-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathmain_arrayjob_using_precalculated_M.py
241 lines (187 loc) · 8.44 KB
/
main_arrayjob_using_precalculated_M.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
from sys import argv, exit
import os
import numpy as np
import copy
from math import sqrt
from sklearn import preprocessing, metrics
from pairwisemkl.learner.compute_a_regression import *
from pairwisemkl.learner.optimize_kernel_weights import *
from pairwisemkl.learner.cg_kron_rls import CGKronRLS
try:
i_out = int(argv[1])
except:
exit()
data_path = './drug_response_data'
new_path = data_path + "/ArrayJob_results"
if i_out==0 and not os.path.exists(new_path):
os.makedirs(new_path)
# Drug kernels
# Read file names of drug kernels
fn_kd = open(data_path + '/Drug_kernels/Drug_kernel_file_names.txt', 'r')
kd_file_names = fn_kd.readlines()
fn_kd.close()
kd_file_names = [x.split('\n')[0] for x in kd_file_names]
# Prepare a list of drug kernels
kd_list = []
for kd in kd_file_names:
f_kd = open(data_path + '/Drug_kernels/' + kd, 'r')
kd_list.append(np.loadtxt(f_kd))
f_kd.close()
# Cell line kernels
# Read file names of cell line kernels
fn_kc = open(data_path + '/Cell_line_kernels/Cell_kernel_file_names.txt', 'r')
kc_file_names = fn_kc.readlines()
fn_kc.close()
kc_file_names = [x.split('\n')[0] for x in kc_file_names]
kc_list = []
# Prepare a list of cell line kernels
for kc in kc_file_names:
f_kc = open(data_path + '/Cell_line_kernels/' + kc, 'r')
kc_list.append(np.loadtxt(f_kc))
f_kc.close()
# Number of pairwise kernels
P = len(kd_list)*len(kc_list)
# Generate pairwise kernel ids
kd_ids, kc_ids = np.unravel_index(np.arange(P), (len(kd_list),len(kc_list)), order = 'C')
# Assemble matrix M precomputed using array jobs
M = np.empty([P,P]); M[:] = np.NAN
for row in range(P):
f_m = open(data_path + '/M/M__row_' + str(row) + '.txt')
m = np.loadtxt(f_m)
f_m.close()
M[row, row:P] = m[row:P]
M[row:P, row] = m[row:P]
# Labels
# Read matrix with drug responses in cancer cell lines
f = open(data_path + '/Labels.txt', 'r') # rows: drugs, columns: cell lines
Y = np.loadtxt(f)
f.close()
# Number of drugs and cell lines
n_d, n_c = Y.shape
# Labels in the vector form
y_vec = Y.ravel(order = 'C')
# Create indicies
ids = np.arange(n_d*n_c)
drug_ids, cell_ids = np.unravel_index(ids, (n_d,n_c), order = 'C')
# Remove missing values (if any) from the label vector as well as the corresponding ids
ids_known = ~np.isnan(y_vec)
y_vec_known = y_vec[ids_known]
drug_ids_known = drug_ids[ids_known]
cell_ids_known = cell_ids[ids_known]
# Values for the regularization parameter \lambda
# (to be optimized in the nested CV)
regparam = [10.**x for x in range(-5, 1)]
# CV (10 outer folds, 3 inner folds)
# Each outer CV loop is run as a separate array job
# Vector where predicted drug responses will be stored
y_pred_outer_vec = np.empty([len(ids[ids_known])]); y_pred_outer_vec[:] = np.NAN
# Read pre-defined outer folds used in the experiments presented in the pairwiseMKL paper
outer_folds = np.loadtxt(data_path + '/Folds/outer_folds.txt').astype(int)
# Outer CV loop
print('Outer loop ' + str(i_out+1) + '\n')
test_ids = np.array(np.where(outer_folds==i_out)).squeeze()
train_ids = np.array(np.where(outer_folds!=i_out)).squeeze()
# Test data
y_test = y_vec_known[test_ids]
drug_ids_test = drug_ids_known[test_ids]
cell_ids_test = cell_ids_known[test_ids]
# Training data
# Training labels in the vector form
y_train = y_vec_known[train_ids]
# Training labels in the matrix form
Y_train = copy.deepcopy(Y)
Y_train[drug_ids_test,cell_ids_test] = np.nan
drug_ids_train = drug_ids_known[train_ids]
cell_ids_train = cell_ids_known[train_ids]
# Stage 1 of determining pairwise kernel weights
# Compute vector a needed to optimize kernel weights
a = compute_a_regression(kd_list, kc_list, Y_train)
# Optimize pairwise kernel weights
k_weights_outer = optimize_kernel_weights(a, M)
# Stage 2 of pairwise model training
# Find pairwise kernels with weights different from 0 (>10**-3).
ix = np.where(k_weights_outer[0,:] > 10**-3)[0]
# Corresponding kernel weights
w = k_weights_outer[0,ix]/sum(k_weights_outer[0,ix])
kd_list_selected = []
kc_list_selected = []
for i_p in range(len(w)):
kd_list_selected.append(kd_list[kd_ids[ix[i_p]]])
kc_list_selected.append(kc_list[kc_ids[ix[i_p]]])
# Inner CV loop
rmse_inner = np.empty([3, len(regparam)]); rmse_inner[:] = np.NAN
# Read pre-defined inner folds used in the experiments presented in pairwiseMKL paper
inner_folds = np.loadtxt(data_path + '/Folds/inner_folds_outer%d.txt'%i_out).astype(int)
for i_in in range(3):
print(' Inner loop ' + str(i_in+1))
inner_test_ids = np.array(np.where(inner_folds==i_in)).squeeze()
inner_train_ids = np.array(np.where((inner_folds!=i_in) & (inner_folds!=-1))).squeeze()
y_test_inner = y_vec_known[inner_test_ids]
drug_ids_test_inner = drug_ids_known[inner_test_ids]
cell_ids_test_inner = cell_ids_known[inner_test_ids]
y_train_inner = y_vec_known[inner_train_ids]
drug_ids_train_inner = drug_ids_known[inner_train_ids]
cell_ids_train_inner = cell_ids_known[inner_train_ids]
# Find optimal \lambda
for i_param in range(len(regparam)):
# Training
learner = CGKronRLS(K1 = kd_list_selected,
K2 = kc_list_selected,
weights = w.tolist(),
Y = y_train_inner,
label_row_inds = [drug_ids_train_inner for i in range(len(w))],
label_col_inds = [cell_ids_train_inner for i in range(len(w))],
regparam = regparam[i_param],
maxiter = 400)
# Prediction
pred_inner = learner.predict(kd_list_selected, kc_list_selected, [drug_ids_test_inner for i in range(len(w))], [cell_ids_test_inner for i in range(len(w))])
# RMSE
rmse_inner[i_in,i_param] = sqrt(((y_test_inner - pred_inner) ** 2).mean(axis=0))
# \lambda with the lowest RMSE
model = regparam[np.argmin(np.mean(rmse_inner, axis=0))]
# Model training with selected \lambda
learner = CGKronRLS(K1 = kd_list_selected,
K2 = kc_list_selected,
weights = w.tolist(),
Y = y_train,
label_row_inds = [drug_ids_train for i in range(len(w))],
label_col_inds = [cell_ids_train for i in range(len(w))],
regparam = model,
maxiter = 400)
# Prediction
y_pred_outer_vec[test_ids] = learner.predict(kd_list_selected, kc_list_selected, [drug_ids_test for i in range(len(w))], [cell_ids_test for i in range(len(w))])
# RMSE
rmse_outer = sqrt(((y_test - y_pred_outer_vec[test_ids]) ** 2).mean(axis=0))
# Pearson correlation
pearson_outer = np.corrcoef(y_test, y_pred_outer_vec[test_ids])[0,1]
# F1 score
y_test_binary = copy.deepcopy(y_test)
y_test_binary = preprocessing.binarize(y_test_binary.reshape(1,-1), threshold=5, copy=False)[0]
y_pred_binary = copy.deepcopy(y_pred_outer_vec[test_ids])
y_pred_binary = preprocessing.binarize(y_pred_binary.reshape(1,-1), threshold=5, copy=False)[0]
f1_outer = metrics.f1_score(y_test_binary, y_pred_binary)
# Predicted drug responses
np.savetxt(new_path + '/y_pred_vec__fold_' + str(i_out+1) + '.txt', y_pred_outer_vec, delimiter='\t')
# Test ids
np.savetxt(new_path + '/test_ids__fold_' + str(i_out+1) + '.txt', test_ids, delimiter='\t')
# In the below files, each row corresponds to the result from a single outer CV fold
# RMSE
np.savetxt(new_path + '/RMSE__fold_' + str(i_out+1) + '.txt', np.asarray([rmse_outer]))
# Pearson correlation
np.savetxt(new_path + '/Pearson_correlation__fold_' + str(i_out+1) + '.txt', np.asarray([pearson_outer]))
# F1 score
np.savetxt(new_path + '/F1_score__fold_' + str(i_out+1) + '.txt', np.asarray([f1_outer]))
# Optimal value of the regularization parameter \lambda
np.savetxt(new_path + '/selected_lambda__fold_' + str(i_out+1) + '.txt', np.asarray([model]))
# Pairwise kernel weights
np.savetxt(new_path + '/pairwise_kernel_weights__fold_' + str(i_out+1) + '.txt', k_weights_outer, delimiter='\t')
# Vector a
np.savetxt(new_path +'/a__fold_' + str(i_out+1) + '.txt', a, delimiter='\t')
# File names of the corresponding pairwise kernels, in the same order as in the file
# with pairwise kernel weights
if i_out==0:
thefile = open(new_path + '/pairwise_kernel_names.txt', 'w')
for i in range(P):
thefile.write("%s_KRONECKER_%s\t" %(kd_file_names[kd_ids[i]], kc_file_names[kc_ids[i]]))
thefile.close()
print('\nSuccess!')