def main():
#load the config_json into the optionhandler object
config_json = FLAGS.config_json
with open(config_json) as config_fobj:
config_dict = json.load(config_fobj)
optionhandler = helpers.OptionHandler(config_dict)
#Handle the summaries_dir output directory
summaries_dir = optionhandler._summariesdir
if FLAGS.output_dir:
summaries_dir = FLAGS.output_dir
if not os.path.exists(summaries_dir):
os.mkdir(summaries_dir)
if not os.path.exists(os.path.join(summaries_dir, 'scripts')):
os.mkdir(os.path.join(summaries_dir, 'scripts'))
#Enable overwriting the sequence path
if FLAGS.sequence:
optionhandler.seqfile = FLAGS.sequence
#Initialize
classifier = DeeProtein(optionhandler)
classifier.init_for_machine_infer(optionhandler._batchsize)
evolver = GeneticAlg(optionhandler, classifier,
helpers.TFrecords_generator(optionhandler))
evolver.evolve()
def main():
with open(FLAGS.config_json) as config_fobj:
config_dict = json.load(config_fobj)
# set the gpu context
if not FLAGS.gpu:
if config_dict["gpu"] == 'True':
config_dict["gpu"] = "False"
optionhandler = helpers.OptionHandler(config_dict)
model = DeeProtein(optionhandler)
model.validate()
def main():
with open(FLAGS.config_json) as config_fobj:
config_dict = json.load(config_fobj)
# set the gpu context
if not FLAGS.gpu:
if config_dict["gpu"] == 'True':
config_dict["gpu"] = "False"
optionhandler = helpers.OptionHandler(config_dict)
model = DeeProtein(optionhandler)
model.train(restore_whole=FLAGS.restore_whole, binary=True)
def main():
with open(FLAGS.config_json) as config_fobj:
config_dict = json.load(config_fobj)
# set the gpu context
if not FLAGS.gpu:
if config_dict["gpu"] == 'True':
config_dict["gpu"] = "False"
optionhandler = helpers.OptionHandler(config_dict)
model = DeeProtein(optionhandler)
model.generate_embedding(embedding_dims=512, reduce_dims=False)
def main():
if FLAGS.obo_file:
obo_file = FLAGS.obo_file
else: # no GOdag file was specified -> download it.
url = 'http://purl.obolibrary.org/obo/go.obo'
obo_file = wget.download(url)
GOdag = GODag(obo_file, optional_attrs=['relationship'])
with open(FLAGS.config_json) as config_fobj:
config_dict = json.load(config_fobj)
# set the gpu context
if not FLAGS.gpu:
if config_dict["gpu"] == 'True':
config_dict["gpu"] = "False"
opts = helpers.OptionHandler(config_dict)
deeprotein = DeeProtein(opts, inference=True)
with tf.Graph().as_default():
deeprotein.initialize_helpers()
# graph for inference:
deeprotein.is_train = False
deeprotein._opts._batchsize = 1
input_seq_node = tf.placeholder(tf.float32, [
deeprotein._opts._batchsize, deeprotein._opts._depth,
deeprotein._opts._windowlength, 1
])
inference_net, _ = deeprotein.model(input_seq_node, valid_mode=False)
double_sigmoid_logits = tf.nn.sigmoid(inference_net.outputs)
# as its a double outlayer take a mean and get the var
mean_inf_sigmoid_logits, var_inf_sigmoid_logits = tf.nn.moments(
double_sigmoid_logits, axes=[2])
deeprotein.session.run(tf.global_variables_initializer())
deeprotein.saver = tf.train.Saver()
# restore the session from ckpt
if FLAGS.restore_from_explicit_ckpt:
deeprotein.restore_model_from_checkpoint(
FLAGS.restore_from_explicit_ckpt, deeprotein.session)
else:
# restore all the weights from dir in checkpointfile:
_ = deeprotein.load_model_weights(
inference_net, session=deeprotein.session,
name='Classifier') #THIS RUNS THE SESSION INTERNALLY
encoded_seq = deeprotein.batchgen._encode_single_seq(FLAGS.sequence)
# expand it to batch dim and channels dim:
assert len(encoded_seq.shape) == 2
encoded_seq = np.reshape(
encoded_seq, [1, encoded_seq.shape[0], encoded_seq.shape[1], 1])
# run the session and retrieve the predicted classes
class_logits, class_variance = deeprotein.session.run(
[mean_inf_sigmoid_logits, var_inf_sigmoid_logits],
feed_dict={input_seq_node: encoded_seq})
# logits is a 2D array of shape [1, nclasses]
thresholded_logits = np.where(class_logits >= 0.5)
# get the IDs of the classes:
predicted_IDs = thresholded_logits[1].tolist()
# get the corresponding classes from the EC_file:
predicted_classes = [
deeprotein.batchgen.id_to_class[_id] for _id in predicted_IDs
]
deeprotein.log_file.write('Predicted classes:\n')
for i, go in enumerate(predicted_classes):
try:
go = go.split('_')[1]
except:
pass
output = '%s\t%s:\t%f, %f\n' % (go, GOdag[go].name, class_logits[
0, predicted_IDs[i]], class_variance[0, predicted_IDs[i]])
deeprotein.log_file.write(output)
print(output)
if not predicted_classes:
output = 'The sequence could not be classified with enough fidelity.\n\n'
deeprotein.log_file.write(output)
print(output)