-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsig_ratio.py
341 lines (242 loc) · 13.6 KB
/
sig_ratio.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
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
import numpy as np
import pandas as pd
import os
import warnings
import ggseg_python
import re
import pickle
import json
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from multiviewae import mVAE, weighted_mVAE, mmVAE, MoPoEVAE, mmJSD, mcVAE, JMVAE, DVCCA, AAE
#import neuroHarmonize
from neuroHarmonize import harmonizationLearn, harmonizationApply, loadHarmonizationModel, saveHarmonizationModel
import seaborn as sns
from functools import reduce
import warnings
warnings.filterwarnings("ignore")
from sklearn.model_selection import KFold, StratifiedKFold
import scipy
import statsmodels
from statsmodels.stats.multitest import fdrcorrection
from statsmodels.stats.multitest import fdrcorrection_twostage
from data_utils import *
from ADNI_KARI_merge_compare import *
from harmonize_combat import *
from dataloaders import *
from multimodal_VAE import *
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torch.autograd import Variable
from torch.nn.parameter import Parameter
from torchvision import datasets, transforms
import random
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import RobustScaler
from sklearn.preprocessing import QuantileTransformer
###########################################################
#-------------- Relevant functions ------------------------
###########################################################
def latent_deviations_mahalanobis_across_sig(cohort, train):
latent_dim = cohort[0].shape[1]
dists = calc_robust_mahalanobis_distance(cohort[0], train[0])
pvals = 1 - chi2.cdf(dists, latent_dim - 1)
return pvals
def calc_mahalanobis_distance(values, train_values):
covariance = np.cov(train_values, rowvar=False)
covariance_pm1 = np.linalg.matrix_power(covariance, -1)
centerpoint = np.mean(train_values, axis=0)
dists = np.zeros((values.shape[0],1))
for i in range(0, values.shape[0]):
p0 = values[i,:]
dist = (p0-centerpoint).T.dot(covariance_pm1).dot(p0-centerpoint)
dists[i,:] = dist
return dists
def calc_robust_mahalanobis_distance(values, train_values):
# fit a MCD robust estimator to data
robust_cov = MinCovDet(random_state=42).fit(train_values)
mahal_robust_cov = robust_cov.mahalanobis(values)
return mahal_robust_cov
def latent_count_ratio(pvals_cohort, pvals_holdout, model_type, cols):
thresh = 0.001
count_cohort = (pvals_cohort <= thresh).sum()
count_holdout = (pvals_holdout <= thresh).sum()
ratio_cohort = count_cohort/pvals_cohort.shape[0]
ratio_holdout = count_holdout/pvals_holdout.shape[0]
ratio = ratio_cohort/ratio_holdout
df = pd.DataFrame(np.array([key, count_cohort, ratio_cohort, count_holdout, ratio_holdout, ratio]).reshape(1,-1),
columns=cols)
return df
def feature_sig_ratio(cohort_recon, holdout_recon, train_recon, key, cols):
thresh = 0.001
dists = calc_robust_mahalanobis_distance(cohort_recon, train_recon)
pvals_cohort = 1 - chi2.cdf(dists, cohort_recon.shape[1] - 1)
dists = calc_robust_mahalanobis_distance(holdout_recon, train_recon)
pvals_holdout = 1 - chi2.cdf(dists, cohort_recon.shape[1] - 1)
count_cohort = (pvals_cohort <= thresh).sum()
count_holdout = (pvals_holdout <= thresh).sum()
ratio_cohort = count_cohort/pvals_cohort.shape[0]
ratio_holdout = count_holdout/pvals_holdout.shape[0]
ratio = ratio_cohort/ratio_holdout
df = pd.DataFrame(np.array([key, count_cohort, ratio_cohort, count_holdout, ratio_holdout, ratio]).reshape(1,-1),
columns=cols)
return df
###########################################################
#-------------- Main function ------------------------
###########################################################
a_n_merged_harm = pd.read_csv('./saved_dataframes/a_n_merged_harm.csv')
a_n_merged = pd.read_csv('./saved_dataframes/a_n_merged.csv')
## Harmonized ADNI and KARI --> a_n_merged_harm
## Concatenated non-harmonized ADNI and KARI --> a_n_merged
a_n_merged_adni = a_n_merged.loc[a_n_merged.dataset == 'ADNI'].reset_index(drop = True)
a_n_merged_kari = a_n_merged.loc[a_n_merged.dataset == 'KARI'].reset_index(drop = True)
CN_model_adni, CN_held_val_adni, only_CN_test_adni, X_test_org_adni = split_train_test(a_n_merged_adni, 0.15, 'ADNI')
CN_model_kari, CN_held_val_kari, only_CN_test_kari, X_test_org_kari = split_train_test(a_n_merged_kari, 0.15, 'KARI')
with open("./saved_dataframes/MRI_vol_cols", "rb") as fp:
MRI_vol_cols = pickle.load(fp)
with open("./saved_dataframes/amyloid_SUVR_cols", "rb") as fp:
amyloid_SUVR_cols = pickle.load(fp)
############# Results only for a single dataset (e.g. KARI) #############
m1_train = CN_model_kari[MRI_vol_cols]
m2_train = CN_model_kari[amyloid_SUVR_cols]
m1_holdout = CN_held_val_kari[MRI_vol_cols]
m2_holdout = CN_held_val_kari[amyloid_SUVR_cols]
m1_test = X_test_org_kari[MRI_vol_cols]
m2_test = X_test_org_kari[amyloid_SUVR_cols]
#-------- Scaling train, test and holdout based on parameters from train-----
mean_controls_m1 = np.mean(m1_train, axis=0)
sd_controls_m1 = np.std(m1_train, axis=0)
mean_controls_m2 = np.mean(m2_train, axis=0)
sd_controls_m2 = np.std(m2_train, axis=0)
m1_train_scaled = (m1_train - mean_controls_m1)/sd_controls_m1
m2_train_scaled = (m2_train - mean_controls_m2)/sd_controls_m2
m1_holdout_scaled = (m1_holdout - mean_controls_m1)/sd_controls_m1
m2_holdout_scaled = (m2_holdout - mean_controls_m2)/sd_controls_m2
m1_test_scaled = (m1_test - mean_controls_m1)/sd_controls_m1
m2_test_scaled = (m2_test - mean_controls_m2)/sd_controls_m2
train_concat = np.concatenate((m1_train_scaled, m2_train_scaled), axis=1)
holdout_concat = np.concatenate((m1_holdout_scaled, m2_holdout_scaled), axis=1)
test_concat = np.concatenate((m1_test_scaled, m2_test_scaled), axis=1)
model_dict = {'MoPoEVAE': './saved_models/MoPoEVAE',
'mmVAE': './saved_models/mmVAE',
'weighted_mVAE': './saved_models/weighted_mVAE',
'mVAE': './saved_models/mVAE',
'mmJSD': './saved_models/mmJSD'
'JMVAE': './saved_models/JMVAE',
'mcVAE': './saved_models/mcVAE',
'mri_only': './saved_models/mri_only',
'amyloid_only': './saved_models/amyloid_only',
'mri_amyloid_concat': './saved_models/mri_amyloid_concat',}
cols = ['model', 'count cohort', 'ratio cohort', 'count holdout', 'ratio holdout', 'significance ratio']
results_df = pd.DataFrame(columns=cols)
'''
model_path = './saved_models/MoPoEVAE'
model = torch.load(join(model_path, 'model.pkl'))
train_latents = model.predict_latents(m1_train_scaled, m2_train_scaled)
holdout_latents = model.predict_latents(m1_holdout_scaled, m2_holdout_scaled)
test_latents = model.predict_latents(m1_test_scaled, m2_test_scaled)
pvals_holdout = latent_deviations_mahalanobis_across_sig(holdout_latents, train_latents)
pvals_test = latent_deviations_mahalanobis_across_sig(test_latents, train_latents)
results = latent_count_ratio(pvals_test, pvals_holdout, key, cols)
results_df = pd.concat([results_df, results], axis=0)
'''
##########################################################
#-------signifcance ratio (latent mahalnobis distance------------------
##########################################################
cols = ['model', 'count cohort', 'ratio cohort', 'count holdout', 'ratio holdout', 'significance ratio']
latent_sig_results = pd.DataFrame(columns=cols)
for key, val in model_dict.items():
model = torch.load(join(val, 'model.pkl'))
print(key)
if key == 'mri_amyloid_concat':
train_latents = model.predict_latents(train_concat)
holdout_latents = model.predict_latents(holdout_concat)
test_latents = model.predict_latents(test_concat)
elif key == 'mri_only':
train_latents = model.predict_latents(m1_train_scaled)
holdout_latents = model.predict_latents(m1_holdout_scaled)
test_latents = model.predict_latents(m1_test_scaled)
elif key == 'amyloid_only':
train_latents = model.predict_latents(m2_train_scaled)
holdout_latents = model.predict_latents(m2_holdout_scaled)
test_latents = model.predict_latents(m2_test_scaled)
elif key == 'mmVAE':
train_latents = model.predict_latents(m1_train_scaled, m2_train_scaled)
train_latents = [np.mean([train_latents[0], train_latents[1]], axis=0)]
holdout_latents = model.predict_latents(m1_holdout_scaled, m2_holdout_scaled)
holdout_latents = [np.mean([holdout_latents[0], holdout_latents[1]], axis=0)]
test_latents = model.predict_latents(m1_test_scaled, m2_test_scaled)
test_latents = [np.mean([test_latents[0], test_latents[1]], axis=0)]
else:
train_latents = model.predict_latents(m1_train_scaled, m2_train_scaled)
holdout_latents = model.predict_latents(m1_holdout_scaled, m2_holdout_scaled)
test_latents = model.predict_latents(m1_test_scaled, m2_test_scaled)
pvals_holdout = latent_deviations_mahalanobis_across_sig(holdout_latents, train_latents)
pvals_test = latent_deviations_mahalanobis_across_sig(test_latents, train_latents)
results = latent_count_ratio(pvals_test, pvals_holdout, key, cols)
latent_sig_results = pd.concat([results_df, results], axis=0)
latent_sig_results.to_csv('./saved_models/latent_sig_results.csv'.format(date), header=True, index=False)
##########################################################
#-------signifcance ratio (Feature mahalnobis distance)------------------
##########################################################
def deviation(orig, recon, recon_type='abs'):
return np.sqrt((orig - recon)**2)
cols = ['model', 'count cohort', 'ratio cohort', 'count holdout', 'ratio holdout', 'significance ratio']
feature_sig_results = pd.DataFrame(columns=cols)
for key, val in model_dict.items():
model = torch.load(join(val, 'model.pkl'))
if key == 'mri_amyloid_concat':
train_recon = model.predict_reconstruction(train_concat)
holdout_recon = model.predict_reconstruction(holdout_concat)
test_recon = model.predict_reconstruction(test_concat)#
dev_holdout = deviation(holdout_concat, holdout_recon[0][0])
dev_test = deviation(test_concat, test_recon[0][0])
dev_train = deviation(train_concat, train_recon[0][0])
# results = deviation_sig_count(dev_test, dev_holdout, dev_train, key, cols, zscore=zscore)
# results_df = pd.concat([results_df, results], axis=0)
elif key == 'mri_only':
train_recon = model.predict_reconstruction(m1_train_scaled)
holdout_recon = model.predict_reconstruction(m1_holdout_scaled)
test_recon = model.predict_reconstruction(m1_test_scaled)
dev_holdout = deviation(m1_holdout_scaled, holdout_recon[0][0])
dev_test = deviation(m1_test_scaled, test_recon[0][0])
dev_train = deviation(m1_train_scaled, train_recon[0][0])
elif key == 'amyloid_only':
train_recon = model.predict_reconstruction(m2_train_scaled)
holdout_recon = model.predict_reconstruction(m2_holdout_scaled)
test_recon = model.predict_reconstruction(m2_test_scaled)
dev_holdout = deviation(m2_holdout_scaled, holdout_recon[0][0])
dev_test = deviation(m2_test_scaled, test_recon[0][0])
dev_train = deviation(m2_train_scaled, train_recon[0][0])
elif key == 'mmVAE':
train_recon = model.predict_reconstruction(m1_train_scaled, m2_train_scaled)
holdout_recon = model.predict_reconstruction(m1_holdout_scaled, m2_holdout_scaled)
test_recon = model.predict_reconstruction(m1_test_scaled, m2_test_scaled)
m1_dev_holdout = deviation(m1_holdout_scaled, holdout_recon[0][0])
m2_dev_holdout = deviation(m2_holdout_scaled, holdout_recon[1][1])
m1_dev_train = deviation(m1_train_scaled, train_recon[0][0])
m2_dev_train = deviation(m2_train_scaled, train_recon[1][1])
m1_dev_test = deviation(m1_test_scaled, test_recon[0][0])
m2_dev_test = deviation(m2_test_scaled, test_recon[1][1])
dev_test = np.concatenate((m1_dev_test, m2_dev_test), axis=1)
dev_holdout = np.concatenate((m1_dev_holdout, m2_dev_holdout), axis=1)
dev_train = np.concatenate((m1_dev_train, m2_dev_train), axis=1)
else:
train_recon = model.predict_reconstruction(m1_train_scaled, m2_train_scaled)
holdout_recon = model.predict_reconstruction(m1_holdout_scaled, m2_holdout_scaled)
test_recon = model.predict_reconstruction(m1_test_scaled, m2_test_scaled)
m1_dev_holdout = deviation(m1_holdout_scaled, holdout_recon[0][0])
m2_dev_holdout = deviation(m2_holdout_scaled, holdout_recon[0][1])
m1_dev_test = deviation(m1_test_scaled, test_recon[0][0])
m2_dev_test = deviation(m2_test_scaled, test_recon[0][1])
m1_dev_train = deviation(m1_train_scaled, train_recon[0][0])
m2_dev_train = deviation(m2_train_scaled, train_recon[0][1])
dev_test = np.concatenate((m1_dev_test, m2_dev_test), axis=1)
dev_holdout = np.concatenate((m1_dev_holdout, m2_dev_holdout), axis=1)
dev_train = np.concatenate((m1_dev_train, m2_dev_train), axis=1)
results = feature_sig_ratio(dev_test, dev_holdout, dev_train, key, cols)
feature_sig_results = pd.concat([results_df, results], axis=0)
feature_sig_results.to_csv('./saved_models/feature_sig_results.csv'.format(date), header=True, index=False)