sequences)
def test_no_files_fastq(self):
sequences = self.importer.import_fastq([], True)
np.testing.assert_array_equal(sequences, [])
def test_bogus_file_fastq(self):
try:
self.importer.import_fastq(['data/blahblahblah.fastq'])
self.fail()
except FileNotFoundError as e:
pass
except Exception as e:
self.fail()
def test_no_files_fasta(self):
sequences = self.importer.import_fasta([])
np.testing.assert_array_equal(sequences, [])
def test_bogus_file_fasta(self):
try:
self.importer.import_fasta(['data/blahblahblah.fasta'])
self.fail()
except FileNotFoundError as e:
pass
except Exception as e:
self.fail()
def predict_reference(weights_path,
gap_id,
fasta_path,
embedding_dim,
latent_dim,
min_seed_length,
plots=False,
base_path=None):
importer = SequenceImporter()
reverser = SequenceReverser()
sequences = importer.import_fasta([fasta_path])[0:2]
stateful_model = GapPredictModel(min_seed_length=min_seed_length,
stateful=True,
batch_size=1,
embedding_dim=embedding_dim,
latent_dim=latent_dim,
with_gpu=True)
stateless_model = GapPredictModel(min_seed_length=min_seed_length,
stateful=False,
batch_size=1,
embedding_dim=embedding_dim,
latent_dim=latent_dim,
with_gpu=True)
stateful_model.load_weights(weights_path + 'my_model_weights.h5')
stateless_model.load_weights(weights_path + 'my_model_weights.h5')
terminal_directory_character = UTILS.get_terminal_directory_character()
first_directory = "regenerate_seq" + terminal_directory_character
forward_left_flank = None
rc_left_flank = None
forward_right_flank = None
rc_right_flank = None
for i in range(len(sequences)):
sequence = sequences[i]
#TODO: messy since we assume that i will always be [0, 1]
if i == 0:
static_path = first_directory + "left_flank" + terminal_directory_character
elif i == 1:
static_path = first_directory + "right_flank" + terminal_directory_character
forward_predict = predict_flanks(stateful_model,
stateless_model,
min_seed_length,
sequence,
base_path=base_path,
directory=static_path + "forward")
reverse_predict = predict_flanks(stateful_model,
stateless_model,
min_seed_length,
reverser.reverse_complement(sequence),
base_path=base_path,
directory=static_path +
"reverse_complement")
if i == 0:
forward_left_flank = forward_predict
rc_left_flank = reverse_predict
elif i == 1:
forward_right_flank = forward_predict
rc_right_flank = reverse_predict
writer = DataWriter(root_directory=base_path, directory=first_directory)
writer.write_flank_predict_fasta(forward_left_flank, rc_left_flank,
forward_right_flank, rc_right_flank,
latent_dim, gap_id)