def in_silico_mutagenesis(model: Model,
encoded_sequences: np.ndarray) -> np.ndarray:
"""Computes in-silico-mutagenesis scores
Parameters
----------
model: Model
This can be any model that accepts inputs of the required shape and produces
an output of shape `(N_sequences, N_tasks)`.
encoded_sequences: np.ndarray
A numpy array of shape `(N_sequences, N_letters, sequence_length, 1)`
Returns
-------
np.ndarray
A numpy array of ISM scores. The shape is `(num_task, N_sequences, N_letters, sequence_length, 1)`.
"""
# Shape (N_sequences, num_tasks)
wild_type_predictions = model.predict(NumpyDataset(encoded_sequences))
# check whether wild_type_predictions is np.ndarray or not
assert isinstance(wild_type_predictions, np.ndarray)
num_tasks = wild_type_predictions.shape[1]
# Shape (N_sequences, N_letters, sequence_length, 1, num_tasks)
mutagenesis_scores = np.empty(
encoded_sequences.shape + (num_tasks,), dtype=np.float32)
# Shape (N_sequences, num_tasks, 1, 1, 1)
wild_type_predictions = wild_type_predictions[:, np.newaxis, np.newaxis,
np.newaxis]
for sequence_index, (sequence, wild_type_prediction) in enumerate(
zip(encoded_sequences, wild_type_predictions)):
# Mutates every position of the sequence to every letter
# Shape (N_letters * sequence_length, N_letters, sequence_length, 1)
# Breakdown:
# Shape of sequence[np.newaxis] (1, N_letters, sequence_length, 1)
mutated_sequences = np.repeat(
sequence[np.newaxis], np.prod(sequence.shape), axis=0)
# remove wild-type
# len(arange) = N_letters * sequence_length
arange = np.arange(len(mutated_sequences))
# len(horizontal cycle) = N_letters * sequence_length
horizontal_cycle = np.tile(np.arange(sequence.shape[1]), sequence.shape[0])
mutated_sequences[arange, :, horizontal_cycle, :] = 0
# add mutant
vertical_repeat = np.repeat(np.arange(sequence.shape[0]), sequence.shape[1])
mutated_sequences[arange, vertical_repeat, horizontal_cycle, :] = 1
# make mutant predictions
mutated_predictions = model.predict(NumpyDataset(mutated_sequences))
# check whether wild_type_predictions is np.ndarray or not
assert isinstance(mutated_predictions, np.ndarray)
mutated_predictions = mutated_predictions.reshape(sequence.shape +
(num_tasks,))
mutagenesis_scores[
sequence_index] = wild_type_prediction - mutated_predictions
rolled_scores = np.rollaxis(mutagenesis_scores, -1)
return rolled_scores
def fit_model(model_name, model_params, model_dir, data_dir):
"""Builds model from featurized data."""
task_type = Model.get_task_type(model_name)
train_dir = os.path.join(data_dir, "train-data")
train = Dataset(train_dir)
task_types = {task: task_type for task in train.get_task_names()}
model_params["data_shape"] = train.get_data_shape()
model = Model.model_builder(model_name, task_types, model_params)
model.fit(train)
model.save(model_dir)
def create_and_eval_model(train_dataset, test_dataset, task_type,
model_params, model_name, model_dir, tasks):
"""Helper method to create model for test."""
# Fit model
task_types = {task: task_type for task in tasks}
model_params["data_shape"] = train_dataset.get_data_shape()
print("Creating Model object.")
import deepchem.models.deep
model = Model.model_builder(model_name, task_types, model_params)
print("About to fit model")
model.fit(train_dataset)
print("Done fitting, about to save...")
model.save(model_dir)
# Eval model on train
evaluator = Evaluator(model, train_dataset, verbose=True)
with tempfile.NamedTemporaryFile() as train_csv_out:
with tempfile.NamedTemporaryFile() as train_stats_out:
_, performance_df = evaluator.compute_model_performance(
train_csv_out, train_stats_out)
print("train_performance_df")
print(performance_df)
evaluator = Evaluator(model, test_dataset, verbose=True)
with tempfile.NamedTemporaryFile() as test_csv_out:
with tempfile.NamedTemporaryFile() as test_stats_out:
_, performance_df = evaluator.compute_model_performance(
test_csv_out, test_stats_out)
print("test_performance_df")
print(performance_df)
return performance_df.iterrows().next()[1]["r2_score"]
def eval_trained_model(model_type, model_dir, data_dir, csv_out, stats_out):
"""Evaluates a trained model on specified data."""
model = Model.load(model_type, model_dir)
data = Dataset(data_dir)
evaluator = Evaluator(model, data, verbose=True)
_, perf_df = evaluator.compute_model_performance(csv_out, stats_out)
print("Model Performance.")
print(perf_df)
def predict(self, dataset, transformers=[], batch_size=None,
pad_batches=False):
"""
Uses self to make predictions on provided Dataset object.
This is overridden to make sure the batch size is always valid for Tensorflow.
Returns:
y_pred: numpy ndarray of shape (n_samples,)
"""
return Model.predict(self, dataset, transformers,
self.model_instance.batch_size, True)
def load(self, model_dir):
"""
Load keras multitask DNN from disk.
"""
filename = Model.get_model_filename(model_dir)
filename, _ = os.path.splitext(filename)
json_filename = "%s.%s" % (filename, "json")
h5_filename = "%s.%s" % (filename, "h5")
with open(json_filename) as file_obj:
model = model_from_json(file_obj.read())
model.load_weights(h5_filename)
self.raw_model = model
def reload(self):
"""
Load keras multitask DNN from disk.
"""
filename = Model.get_model_filename(self.model_dir)
filename, _ = os.path.splitext(filename)
json_filename = "%s.%s" % (filename, "json")
h5_filename = "%s.%s" % (filename, "h5")
with open(json_filename) as file_obj:
model = model_from_json(file_obj.read())
model.load_weights(h5_filename)
self.raw_model = model
def reload(self, custom_objects={}):
"""
Load keras multitask DNN from disk.
"""
filename = Model.get_model_filename(self.model_dir)
filename, _ = os.path.splitext(filename)
json_filename = "%s.%s" % (filename, "json")
h5_filename = "%s.%s" % (filename, "h5")
with open(json_filename) as file_obj:
model = model_from_json(file_obj.read(), custom_objects=custom_objects)
model.load_weights(h5_filename)
self.model_instance = model
def test_API(self):
"""Straightforward test of multitask deepchem classification API."""
splittype = "scaffold"
feature_types = ["ECFP"]
output_transforms = []
input_transforms = []
task_type = "classification"
# TODO(rbharath): There should be some automatic check to ensure that all
# required model_params are specified.
model_params = {"nb_hidden": 10, "activation": "relu",
"dropout": .5, "learning_rate": .01,
"momentum": .9, "nesterov": False,
"decay": 1e-4, "batch_size": 5,
"nb_epoch": 2}
model_name = "multitask_deep_classifier"
# Featurize input
featurizer = DataFeaturizer(tasks=self.tasks,
smiles_field=self.smiles_field,
verbose=True)
feature_files = featurizer.featurize(self.input_file, feature_types, self.feature_dir)
# Transform data into arrays for ML
samples = FeaturizedSamples(self.samplesdir, feature_files,
reload_data=False)
# Split into train/test
train_samples, test_samples = samples.train_test_split(
splittype, self.train_dir, self.test_dir)
train_dataset = Dataset(self.train_dir, train_samples, feature_types)
test_dataset = Dataset(self.test_dir, test_samples, feature_types)
# Transforming train/test data
train_dataset.transform(input_transforms, output_transforms)
test_dataset.transform(input_transforms, output_transforms)
# Fit model
task_types = {task: task_type for task in self.tasks}
model_params["data_shape"] = train_dataset.get_data_shape()
model = Model.model_builder(model_name, task_types, model_params)
model.fit(train_dataset)
model.save(self.model_dir)
# Eval model on train
evaluator = Evaluator(model, test_dataset, verbose=True)
with tempfile.NamedTemporaryFile() as test_csv_out:
with tempfile.NamedTemporaryFile() as test_stats_out:
evaluator.compute_model_performance(test_csv_out, test_stats_out)
def save(self, out_dir):
"""
Saves underlying keras model to disk.
"""
super(KerasModel, self).save(out_dir)
model = self.get_raw_model()
filename, _ = os.path.splitext(Model.get_model_filename(out_dir))
# Note that keras requires the model architecture and weights to be stored
# separately. A json file is generated that specifies the model architecture.
# The weights will be stored in an h5 file. The pkl.gz file with store the
# target name.
json_filename = "%s.%s" % (filename, "json")
h5_filename = "%s.%s" % (filename, "h5")
# Save architecture
json_string = model.to_json()
with open(json_filename, "wb") as file_obj:
file_obj.write(json_string)
model.save_weights(h5_filename, overwrite=True)
def save(self, out_dir):
"""
Saves underlying keras model to disk.
"""
super(KerasModel, self).save(out_dir)
model = self.get_raw_model()
filename, _ = os.path.splitext(Model.get_model_filename(out_dir))
# Note that keras requires the model architecture and weights to be stored
# separately. A json file is generated that specifies the model architecture.
# The weights will be stored in an h5 file. The pkl.gz file with store the
# target name.
json_filename = "%s.%s" % (filename, "json")
h5_filename = "%s.%s" % (filename, "h5")
# Save architecture
json_string = model.to_json()
with open(json_filename, "wb") as file_obj:
file_obj.write(json_string)
model.save_weights(h5_filename, overwrite=True)
def _create_model(self, splittype, feature_types, input_transforms,
output_transforms, task_type, model_params, model_name,
input_file, tasks, protein_pdb_field=None, ligand_pdb_field=None):
"""Helper method to create model for test."""
# Featurize input
input_file = os.path.join(self.current_dir, input_file)
featurizer = DataFeaturizer(tasks=tasks,
smiles_field=self.smiles_field,
protein_pdb_field=protein_pdb_field,
ligand_pdb_field=ligand_pdb_field,
verbose=True)
feature_files = featurizer.featurize(input_file, feature_types, self.feature_dir)
# Transform data into arrays for ML
samples = FeaturizedSamples(self.samplesdir, feature_files,
reload_data=False)
# Split into train/test
train_samples, test_samples = samples.train_test_split(
splittype, self.train_dir, self.test_dir)
train_dataset = Dataset(self.train_dir, train_samples, feature_types)
test_dataset = Dataset(self.test_dir, test_samples, feature_types)
# Transforming train/test data
train_dataset.transform(input_transforms, output_transforms)
test_dataset.transform(input_transforms, output_transforms)
# Fit model
task_types = {task: task_type for task in tasks}
model_params["data_shape"] = train_dataset.get_data_shape()
model = Model.model_builder(model_name, task_types, model_params)
model.fit(train_dataset)
model.save(self.model_dir)
# Eval model on train
evaluator = Evaluator(model, test_dataset, verbose=True)
with tempfile.NamedTemporaryFile() as test_csv_out:
with tempfile.NamedTemporaryFile() as test_stats_out:
_, _ = evaluator.compute_model_performance(
test_csv_out, test_stats_out)
def reload(self):
"""Loads sklearn model from joblib file on disk."""
self.model_instance = load_from_disk(
Model.get_model_filename(self.model_dir))
nb_tasks = len(sorted_tasks)
y_pred = np.zeros((nb_samples, nb_tasks))
for ind, task in enumerate(sorted_tasks):
task_type = self.task_types[task]
taskname = "task%d" % ind
if task_type == "classification":
# Class probabilities are predicted for classification outputs. Instead,
# output the most likely class.
y_pred_task = np.squeeze(np.argmax(y_pred_dict[taskname], axis=1))
else:
y_pred_task = np.squeeze(y_pred_dict[taskname])
y_pred[:, ind] = y_pred_task
y_pred = np.squeeze(y_pred)
return y_pred
Model.register_model_type("multitask_deep_regressor", MultiTaskDNN)
Model.register_model_type("multitask_deep_classifier", MultiTaskDNN)
class SingleTaskDNN(MultiTaskDNN):
"""
Abstract base class for different ML models.
"""
def __init__(self, model_type, task_types, model_params, initialize_raw_model=True):
super(SingleTaskDNN, self).__init__(model_type, task_types, model_params,
initialize_raw_model)
Model.register_model_type("singletask_deep_regressor", SingleTaskDNN)
Model.register_model_type("singletask_deep_classifier", SingleTaskDNN)
def to_one_hot(y):
"""Transforms label vector into one-hot encoding.
model.add(Activation('relu'))
model.add(MaxPooling3D(pool_size=(nb_pool[2], nb_pool[2], nb_pool[2])))
model.add(Flatten())
# TODO(rbharath): If we change away from axis-size 32, this code will break.
# Eventually figure out a more general rule that works for all axis sizes.
model.add(Dense(16, init='normal'))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(1, init='normal'))
sgd = RMSprop(lr=learning_rate, decay=1e-6, momentum=0.9, nesterov=True)
print("About to compile model")
model.compile(loss=loss_function, optimizer=sgd)
self.raw_model = model
def fit_on_batch(self, X, y, w):
X = shuffle_data(X)
loss = self.raw_model.train_on_batch(X, y)
print("Loss: %f" % loss)
def predict_on_batch(self, X):
if len(np.shape(X)) != 5:
raise ValueError(
"Tensorial datatype must be of shape (n_samples, N, N, N, n_channels).")
X = shuffle_data(X)
y_pred = self.raw_model.predict_on_batch(X)
y_pred = np.squeeze(y_pred)
return y_pred
Model.register_model_type("convolutional_3D_regressor", DockingDNN)
for ind, task in enumerate(sorted_tasks):
task_type = self.task_types[task]
taskname = "task%d" % ind
if task_type == "classification":
# Class probabilities are predicted for classification outputs. Instead,
# output the most likely class.
y_pred_task = np.squeeze(
np.argmax(y_pred_dict[taskname], axis=1))
else:
y_pred_task = np.squeeze(y_pred_dict[taskname])
y_pred[:, ind] = y_pred_task
y_pred = np.squeeze(y_pred)
return y_pred
Model.register_model_type(MultiTaskDNN)
class SingleTaskDNN(MultiTaskDNN):
"""
Abstract base class for different ML models.
"""
def __init__(self, task_types, model_params, initialize_raw_model=True):
super(SingleTaskDNN,
self).__init__(task_types,
model_params,
initialize_raw_model=initialize_raw_model)
Model.register_model_type(SingleTaskDNN)
"""
Makes predictions on dataset.
"""
# Sets batch_size which the default impl in Model expects
#TODO(enf/rbharath): This is kludgy. Fix later.
if "batch_size" not in self.model_params.keys():
self.model_params["batch_size"] = 32
return super(SklearnModel, self).predict(X)
def save(self, out_dir):
"""Saves sklearn model to disk using joblib."""
super(SklearnModel, self).save(out_dir)
save_to_disk(self.raw_model, self.get_model_filename(out_dir))
def load(self, model_dir):
"""Loads sklearn model from joblib file on disk."""
self.raw_model = load_from_disk(Model.get_model_filename(model_dir))
Model.register_model_type(SklearnModel)
#TODO(enf/rbharath): deprecate the following if __init__.py functions as planned.
'''
Model.register_model_type("logistic", SklearnModel)
Model.register_model_type("rf_classifier", SklearnModel)
Model.register_model_type("rf_regressor", SklearnModel)
Model.register_model_type("linear", SklearnModel)
Model.register_model_type("ridge", SklearnModel)
Model.register_model_type("lasso", SklearnModel)
Model.register_model_type("lasso_lars", SklearnModel)
Model.register_model_type("elastic_net", SklearnModel)
'''
for (X, y, _, _) in numpy_dataset.itershards():
Xs.append(X)
ys.append(y)
X = np.concatenate(Xs)
y = np.concatenate(ys).ravel()
self.raw_model.fit(X, y)
def predict_on_batch(self, X):
"""
Makes predictions on given batch of new data.
"""
return self.raw_model.predict(X)
def save(self, out_dir):
"""Saves sklearn model to disk using joblib."""
super(SklearnModel, self).save(out_dir)
save_to_disk(self.raw_model, self.get_model_filename(out_dir))
def load(self, model_dir):
"""Loads sklearn model from joblib file on disk."""
self.raw_model = load_from_disk(Model.get_model_filename(model_dir))
Model.register_model_type("logistic", SklearnModel)
Model.register_model_type("rf_classifier", SklearnModel)
Model.register_model_type("rf_regressor", SklearnModel)
Model.register_model_type("linear", SklearnModel)
Model.register_model_type("ridge", SklearnModel)
Model.register_model_type("lasso", SklearnModel)
Model.register_model_type("lasso_lars", SklearnModel)
Model.register_model_type("elastic_net", SklearnModel)
def reload(self): """Loads sklearn model from joblib file on disk.""" self.model_instance = load_from_disk(Model.get_model_filename(self.model_dir))
def load(self, model_dir):
"""Loads sklearn model from joblib file on disk."""
self.raw_model = load_from_disk(Model.get_model_filename(model_dir))
Makes predictions on dataset.
"""
# Sets batch_size which the default impl in Model expects
#TODO(enf/rbharath): This is kludgy. Fix later.
if "batch_size" not in self.model_params.keys():
self.model_params["batch_size"] = 32
return super(SklearnModel, self).predict(X)
def save(self, out_dir):
"""Saves sklearn model to disk using joblib."""
super(SklearnModel, self).save(out_dir)
save_to_disk(self.raw_model, self.get_model_filename(out_dir))
def load(self, model_dir):
"""Loads sklearn model from joblib file on disk."""
self.raw_model = load_from_disk(Model.get_model_filename(model_dir))
Model.register_model_type(SklearnModel)
#TODO(enf/rbharath): deprecate the following if __init__.py functions as planned.
'''
Model.register_model_type("logistic", SklearnModel)
Model.register_model_type("rf_classifier", SklearnModel)
Model.register_model_type("rf_regressor", SklearnModel)
Model.register_model_type("linear", SklearnModel)
Model.register_model_type("ridge", SklearnModel)
Model.register_model_type("lasso", SklearnModel)
Model.register_model_type("lasso_lars", SklearnModel)
Model.register_model_type("elastic_net", SklearnModel)
'''
def load(self, model_dir): """Loads sklearn model from joblib file on disk.""" self.raw_model = load_from_disk(Model.get_model_filename(model_dir))
nb_tasks = len(sorted_tasks)
y_pred = np.zeros((nb_samples, nb_tasks))
for ind, task in enumerate(sorted_tasks):
task_type = self.task_types[task]
taskname = "task%d" % ind
if task_type == "classification":
# Class probabilities are predicted for classification outputs. Instead,
# output the most likely class.
y_pred_task = np.squeeze(np.argmax(y_pred_dict[taskname], axis=1))
else:
y_pred_task = np.squeeze(y_pred_dict[taskname])
y_pred[:, ind] = y_pred_task
y_pred = np.squeeze(y_pred)
return y_pred
Model.register_model_type(MultiTaskDNN)
class SingleTaskDNN(MultiTaskDNN):
"""
Abstract base class for different ML models.
"""
def __init__(self, task_types, model_params, initialize_raw_model=True):
super(SingleTaskDNN, self).__init__(task_types, model_params,
initialize_raw_model=initialize_raw_model)
Model.register_model_type(SingleTaskDNN)
def to_one_hot(y):
"""Transforms label vector into one-hot encoding.
Turns y into vector of shape [n_samples, 2] (assuming binary labels).
