-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathexplainable_models.py
360 lines (273 loc) · 14.5 KB
/
explainable_models.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
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
import numpy as np
import pandas as pd
from config import Config
from matplotlib import pyplot as plt
from DBM_toolbox.data_manipulation.data_utils import pickle_objects, unpickle_objects, merge_and_clean
from sklearn.metrics import roc_curve, auc
import seaborn as sns
import glob
import logging
from sklearn.tree import DecisionTreeClassifier
from sklearn.svm import LinearSVC
from sklearn.linear_model import LinearRegression, LogisticRegression
from sklearn.dummy import DummyClassifier
from sklearn import tree
from sklearn.base import clone
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, balanced_accuracy_score, classification_report, confusion_matrix, roc_curve, auc
from mpl_toolkits.mplot3d import Axes3D
import warnings
warnings.filterwarnings("ignore", category=RuntimeWarning)
def partial_dependency(model, X, feature1_idx, feature2_idx, fixed_feature_idx, fixed_feature_value,
grid_resolution=100):
feature1_values = np.linspace(X.iloc[:, feature1_idx].min(), X.iloc[:, feature1_idx].max(), grid_resolution)
feature2_values = np.linspace(X.iloc[:, feature2_idx].min(), X.iloc[:, feature2_idx].max(), grid_resolution)
xx, yy = np.meshgrid(feature1_values, feature2_values)
pdp_data = np.c_[xx.ravel(), yy.ravel()]
X_temp = X.copy()
pdp_results = np.zeros((grid_resolution, grid_resolution))
X_temp.iloc[:, fixed_feature_idx] = fixed_feature_value
for i, f1_val in enumerate(feature1_values):
for j, f2_val in enumerate(feature2_values):
X_temp.iloc[:, feature1_idx] = f1_val
X_temp.iloc[:, feature2_idx] = f2_val
pdp_results[j, i] = np.mean(model.predict(X_temp))
return feature1_values, feature2_values, pdp_results
old_dataset = unpickle_objects('FINAL_preprocessed_data_2023-02-16-10-30-39-935233.pkl')
# old_final_results = unpickle_objects('FINAL_results_2023-02-20-12-35-21-577662.pkl')
config = Config("testall/config_explain.yaml")
#
#
# logging.info("Reading data")
# data, ActAreas, ic50s, dose_responses = config.read_data()
#
# logging.info("Filtering data")
# filtered_data, filters = config.filter_data(data)
#####
#
# filtered_data = unpickle_objects('REV_filtered_2024-09-15-05-25-10-904618.pkl')
#
# logging.info("Selecting subsets for feature engineering")
# selected_subset = config.select_subsets(filtered_data)
#
# logging.info("Engineering features")
# engineered_features = config.engineer_features(filtered_data)
#
# logging.info("Merging engineered features")
# engineered_data = filtered_data.merge_with(engineered_features)
#
# logging.info("Quantizing targets")
# quantized_data = config.quantize(engineered_data, target_omic="DRUGS", ic50s=ic50s)
#
# final_data = quantized_data.normalize().optimize_formats()
# config.save(to_save=final_data, name="FINAL_explain_preprocessed_data")
final_data = unpickle_objects('FINAL_preprocessed_data_2023-02-16-10-30-39-935233.pkl')
missing_data = final_data.dataframe.loc[:, final_data.dataframe.isnull().any(axis=0)]
######
# logging.info("Getting optimized models")
# trained_models = config.get_models(dataset=final_data, method="standard")
# config.save(to_save=trained_models, name="FINAL_explain_pre-models")
########### from here
#######################
######################
final_data = unpickle_objects("FINAL_explain_preprocessed_data_2024-06-25-21-07-10-041579.pkl")
trained_models = unpickle_objects("FINAL_explain_pre-models_2024-06-25-22-59-37-868197.pkl")
predictors = {
'PD-0325901': ['ETV4', 'SPRY2', 'RP11-93B14.5', 'ETV5', 'ZNF502', 'SPRY1', 'TOR4A', 'CMTM7', 'DUSP6'],
'PD0325901': ['ETV4', 'SPRY2', 'RP11-93B14.5', 'ETV5', 'ZNF502', 'SPRY1', 'TOR4A', 'CMTM7', 'DUSP6'],
'Irinotecan': ['SLFN11', 'RP11-177C12.1', 'HNRNPA1', 'KHDC1'],
'Erlotinib': ['RP11-902B17.1', 'RP11-47I22.1', 'IFI27', 'CORO2A', 'TSTD1'],
'Lapatinib': ['RP11-902B17.1', 'RP11-47I22.1', 'PRKCH', 'ARHGAP27', 'DYRK3', 'SYTL1', 'GPX3', 'ADORA1', 'GPR135'],
}
training_labels = old_dataset.dataframe.index
new_labels = final_data.dataframe.index
validation_labels = [x for x in new_labels if x not in training_labels]
train_dataset, test_dataset = final_data.split(train_index=training_labels, test_index=validation_labels)
targets_list = ['PD-0325901_ActArea', 'Irinotecan_ActArea', 'Erlotinib_ActArea', 'Lapatinib_ActArea']# final_data.to_pandas(omic='DRUGS').columns
predictions = {}
conf_matrices = {}
models = {}
accuracies = pd.DataFrame(index=targets_list, columns=[''])
for target_name in targets_list:
print(f"Target: {target_name}")
this_predictors = predictors[target_name.split('_Act')[0]]
colnames_predictors = [x for y in this_predictors for x in train_dataset.dataframe.columns if x.startswith(y+'_')]
y_train = train_dataset.dataframe[target_name]
X_train = train_dataset.dataframe.loc[:, colnames_predictors]
y_test = test_dataset.dataframe[target_name]
X_test = test_dataset.dataframe.loc[:, colnames_predictors]
y_train_clean = y_train[(y_train == 0) | (y_train == 1)]
X_train_clean = X_train.loc[y_train_clean.index]
y_test_clean = y_test[(y_test == 0) | (y_test == 1)]
X_test_clean = X_test.loc[y_test_clean.index]
print(f"train: {X_train_clean.shape[0]}")
print(f"test: {X_test_clean.shape[0]}")
# Initialize the classifiers
linear_classifier = LinearRegression()
logistic_classifier = LogisticRegression(random_state=42)
decision_tree = DecisionTreeClassifier(random_state=42)
rule_based_classifier = DummyClassifier(strategy="most_frequent", random_state=42)
# Train the classifiers
linear_classifier.fit(X_train_clean, y_train_clean)
logistic_classifier.fit(X_train_clean, y_train_clean)
decision_tree.fit(X_train_clean, y_train_clean)
rule_based_classifier.fit(X_train_clean, y_train_clean)
# Test the classifiers and get accuracy
linear_accuracy = balanced_accuracy_score(y_test_clean, linear_classifier.predict(X_test_clean)>0.5)
logistic_accuracy = balanced_accuracy_score(y_test_clean, logistic_classifier.predict(X_test_clean))
decision_tree_accuracy = balanced_accuracy_score(y_test_clean, decision_tree.predict(X_test_clean))
rule_based_accuracy = balanced_accuracy_score(y_test_clean, rule_based_classifier.predict(X_test_clean))
# Print classifier performance
print("Linear classifier accuracy:", linear_accuracy)
print("Logistic classifier accuracy:", logistic_accuracy)
print("Decision tree accuracy:", decision_tree_accuracy)
print("Rule-based classifier accuracy:", rule_based_accuracy)
classifier_names = ["linear", "logistic", "decision_tree", "dummy"]
accuracies = [linear_accuracy, logistic_accuracy, decision_tree_accuracy, rule_based_accuracy]
best_classifier_name = classifier_names[np.argmax(accuracies)]
clf = {'linear': LinearRegression(),
'logistic': LogisticRegression(random_state=42),
'decision_tree': DecisionTreeClassifier(random_state=42),
'dummy': DummyClassifier(strategy="most_frequent", random_state=42),
}[best_classifier_name]
print(f"Best classifier: {best_classifier_name}, with balanced accuracy {accuracies[np.argmax(accuracies)]}")
clf.fit(X_train_clean, y_train_clean)
########### Now try with GDSC
df_drugs = pd.read_excel('data/Drugs_AUC_4drugs.xlsx')
df_pivoted = df_drugs.pivot(index=['Cell Line Name', 'Cosmic ID'], columns='Drug Name', values='AUC')
df_pivoted.columns = [f"{drug}_ActArea" for drug in df_pivoted.columns]
df_pivoted.reset_index(inplace=True)
print(df_pivoted.head())
df_pivoted.to_excel('Drugs_AUC_reformatted.xlsx', index=False)
df_genes = pd.read_csv('data/Cell_line_RMA_proc_basalExp_transposed.tsv', sep='\t')
df_genes = df_genes.iloc[1:].reset_index(drop=True)
df_genes.columns = ['Cosmic ID'] + list(df_genes.columns[1:])
df_merged = pd.merge(df_genes, df_pivoted, how='inner', on='Cosmic ID')
df_merged_clean = df_merged.loc[:, ~df_merged.columns.str.startswith('Unnamed')]
actarea_columns = df_merged_clean.filter(like='ActArea')
def quantize_column(col):
return pd.qcut(col, q=3, labels=[0, 0.5, 1])
df_merged_clean[actarea_columns.columns] = actarea_columns.apply(quantize_column)
train_df, test_df = train_test_split(df_merged_clean, test_size=0.2, random_state=42)
targets_list = ['PD0325901_ActArea', 'Irinotecan_ActArea', 'Erlotinib_ActArea',
'Lapatinib_ActArea']
predictions = {}
conf_matrices = {}
models = {}
accuracies = pd.DataFrame(index=targets_list, columns=[''])
for target_name in targets_list:
print(f"Target: {target_name}")
this_predictors = predictors[target_name.split('_Act')[0]]
colnames_predictors = [x for y in this_predictors for x in train_df.columns if
x.startswith(y)]
y_train = train_df[target_name]
X_train = train_df.loc[:, colnames_predictors]
y_test = test_df[target_name]
X_test = test_df.loc[:, colnames_predictors]
y_train_clean = y_train[(y_train == 0) | (y_train == 1)]
X_train_clean = X_train.loc[y_train_clean.index]
y_test_clean = y_test[(y_test == 0) | (y_test == 1)]
X_test_clean = X_test.loc[y_test_clean.index]
print(f"train: {X_train_clean.shape[0]}")
print(f"test: {X_test_clean.shape[0]}")
# Initialize the classifiers
linear_classifier = LinearRegression()
logistic_classifier = LogisticRegression(random_state=42)
decision_tree = DecisionTreeClassifier(random_state=42)
rule_based_classifier = DummyClassifier(strategy="most_frequent", random_state=42)
# Train the classifiers
linear_classifier.fit(X_train_clean, y_train_clean)
logistic_classifier.fit(X_train_clean, y_train_clean)
decision_tree.fit(X_train_clean, y_train_clean)
rule_based_classifier.fit(X_train_clean, y_train_clean)
# Test the classifiers and get accuracy
linear_accuracy = balanced_accuracy_score(y_test_clean, linear_classifier.predict(X_test_clean) > 0.5)
logistic_accuracy = balanced_accuracy_score(y_test_clean, logistic_classifier.predict(X_test_clean))
decision_tree_accuracy = balanced_accuracy_score(y_test_clean, decision_tree.predict(X_test_clean))
rule_based_accuracy = balanced_accuracy_score(y_test_clean, rule_based_classifier.predict(X_test_clean))
# Print classifier performance
print("Linear classifier accuracy:", linear_accuracy)
print("Logistic classifier accuracy:", logistic_accuracy)
print("Decision tree accuracy:", decision_tree_accuracy)
print("Rule-based classifier accuracy:", rule_based_accuracy)
classifier_names = ["linear", "logistic", "decision_tree", "dummy"]
accuracies = [linear_accuracy, logistic_accuracy, decision_tree_accuracy, rule_based_accuracy]
best_classifier_name = classifier_names[np.argmax(accuracies)]
clf = {'linear': LinearRegression(),
'logistic': LogisticRegression(random_state=42),
'decision_tree': DecisionTreeClassifier(random_state=42),
'dummy': DummyClassifier(strategy="most_frequent", random_state=42),
}[best_classifier_name]
print(f"Best classifier: {best_classifier_name}, with balanced accuracy {accuracies[np.argmax(accuracies)]}")
clf.fit(X_train_clean, y_train_clean)
try:
y_pred = clf.predict_proba(X_test_clean)[:, 1]
except:
y_pred = clf.predict(X_test_clean)
predictions[target_name + '_truth'] = y_test_clean
predictions[target_name] = y_pred
fpr, tpr, thresholds = roc_curve(y_test_clean, y_pred)
distances = np.sqrt(np.square(1 - tpr) + np.square(fpr))
min_distance_index = np.argmin(distances)
best_threshold = thresholds[min_distance_index]
plt.subplots(figsize=(8, 8))
hfont = {'fontname': 'Times New Roman'}
roc_auc = auc(fpr, tpr)
plt.plot(fpr, tpr, linewidth=3, label=f"AUC: {round(roc_auc, 3)}")
plt.plot([0, 1], [0, 1], color="black", linestyle="--")
plt.plot(fpr[min_distance_index], tpr[min_distance_index], marker='o', color='blue', label='Best Threshold = %0.2f' % best_threshold, markersize=6)
plt.xlim([0.0, 1.05])
plt.ylim([0.0, 1.05])
plt.xlabel("False Positive Rate", fontsize=25, **hfont)
plt.ylabel("True Positive Rate", fontsize=25, **hfont)
plt.title(f"expl_{best_classifier_name}_{target_name.split('_')[0]}_n={len(y_pred)}", fontsize=25, **hfont)
plt.legend(loc="lower right", fontsize=20)
plt.savefig(f'FINAL_redo_Expl_ROC_{target_name}.tif')
plt.close()
# plt.subplots(figsize=(20, 20))
# feat_names = [x.split('_')[0] for x in clf.feature_names_in_]
# tree.plot_tree(clf, feature_names=feat_names)
# plt.savefig(f'FINAL_ExplDT_struct_{target_name}.tif')
# plt.close()
#
# predictions['model'] = clf.copy()
#
# accuracy = accuracy_score(y_test_clean, y_pred)
# ba = balanced_accuracy_score(y_test_clean, y_pred)
# report = classification_report(y_test_clean, y_pred)
y_pred = y_pred > best_threshold
conf_matrix = confusion_matrix(y_test_clean, y_pred)
conf_matrices[target_name] = conf_matrix
models[target_name] = clone(clf)
feature_names_all = clf.feature_names_in_
try:
feature_score_all = np.abs(clf.coef_)
except:
feature_score_all = clf.feature_importances_
if len(feature_score_all.shape) > 1:
feature_score_all = feature_score_all.ravel()
top3_indices = np.argsort(feature_score_all)[-3:]
top3_features = [feature_names_all[i] for i in top3_indices]
feature1_idx = top3_indices[2]
feature2_idx = top3_indices[1]
fixed_feature_idx = top3_indices[0]
percentiles = [10, 25, 50, 75, 90]
fixed_feature_values = np.percentile(X_train_clean.loc[:, top3_features[2]], percentiles)
# Set up the figure and the subplots
fig, axs = plt.subplots(1, len(percentiles), figsize=(18, 4))
for i, fixed_value in enumerate(fixed_feature_values):
feature1_values, feature2_values, pdp_results = partial_dependency(clf, X_train_clean, feature1_idx, feature2_idx,
fixed_feature_idx, fixed_value)
# Create a contour plot
x, y = np.meshgrid(feature1_values, feature2_values)
cs = axs[i].contourf(x, y, pdp_results, cmap=plt.cm.coolwarm)
# Customize the subplot
axs[i].set_title(f'{percentiles[i]}th percentile of {top3_features[0].split("_")[0]}')
axs[i].set_xlabel(top3_features[2].split("_")[0])
axs[i].set_ylabel(top3_features[1].split("_")[0])
# Add a colorbar to the figure
fig.subplots_adjust(right=0.8)
cbar_ax = fig.add_axes([0.85, 0.15, 0.05, 0.7])
fig.colorbar(cs, cax=cbar_ax)
plt.savefig(f'FINAL_redo_Expl_PDP_{target_name}.tif')
plt.close()