-
Notifications
You must be signed in to change notification settings - Fork 1
/
MOGONET.py
296 lines (245 loc) · 12.4 KB
/
MOGONET.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
294
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Thu Nov 17 09:39:37 2022
@author: ltoure
"""
import os
import pandas as pd
import numpy as np
import torch
import torch.nn.functional as F
from models import init_model_dict, init_optim
from utils import one_hot_tensor, cal_sample_weight, gen_adj_mat_tensor, gen_test_adj_mat_tensor, cal_adj_mat_parameter
from train_test import train_test, prepare_trte_data, train_epoch, test_epoch, gen_trte_adj_mat
from sklearn.metrics import confusion_matrix, accuracy_score, balanced_accuracy_score, f1_score, precision_score, recall_score, roc_auc_score
import seaborn as sns
import copy
#### Define device ###
cuda = True if torch.cuda.is_available() else False
#####################################
# Load Data #
#####################################
omics1 = pd.read_csv("path", sep=",", index_col=0)
outcomes = pd.read_csv("path", index_col=0)
omics2 = pd.read_csv("path", sep=",", index_col=0)
TestIndex100 = pd.read_csv("path", sep=" ")
#############################################
# Data transformation #
#############################################
test_i = TestIndex100
test_i = test_i - 1
X = omics1
y = omics2
labels = outcomes
data_folder = 'ROSMAP'
view_list = [1,2] ## The parameter for the number of omics you have. Here I using 2 types of omics. for 3 types : view_list = [1,2,3] and so on.
num_epoch_pretrain = 2500
num_epoch = 1
lr_e_pretrain = 1e-3
lr_e = 5e-4
lr_c = 1e-3
# Choose a data depending of the type of classifiation task. ROSMAP for binary classification and BRCA for multi-class classification.
if data_folder == 'ROSMAP':
num_class = 2
if data_folder == 'BRCA':
num_class = 5
#all_imp = {}
all_auc = []
all_acc = []
all_bacc = []
all_f1 = []
all_precision = []
all_recall = []
all_se = []
all_sp = []
# I train the in split data. I choose 100 splis previous done based on the response data.
#If you want to train the model in more than 1 epoch and you don't have a GPU (It's not a god idea my dear).
for i in range(test_i.shape[1]):
X_tr = X.iloc[np.setdiff1d(np.arange(X.shape[0]), test_i.iloc[:,i]),:]
X_te = X.iloc[test_i.iloc[:,i],:]
y_tr = y.iloc[np.setdiff1d(np.arange(y.shape[0]), test_i.iloc[:,i]),:]
y_te = y.iloc[test_i.iloc[:,i],:]
labels_tr = labels.iloc[np.setdiff1d(np.arange(y.shape[0]), test_i.iloc[:,i]),:]
labels_te = labels.iloc[test_i.iloc[:,i],:]
labels_tr = labels_tr.replace("X1",1.0000+00)
labels_tr = labels_tr.replace("X2",0.0000+00)
labels_te = labels_te.replace("X2",0.0000+00)
labels_te = labels_te.replace("X1",1.0000+00)
feature_name_X_tr=X_tr.columns
feature_name_y=y_tr.columns
X_tr.to_csv("~/MOGONET/ROSMAP/1_tr.csv", header=False, index=False)
X_te.to_csv("~/MOGONET/ROSMAP/1_te.csv", header=False, index=False)
y_tr.to_csv("~MOGONET/ROSMAP/2_tr.csv", header=False, index=False)
y_te.to_csv("~/MOGONET/ROSMAP/2_te.csv", header=False, index=False)
labels_tr.to_csv("~/MOGONET/ROSMAP/labels_tr.csv", header=False, index=False)
labels_te.to_csv("~/MOGONET/ROSMAP/labels_te.csv", header=False, index=False)
feature_name_X_tr=pd.DataFrame(feature_name_X_tr)
feature_name_X_tr.to_csv("~/MOGONET/ROSMAP/1_featname.csv", header=False, index=False)
feature_name_y=pd.DataFrame(feature_name_y)
feature_name_y.to_csv("~/MOGONET/ROSMAP/2_featname.csv", header=False, index=False)
#####################################
# Model train #
#####################################
test_inverval = 50
num_view = len(view_list)
dim_hvcdn = pow(num_class,num_view)
if data_folder == 'ROSMAP':
adj_parameter = 2
dim_he_list = [200,200,100]
if data_folder == 'BRCA':
adj_parameter = 10
dim_he_list = [400,400,200]
data_tr_list, data_trte_list, trte_idx, labels_trte = prepare_trte_data(data_folder, view_list)
labels_tr_tensor = torch.LongTensor(labels_trte[trte_idx["tr"]])
onehot_labels_tr_tensor = one_hot_tensor(labels_tr_tensor, num_class)
sample_weight_tr = cal_sample_weight(labels_trte[trte_idx["tr"]], num_class)
sample_weight_tr = torch.FloatTensor(sample_weight_tr)
if cuda:
labels_tr_tensor = labels_tr_tensor.cuda()
onehot_labels_tr_tensor = onehot_labels_tr_tensor.cuda()
sample_weight_tr = sample_weight_tr.cuda()
adj_tr_list, adj_te_list = gen_trte_adj_mat(data_tr_list, data_trte_list, trte_idx, adj_parameter)
dim_list = [x.shape[1] for x in data_tr_list]
model_dict = init_model_dict(num_view, num_class, dim_list, dim_he_list, dim_hvcdn)
for m in model_dict:
if cuda:
model_dict[m].cuda()
print("\nPretrain GCNs...")
optim_dict = init_optim(num_view, model_dict, lr_e_pretrain, lr_c)
for epoch in range(num_epoch_pretrain):
train_epoch(data_tr_list, adj_tr_list, labels_tr_tensor,
onehot_labels_tr_tensor, sample_weight_tr, model_dict, optim_dict, train_VCDN=False)
# print("\nPretrain GCNs: Epoch {:d}".format(epoch))
print("\nTraining...")
optim_dict = init_optim(num_view, model_dict, lr_e, lr_c)
for epoch in range(num_epoch+1):
train_epoch(data_tr_list, adj_tr_list, labels_tr_tensor,
onehot_labels_tr_tensor, sample_weight_tr, model_dict, optim_dict)
if epoch % test_inverval == 0:
te_prob = test_epoch(data_trte_list, adj_te_list, trte_idx["te"], model_dict)
print("\nTest: Epoch {:d}".format(epoch))
if num_class == 2:
print("Test ACC: {:.3f}".format(accuracy_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))))
print("Test F1: {:.3f}".format(f1_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))))
print("Test AUC: {:.3f}".format(roc_auc_score(labels_trte[trte_idx["te"]], te_prob[:,1])))
te_prob = test_epoch(data_trte_list, adj_te_list, trte_idx["te"], model_dict)
auc = roc_auc_score(labels_trte[trte_idx["te"]], te_prob[:,1])
f1 = f1_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))
acc = accuracy_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))
bacc = balanced_accuracy_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))
precision = precision_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))
recall = recall_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))
cm = confusion_matrix(labels_trte[trte_idx["te"]], te_prob.argmax(1))
sensitivity = cm[0,0]/(cm[0,0]+cm[0,1])
specificity = cm[1,1]/(cm[1,0]+cm[1,1])
all_auc.append(auc)
all_f1.append(f1)
all_acc.append(acc)
all_bacc.append(bacc)
all_precision.append(precision)
all_recall.append(recall)
all_se.append(sensitivity)
all_sp.append(specificity)
else:
print("Test ACC: {:.3f}".format(accuracy_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))))
print("Test F1 weighted: {:.3f}".format(f1_score(labels_trte[trte_idx["te"]], te_prob.argmax(1), average='weighted')))
print("Test F1 macro: {:.3f}".format(f1_score(labels_trte[trte_idx["te"]], te_prob.argmax(1), average='macro')))
print()
#####################################
# Feature Importance #
#####################################
data_tr_list, data_trte_list, trte_idx, labels_trte = prepare_trte_data(data_folder, view_list)
adj_tr_list, adj_te_list = gen_trte_adj_mat(data_tr_list, data_trte_list, trte_idx, adj_parameter)
featname_list = []
for v in view_list:
df = pd.read_csv(os.path.join(data_folder, str(v)+"_featname.csv"), header=None)
featname_list.append(df.values.flatten())
dim_list = [x.shape[1] for x in data_tr_list]
model_dict = init_model_dict(num_view, num_class, dim_list, dim_he_list, dim_hvcdn)
for m in model_dict:
if cuda:
model_dict[m].cuda()
#model_dict = load_model_dict(model_folder, model_dict)
te_prob = test_epoch(data_trte_list, adj_te_list, trte_idx["te"], model_dict)
if num_class == 2:
f1 = f1_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))
else:
f1 = f1_score(labels_trte[trte_idx["te"]], te_prob.argmax(1), average='macro')
feat_imp_list = []
for i in range(len(featname_list)):
feat_imp = {"feat_name":featname_list[i]}
feat_imp['imp'] = np.zeros(dim_list[i])
for j in range(dim_list[i]):
feat_tr = data_tr_list[i][:,j].clone()
feat_trte = data_trte_list[i][:,j].clone()
data_tr_list[i][:,j] = 0
data_trte_list[i][:,j] = 0
adj_tr_list, adj_te_list = gen_trte_adj_mat(data_tr_list, data_trte_list, trte_idx, adj_parameter)
te_prob = test_epoch(data_trte_list, adj_te_list, trte_idx["te"], model_dict)
if num_class == 2:
f1_tmp = f1_score(labels_trte[trte_idx["te"]], te_prob.argmax(1))
else:
f1_tmp = f1_score(labels_trte[trte_idx["te"]], te_prob.argmax(1), average='macro')
feat_imp['imp'][j] = (f1-f1_tmp)*dim_list[i]
data_tr_list[i][:,j] = feat_tr.clone()
data_trte_list[i][:,j] = feat_trte.clone()
feat_imp_list.append(pd.DataFrame(data=feat_imp))
featimp_list_list = []
featimp_list_list.append(copy.deepcopy(feat_imp_list))
#def summarize_imp_feat(featimp_list_list, topn=30):
num_rep = len(featimp_list_list)
num_view = len(featimp_list_list[0])
df_tmp_list = []
for v in range(num_view):
df_tmp = copy.deepcopy(featimp_list_list[0][v])
df_tmp['omics'] = np.ones(df_tmp.shape[0], dtype=int)*v
df_tmp_list.append(df_tmp.copy(deep=True))
df_featimp = pd.concat(df_tmp_list).copy(deep=True)
for r in range(1,num_rep):
for v in range(num_view):
df_tmp = copy.deepcopy(featimp_list_list[r][v])
df_tmp['omics'] = np.ones(df_tmp.shape[0], dtype=int)*v
df_featimp = df_featimp.append(df_tmp.copy(deep=True), ignore_index=True)
df_featimp_top = df_featimp.groupby(['feat_name', 'omics'])['imp'].sum()
df_featimp_top = df_featimp_top.reset_index()
df_featimp_top = df_featimp_top.sort_values(by='imp',ascending=False)
df_featimp_top = df_featimp_top.iloc[:30]
print('{:}\t{:}'.format('Rank','Feature name'))
for i in range(len(df_featimp_top)):
print('{:}\t{:}'.format(i+1,df_featimp_top.iloc[i]['feat_name']))
#####################################
# Performance #
#####################################
feat_imp=sns.barplot(data=df_featimp_top, x="imp", y="feat_name")
fig = feat_imp.get_figure()
fig.savefig('Feature_imp_MOGONET.png')
df_featimp_top.to_csv("/home/ldap/ltoure/MOGONET/Featimp.csv")
hist_auc = sns.displot(all_auc)
hist_auc.savefig('AUC_MOGONET_100_splits.png')
hist_acc = sns.displot(all_acc)
hist_acc.savefig('Accuracy_MOGONET_100_splits.png')
hist_bacc = sns.displot(all_bacc)
hist_bacc.savefig('BalAccuracy_MOGONET_100_splits.png')
hist_f1 = sns.displot(all_f1)
hist_f1.savefig('F1_MOGONET_100_splits.png')
hist_precision = sns.displot(all_precision)
hist_precision.savefig('Precision_MOGONET_100_splits.png')
hist_recall = sns.displot(all_recall)
hist_recall.savefig('Recall_MOGONET_100_splits.png')
hist_se = sns.displot(all_se)
hist_se.savefig('Sensitivity_MOGONET_100_splits.png')
hist_sp = sns.displot(all_sp)
hist_sp.savefig('Specificity_MOGONET_100_splits.png')
mean_perf = {'auc' : np.mean(all_auc),
'acc' : np.mean(all_acc),
'bacc' : np.mean(all_bacc),
'f1' : np.mean(all_f1),
'prec' : np.mean(all_precision),
'recall' : np.mean(all_recall),
'sens' : np.mean(all_se),
'spec' : np.mean(all_sp)}
mean_perf_out = open("MeanPerf_MOGONET_100_splits", "w")
for key in mean_perf.keys():
mean_perf_out.write(key + " " + str(mean_perf[key]) + "\n")
mean_perf_out.close()