-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathGenerateGeneLatentFeatures.py
180 lines (145 loc) · 7.81 KB
/
GenerateGeneLatentFeatures.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
# -*- coding: utf-8 -*-
"""
@author: SURAJ
"""
import pickle
import random
import numpy as np
import keras_tuner as kt
import pandas as pd
from sklearn.model_selection import KFold
from utils.SurvivalUtils import km_plot
from sklearn.model_selection import train_test_split
import tensorflow as tf
from utils.DataLoader import load_data
from keras.callbacks import EarlyStopping
from models.GenePathwayAutoencoder import GeneCox_model
from sklearn.preprocessing import MinMaxScaler
"""# K-Fold cross validation to fit the model
"""
SEED = 20
random.seed(SEED)
batch_size = 26
search_epochs = 20
epochs = 100
max_trial = 3
exec_per_trial = 1
rnaseq_scaled_df, pathway_mask, y_event, y_survival_time = load_data('gene-pathwaymask')
lusc_rnaseq_df, pathway_mask, lusc_y_event, lusc_y_survival_time = load_data('lusc_gene-pathwaymask')
luad_rnaseq_df, pathway_mask, luad_y_event, luad_y_survival_time = load_data('luad_gene-pathwaymask')
lusc_rnaseq_df = pd.DataFrame(MinMaxScaler().fit_transform(lusc_rnaseq_df), index=lusc_rnaseq_df.index)
luad_rnaseq_df = pd.DataFrame(MinMaxScaler().fit_transform(luad_rnaseq_df), index=luad_rnaseq_df.index)
latent_dim = 500
original_dim = rnaseq_scaled_df.shape[1]
with tf.device('/GPU:0'):
genecox_model = GeneCox_model(original_dim, latent_dim, pathway_mask)
# evaluate a model using k-fold cross-validation
def nested_cross_validate_model(feature_df, y_event, y_time, n_folds=5):
c_index, best_model_list, pred_list, latent_feature, time, event, luad_latent_features, lusc_latent_features = list(
), list(), list(), list(), list(), list(), list(), list()
sample_id = feature_df.index
feature_df.reset_index(drop=True, inplace=True)
y_event.reset_index(drop=True, inplace=True)
y_time.reset_index(drop=True, inplace=True)
# prepare cross validation
k_outer_fold = KFold(n_folds, shuffle=True, random_state=SEED)
# enumerate splits
i = 1
# outer fold
for train_test_ix, validation_idx in k_outer_fold.split(feature_df):
train_test_rna = feature_df.loc[train_test_ix].reset_index(drop=True)
validation_rna = feature_df.loc[validation_idx].reset_index(drop=True)
train_test_event = y_event.loc[train_test_ix].reset_index(drop=True)
validation_test_event = y_event.loc[validation_idx].reset_index(drop=True)
train_test_time = y_time.loc[train_test_ix].reset_index(drop=True)
validation_time = y_time.loc[validation_idx].reset_index(drop=True)
val_sample_id = sample_id[validation_idx]
train_test_rna = MinMaxScaler().fit_transform(train_test_rna)
validation_rna = MinMaxScaler().fit_transform(validation_rna)
print(train_test_rna)
tuner = kt.BayesianOptimization(
hypermodel=genecox_model.build_model,
objective=kt.Objective("val_cindex_metric", direction="max"),
max_trials=max_trial,
overwrite=True,
executions_per_trial=exec_per_trial,
directory="G_XVAE-hyper_dir",
project_name='Gene_XVAE-Cox{}'.format(i),
)
# stratified train test split
train_rna, test_rna, train_event, test_event, train_time, test_time = \
train_test_split(train_test_rna, train_test_event, train_test_time, test_size=0.15,
stratify=train_test_event, shuffle=True, random_state=SEED)
tuner.search(train_rna, [train_event, train_time],
batch_size=batch_size,
epochs=search_epochs,
validation_data=(test_rna, [test_event, test_time]))
best_hps = tuner.get_best_hyperparameters(1)[0]
# best_hyperparameters.append(best_hps.values)
print(best_hps.values)
print(tuner.results_summary())
with tf.device('/CPU:0'):
# best_model = tuner.get_best_models()[0]
early_stopping = EarlyStopping(monitor='val_cindex_metric_1', patience=30, min_delta=0.0001, mode='max')
best_model = tuner.hypermodel.build(best_hps)
history = best_model.fit(train_test_rna, [train_test_event, train_test_time],
batch_size=batch_size,
epochs=epochs,
validation_data=(validation_rna, [validation_test_event, validation_time]),
callbacks=[early_stopping])
c_idx = best_model.evaluate(validation_rna, [validation_test_event, validation_time], verbose=0)
print('cindex for fold_{} is : {}'.format(i, c_idx))
# perform high risk and low risk analysis from the best model
z, pred_risk = best_model.encoder.predict(validation_rna)
# generate latent features of LUAD and LUSC images for OOD validation
luad_z, luad_pred_risk = best_model.encoder.predict(luad_rnaseq_df)
lusc_z, lusc_pred_risk = best_model.encoder.predict(lusc_rnaseq_df)
# best_model.encoder.save('SavedObjects/H_VAE/GeneSparseEncoder_{}.h5'.format(i))
# best_model.decoder.save('SavedObjects/H_VAE/GeneSparseDecoder_{}.h5'.format(i))
lat_feat = pd.DataFrame(z, index=val_sample_id)
luad_lat_feat = pd.DataFrame(luad_z, index=luad_rnaseq_df.index)
lusc_lat_feat = pd.DataFrame(lusc_z, index=lusc_rnaseq_df.index)
c_index.append(c_idx)
pred_list.append(pred_risk)
time.append(validation_time)
event.append(validation_test_event)
latent_feature.append(lat_feat)
luad_latent_features.append(luad_lat_feat)
lusc_latent_features.append(lusc_lat_feat)
best_model_list.append(best_model)
print('loop validation c_index: ', c_index)
i += 1
return c_index, latent_feature, pred_list, time, event, best_model_list, luad_latent_features, lusc_latent_features
c_index, latent_feature, pred_risk_list, time, event, best_model_list, luad_latent_features, lusc_latent_features = nested_cross_validate_model(
rnaseq_scaled_df, y_event, y_survival_time, n_folds=5)
print('mean c_index', np.mean(c_index))
print('std c_index', np.std(c_index))
# %%
for i in range(len(best_model_list)):
best_model_list[i].encoder.save('SavedObjects/H_VAE/GeneSparseEncoder_{}.h5'.format(i))
best_model_list[i].decoder.save('SavedObjects/H_VAE/GeneSparseDecoder_{}.h5'.format(i))
# %%
lusc_z, pred_risk = best_model_list[4].encoder.predict(lusc_rnaseq_df)
lusc_lat_feat = pd.DataFrame(lusc_z, index=lusc_rnaseq_df.index)
lusc_lat_feat.to_csv('Results/tcga_lusc_cox_gene_latent.csv')
luad_z, pred_risk = best_model_list[4].encoder.predict(luad_rnaseq_df)
luad_lat_feat = pd.DataFrame(luad_z, index=luad_rnaseq_df.index)
luad_lat_feat.to_csv('Results/tcga_luad_cox_gene_latent.csv')
# %%
lat_feature_df = pd.DataFrame(columns=latent_feature[0].columns)
for lat in latent_feature:
print(type(lat))
lat_feature_df = pd.concat([lat_feature_df, lat])
lat_feature_df.sort_index().to_csv('Results/cox_gene_latent.csv')
# %%
hazard_list = np.array(pred_risk_list).reshape(-1)
time_list = np.array(time).reshape(-1)
event_list = np.array(event).reshape(-1)
km_plot(pd.DataFrame(event_list, columns=['Survival Status']), pd.DataFrame(time_list, columns=['survival_time']), hazard_list)
median_risk = np.median(hazard_list)
df = pd.DataFrame()
df['Hazard'] = np.array(hazard_list)
df['Risk'] = np.where(df.Hazard > median_risk, 1, 0)
risk_group = pd.concat([pd.DataFrame(event_list, columns=['Survival Status']), pd.DataFrame(
time_list, columns=['survival_time']), df['Hazard'], df['Risk']], axis=1)
risk_group.to_csv('Results/GeneOnlySurvivalRiskGroup.csv')