forked from ecemecemk/HoliLoc
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathHoliLoc_Train_Reproduce.py
293 lines (210 loc) · 13.2 KB
/
HoliLoc_Train_Reproduce.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
# Suppress warnings
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
import argparse
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
import tensorflow as tf
from tensorflow.keras.models import Sequential, Model
from tensorflow.keras.layers import (
Concatenate,
Dense,
Dropout,
BatchNormalization,
Activation,
Flatten,
Conv2D,
MaxPooling2D,
)
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import ReduceLROnPlateau
from sklearn.preprocessing import MinMaxScaler
seed_value = 29
tf.random.set_seed(41)
scaler = MinMaxScaler(feature_range=(0, 1))
def parse_args():
parser = argparse.ArgumentParser(description='Holiloc Model Training Script')
parser.add_argument('--train_data', type=str, help='Path to the training data CSV file', required=True)
parser.add_argument('--img_feature_vectors', type=str, help='Path to the image feature vectors file', required=True)
parser.add_argument('--sequence_embeddings', type=str, help='Path to the sequence embeddings file', required=True)
parser.add_argument('--ppi_embeddings', type=str, help='Path to the PPI embeddings file', required=True)
parser.add_argument('--output_model', type=str, help='Path to save the trained Holiloc model', required=True)
return parser.parse_args()
def load_data(train_data_path, img_feature_vectors_path, sequence_embeddings_path, ppi_embeddings_path):
train = pd.read_csv(train_data_path)
y = np.array(train.drop(['Cluster_ID', 'UNIPROT', 'CELLLINE', 'IMAGE_URL'], axis=1))
X_img = np.load(img_feature_vectors_path)
X_seq = np.load(sequence_embeddings_path, allow_pickle=True)
X_seq = np.vstack(X_seq)
X_inta = np.load(ppi_embeddings_path)
X_train_seq, X_test_seq, y_train_seq, y_test_seq = train_test_split(
X_seq, y, random_state=104, test_size=0.2, shuffle=True
)
X_train_seq_normalized = scaler.fit_transform(X_train_seq)
X_test_seq_normalized = scaler.fit_transform(X_test_seq)
X_train_seq = X_train_seq_normalized
X_test_seq = X_test_seq_normalized
X_train_int, X_test_int, y_train_int, y_test_int = train_test_split(
X_inta, y, random_state=104, test_size=0.2, shuffle=True
)
X_train_int_normalized = scaler.fit_transform(X_train_int)
X_test_int_normalized = scaler.fit_transform(X_test_int)
X_train_int = X_train_int_normalized
X_test_int = X_test_int_normalized
X_train_img, X_test_img, y_train_img, y_test_img = train_test_split(
X_img, y, random_state=104, test_size=0.2, shuffle=True
)
return ( X_train_img, y_train_img, X_test_img, y_test_img, X_train_seq, y_train_seq, X_test_seq, y_test_seq, X_train_int, y_train_int, X_test_int, y_test_int)
def create_image_model():
image_model = Sequential()
image_model.add(Conv2D(filters=16, kernel_size=(5, 5), activation="relu", input_shape=(224, 224, 3), name='imagemodel/conv2d_1'))
image_model.add(MaxPooling2D(pool_size=(2, 2), name='imagemodel/max_pooling2d_1'))
image_model.add(Dropout(rate=0.3,seed=seed_value, name='imagemodel/dropout_1'))
image_model.add(Conv2D(filters=32, kernel_size=(5, 5), activation='relu', name='imagemodel/conv2d_2'))
image_model.add(MaxPooling2D(pool_size=(2, 2), name='imagemodel/max_pooling2d_2'))
image_model.add(Dropout(rate=0.5,seed=seed_value, name='imagemodel/dropout_2'))
image_model.add(Conv2D(filters=64, kernel_size=(5, 5), activation="relu", name='imagemodel/conv2d_3'))
image_model.add(MaxPooling2D(pool_size=(2, 2), name='imagemodel/max_pooling2d_3'))
image_model.add(Dropout(rate=0.3,seed=seed_value,name='imagemodel/dropout_3'))
image_model.add(Conv2D(filters=64, kernel_size=(5, 5), activation='relu', name='imagemodel/conv2d_4'))
image_model.add(MaxPooling2D(pool_size=(2, 2), name='imagemodel/max_pooling2d_4'))
image_model.add(Dropout(rate=0.5,seed=seed_value, name='imagemodel/dropout_4'))
image_model.add(Flatten(name='imagemodel/flatten'))
image_model.add(Dense(128, activation='relu', name='imagemodel/dense_1'))
image_model.add(Dropout(rate=0.3,seed=seed_value, name='imagemodel/dropout_5'))
image_model.add(Dense(64, activation='relu', name='imagemodel/dense_2'))
image_model.add(Dropout(rate=0.3,seed=seed_value, name='imagemodel/dropout_6'))
image_model.add(Dense(22, activation='sigmoid', name='imagemodel/output_layer'))
return image_model
def train_image_model(X_train_img, y_train_img, X_test_img, y_test_img):
model_img = create_image_model()
initial_learning_rate = 1e-4
optimizer = Adam(learning_rate=initial_learning_rate)
model_img.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])
reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=5, min_lr=1e-16)
epochs = 30
history_image = model_img.fit(X_train_img, y_train_img, validation_data=(X_test_img, y_test_img),
epochs=epochs, callbacks=[reduce_lr])
def create_model_sequence():
sequence_model = Sequential(name='sequence_model')
sequence_model.add(Dense(units=256, input_shape=(1024,), name='sequence_model/dense_1'))
sequence_model.add(BatchNormalization(name='sequence_model/batch_norm_1'))
sequence_model.add(Activation('relu', name='sequence_model/activation_1'))
sequence_model.add(Dropout(rate=0.1, seed=seed_value, name='sequence_model/dropout_1'))
sequence_model.add(Dense(units=128, name='sequence_model/dense_2'))
sequence_model.add(Activation('relu', name='sequence_model/activation_2'))
sequence_model.add(Dropout(rate=0.2, seed=seed_value, name='sequence_model/dropout_2'))
sequence_model.add(Dense(units=64, name='sequence_model/dense_3'))
sequence_model.add(Activation('relu', name='sequence_model/activation_3'))
sequence_model.add(Dropout(rate=0.1, seed=seed_value, name='sequence_model/dropout_3'))
sequence_model.add(Dense(units=32, name='sequence_model/dense_4'))
sequence_model.add(Activation('relu', name='sequence_model/activation_4'))
sequence_model.add(Dropout(rate=0.3, seed=seed_value, name='sequence_model/dropout_4'))
sequence_model.add(Dense(units=22, activation='sigmoid', name='sequence_model/output_layer'))
return sequence_model
def train_sequence_model(X_train_seq, y_train_seq, X_test_seq, y_test_seq):
model_seq = create_model_sequence()
initial_learning_rate = 1e-4
optimizer = Adam(learning_rate=initial_learning_rate)
model_seq.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])
reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=5, min_lr=1e-16)
epochs = 20
history_sequence = model_seq.fit(X_train_seq, y_train_seq, validation_data=(X_test_seq, y_test_seq),
epochs=epochs, callbacks=[reduce_lr])
def create_interactome_model():
interactome_model = Sequential(name='interactome_model')
interactome_model.add(Dense(units=128, input_shape=(224,), name='interactome_model/dense_1'))
interactome_model.add(BatchNormalization(name='interactome_model/batch_norm_1'))
interactome_model.add(Activation('relu', name='interactome_model/activation_1'))
interactome_model.add(Dropout(rate=0.4,seed=seed_value, name='interactome_model/dropout_1'))
interactome_model.add(Dense(units=64, name='interactome_model/dense_2'))
interactome_model.add(Activation('relu', name='interactome_model/activation_2'))
interactome_model.add(Dropout(rate=0.5,seed=seed_value,name='interactome_model/dropout_2'))
interactome_model.add(Dense(units=64, name='interactome_model/dense_3'))
interactome_model.add(Activation('relu', name='interactome_model/activation_3'))
interactome_model.add(Dropout(rate=0.1,seed=seed_value, name='interactome_model/dropout_3'))
interactome_model.add(Dense(units=32, name='interactome_model/dense_4'))
interactome_model.add(Activation('relu', name='interactome_model/activation_4'))
interactome_model.add(Dropout(rate=0.1,seed=seed_value, name='interactome_model/dropout_4'))
interactome_model.add(Dense(units=32, name='interactome_model/dense_5'))
interactome_model.add(BatchNormalization(name='interactome_model/batch_norm_2'))
interactome_model.add(Activation('relu', name='interactome_model/activation_5'))
interactome_model.add(Dropout(rate=0.1,seed=seed_value, name='interactome_model/dropout_5'))
interactome_model.add(Dense(units=22, activation='sigmoid', name='interactome_model/output'))
return interactome_model
def train_interactome_model(X_train_int, y_train_int, X_test_int, y_test_int):
model_inta = create_interactome_model()
initial_learning_rate = 1e-5
optimizer = Adam(learning_rate=initial_learning_rate)
model_inta.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])
reduce_lr = ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=5, min_lr=1e-16)
epochs = 200
history_interactome = model_inta.fit(X_train_int, y_train_int, validation_data=(X_test_int, y_test_int),
epochs=epochs, callbacks=[reduce_lr])
def create_fusion_model(model_img, model_seq, model_inta):
model_img=model_img
model_seq=model_seq
model_inta=model_inta
image_input = model_img.input
sequence_input = model_seq.input
interactome_input = model_inta.input
image_representation = model_img.get_layer('imagemodel/dense_2').output
image_representation = Dense(128, activation='relu', name='image_dense')(image_representation)
sequence_representation = model_seq.get_layer('sequence_model/dense_3').output
sequence_representation = Dense(128, activation='relu', name='sequence_dense')(sequence_representation)
interactome_representation = model_inta.get_layer('interactome_model/dense_2').output
interactome_representation = Dense(128, activation='relu', name='interactome_dense')(interactome_representation)
merged_representation = Concatenate()([image_representation, sequence_representation, interactome_representation])
dense1_fusion = Dense(units=1024, activation='relu', name='dense1_fusion')(merged_representation)
bn1_fusion = BatchNormalization(name='bn1_fusion')(dense1_fusion)
dropout1_fusion = Dropout(0.2, seed=seed_value, name='dropout1_fusion')(bn1_fusion)
dense2_fusion = Dense(units=1024, activation='relu', name='dense2_fusion')(dropout1_fusion)
bn2_fusion = BatchNormalization(name='bn2_fusion')(dense2_fusion)
dropout2_fusion = Dropout(0.1, seed=seed_value, name='dropout2_fusion')(bn2_fusion)
dense3_fusion = Dense(units=1024, activation='relu', name='dense3_fusion')(dropout2_fusion)
bn3_fusion = BatchNormalization(name='bn3_fusion')(dense3_fusion)
dropout3_fusion = Dropout(0.1, seed=seed_value, name='dropout3_fusion')(bn3_fusion)
dense4_fusion = Dense(units=448, activation='relu', name='dense4_fusion')(dropout3_fusion)
bn4_fusion = BatchNormalization(name='bn4_fusion')(dense4_fusion)
dropout4_fusion = Dropout(0.2, seed=seed_value, name='dropout4_fusion')(bn4_fusion)
dense5_fusion = Dense(units=704, activation='relu', name='dense5_fusion')(dropout4_fusion)
bn5_fusion = BatchNormalization(name='bn5_fusion')(dense5_fusion)
dropout5_fusion = Dropout(0.2, seed=seed_value, name='dropout5_fusion')(bn5_fusion)
fusion_output = Dense(units=22, activation='sigmoid', name='fusion_output1')(dropout5_fusion)
fusion_model = Model(inputs=[image_input, sequence_input, interactome_input], outputs=fusion_output)
optimizer = tf.keras.optimizers.Adam(learning_rate=1e-4)
fusion_model.compile(optimizer=optimizer, loss='binary_crossentropy', metrics=['accuracy'])
return fusion_model
def train_fusion_model(fusion_model, X_train_img, X_train_seq, X_train_int, y_train_seq, X_test_img, X_test_seq,
X_test_int, y_test_seq):
history_fused =fusion_model.fit([X_train_img, X_train_seq, X_train_int], y_train_seq, epochs=25, batch_size=32, validation_data=([X_test_img, X_test_seq, X_test_int], y_test_seq))
# Save the entire fusion model
fusion_model.save('holiloc_local.h5')
def main():
args = parse_args()
# Load data
(
X_train_img, y_train_img, X_test_img, y_test_img,
X_train_seq, y_train_seq, X_test_seq, y_test_seq,
X_train_int, y_train_int, X_test_int, y_test_int
) = load_data(
args.train_data, args.img_feature_vectors, args.sequence_embeddings, args.ppi_embeddings
)
# Train image model
model_img = create_image_model()
train_image_model(X_train_img, y_train_img, X_test_img, y_test_img)
# Train sequence model
model_seq = create_model_sequence()
train_sequence_model(X_train_seq, y_train_seq, X_test_seq, y_test_seq)
# Train interactome model
model_inta = create_interactome_model()
train_interactome_model(X_train_int, y_train_int, X_test_int, y_test_int)
# Create and train the fusion model
fusion_model = create_fusion_model(model_img, model_seq, model_inta)
train_fusion_model(fusion_model, X_train_img, X_train_seq, X_train_int, y_train_seq, X_test_img, X_test_seq,
X_test_int, y_test_seq)
# Save the entire fusion model
fusion_model.save(args.output_model)
if __name__ == "__main__":
main()