In [2]:

    

from sklearn.model_selection import train_test_split
import pandas as pd
#set verbose to 0 so that it prints from the callback function

from keras.preprocessing import sequence
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation
from keras.layers import Embedding
from keras.layers import Conv1D, GlobalMaxPooling1D,GlobalAveragePooling1D


    

In [3]:

    

#load data

print('Loading data...')

#4 datasets
#CYP1A2_fixpka.smi
#CYP2C19_substrate_fixpka.smi
#CYP2D6_fixpka.smi
#CYP3A4_fixpka.smi

datasetFile='CYP1A2_fixpka.smi'
dataset = pd.read_csv(datasetFile, sep= ' ', header=None)
X_SMILES, y = dataset.iloc[:,:-1], dataset.iloc[:, -1]

#Library of unique characters. The first element is not used for padding
char_lib=["ZMY"]

#Find characters of sequences and build a library
for index, SMILESsequence in X_SMILES.iterrows():
    for letter in SMILESsequence[0]:
        if not letter in char_lib:            
            char_lib.append(letter)

X=[]
#SMILES sequence to an array
for index, SMILESsequence in X_SMILES.iterrows():
    sequenceArray=[]
    for letter in SMILESsequence[0]:
        sequenceArray.append(char_lib.index(letter))
    X.append(sequenceArray)
    
print('data loaded')


    

    




Loading data...
data loaded

In [133]:

    

#add zeros
X=sequence.pad_sequences(X)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=428)

max_features = len(char_lib)
maxlen = X_train.shape[1]


    

In [5]:

    

print(X_train.shape)
print(max_features)
print(maxlen)


    

    




(11922, 226)
43
226

In [192]:

    

import numpy as np
from keras.callbacks import Callback
from sklearn.metrics import roc_curve, auc
class CustomCallbacks(Callback):
    def __init__(self, filename="best_weight.hdf5", monitor='val_acc', patience=10, test_input=None, test_label=None):
        super(Callback, self).__init__()
        self.filename = filename
        self.monitor = monitor
        self.patience = patience
        self.test_input=test_input
        self.test_label=test_label

        self.wait = 0
        
        if 'acc' in self.monitor or 'val_acc' in self.monitor:
                self.monitor_op = np.greater
                self.best = -np.Inf
        else:
                self.monitor_op = np.less
                self.best = np.Inf
        self.losses = []
        self.acc = []
        self.val_acc =[]
        self.val_losses=[]             

    def on_epoch_end(self, epoch, logs={}):
        self.losses.append(logs.get('loss'))
        self.acc.append(logs.get('acc'))
        self.val_acc.append(logs.get('val_acc'))
        self.val_losses.append(logs.get('val_loss'))  
        
        y_score=self.model.predict(X_test)
        y_score_get_value=[]
        for item in y_score:
            y_score_get_value.append(item[0])    
        fpr, tpr, _ = roc_curve(y_test, y_score_get_value)
        val_auc=auc(fpr,tpr)

        print("acc: %0.3f" % logs.get('acc'), 
              ", loss: %0.3f" % logs.get('loss'),
              ", val_acc: %0.3f" % logs.get('val_acc'),
              ", val_loss: %0.3f" % logs.get('val_loss'),
              ", val_auc: %0.3f" % val_auc)        
        
        current = logs.get(self.monitor)
        if current is None:
            warnings.warn('Can save best model only with %s available, '
                                  'skipping.' % (self.monitor), RuntimeWarning)
        else:
            if self.monitor_op(current, self.best):
                print('Epoch %05d: %s improved from %0.5f to %0.5f,'
                                  ' saving model to %s'
                                  % (epoch, self.monitor, self.best,
                                     current, self.filename))
                self.best = current                       
                self.wait = 0
                self.model.save(self.filename, overwrite=True)
            else:
                print('Epoch %05d: %s did not improve' %  (epoch, self.monitor))
                if self.wait >= self.patience:                   
                    self.model.stop_training = True
                self.wait += 1

    def on_train_end(self, logs=None):
        if self.wait >= self.patience:
            print('Warning: early stopping')
        self.wait=0
        if 'acc' in self.monitor or 'val_acc' in self.monitor:
                self.monitor_op = np.greater
                self.best = -np.Inf
        else:
                self.monitor_op = np.less
                self.best = np.In
        
my_callbacks=CustomCallbacks(filename="CYP1A2_conv1.hdf5", monitor='val_acc', patience=np.Inf, test_input=X_test, test_label=y_test)


    

In [83]:

    

from keras.layers.pooling import MaxPooling1D
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.recurrent import LSTM 
from keras.layers import Flatten
from keras import optimizers
from keras.layers import Input
from keras.models import Model
from keras.layers.merge import Concatenate


    

In [195]:

    

inputs = Input(shape=(maxlen,))
embedding_1=Embedding(input_dim=max_features,output_dim=50, input_length=maxlen)(inputs)
dropout_1=Dropout(0.5)(embedding_1)

conv1d_kernel_size=[7,3,5]
conv1d_filters=[128,128,128]
my_globalmaxpoolings=[]
for i in range(0,len(conv1d_kernel_size)):
    conv1d_1=Conv1D(filters=conv1d_filters[i],
                    kernel_size=conv1d_kernel_size[i], 
                    padding='valid', 
                    strides=1)(dropout_1)
   
    leakyrelu_1=LeakyReLU(0.3)(conv1d_1) #better than relu with 0.004 more accuracy...
    dropout_conv=Dropout(0.5)(leakyrelu_1)
    globalmaxpooling1d_1=GlobalMaxPooling1D()(leakyrelu_1)
    my_globalmaxpoolings.append(globalmaxpooling1d_1)

concatenate_1=Concatenate()(my_globalmaxpoolings)
#model.add(MaxPooling1D(pool_size=2))

dense_1=Dense(500)(concatenate_1)
dropout_2=Dropout(0.5)(dense_1)
leakyrelu_hidden=LeakyReLU(0.3)(dropout_2)

dense2=Dense(1)(leakyrelu_hidden)
prediction=Activation('sigmoid')(dense2)


model = Model(inputs=inputs, outputs=prediction)
adam=optimizers.Adam(lr=0.001, beta_1=0.99, beta_2=0.999, epsilon=1e-08, decay=0.0)
model.compile(loss='binary_crossentropy',
              optimizer=adam,
              metrics=['accuracy'])

print(model.summary())


    

    




____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
input_33 (InputLayer)            (None, 226)           0                                            
____________________________________________________________________________________________________
embedding_63 (Embedding)         (None, 226, 50)       2150        input_33[0][0]                   
____________________________________________________________________________________________________
dropout_148 (Dropout)            (None, 226, 50)       0           embedding_63[0][0]               
____________________________________________________________________________________________________
conv1d_99 (Conv1D)               (None, 220, 128)      44928       dropout_148[0][0]                
____________________________________________________________________________________________________
conv1d_100 (Conv1D)              (None, 224, 128)      19328       dropout_148[0][0]                
____________________________________________________________________________________________________
conv1d_101 (Conv1D)              (None, 222, 128)      32128       dropout_148[0][0]                
____________________________________________________________________________________________________
leaky_re_lu_152 (LeakyReLU)      (None, 220, 128)      0           conv1d_99[0][0]                  
____________________________________________________________________________________________________
leaky_re_lu_153 (LeakyReLU)      (None, 224, 128)      0           conv1d_100[0][0]                 
____________________________________________________________________________________________________
leaky_re_lu_154 (LeakyReLU)      (None, 222, 128)      0           conv1d_101[0][0]                 
____________________________________________________________________________________________________
global_max_pooling1d_92 (GlobalM (None, 128)           0           leaky_re_lu_152[0][0]            
____________________________________________________________________________________________________
global_max_pooling1d_93 (GlobalM (None, 128)           0           leaky_re_lu_153[0][0]            
____________________________________________________________________________________________________
global_max_pooling1d_94 (GlobalM (None, 128)           0           leaky_re_lu_154[0][0]            
____________________________________________________________________________________________________
concatenate_27 (Concatenate)     (None, 384)           0           global_max_pooling1d_92[0][0]    
                                                                   global_max_pooling1d_93[0][0]    
                                                                   global_max_pooling1d_94[0][0]    
____________________________________________________________________________________________________
dense_108 (Dense)                (None, 500)           192500      concatenate_27[0][0]             
____________________________________________________________________________________________________
dropout_152 (Dropout)            (None, 500)           0           dense_108[0][0]                  
____________________________________________________________________________________________________
leaky_re_lu_155 (LeakyReLU)      (None, 500)           0           dropout_152[0][0]                
____________________________________________________________________________________________________
dense_109 (Dense)                (None, 1)             501         leaky_re_lu_155[0][0]            
____________________________________________________________________________________________________
activation_52 (Activation)       (None, 1)             0           dense_109[0][0]                  
====================================================================================================
Total params: 291,535
Trainable params: 291,535
Non-trainable params: 0
____________________________________________________________________________________________________
None

In [196]:

    

#data is not shuffled using validation_split. Have to use sk-learn to shuffle it first. 
model.fit(X_train, y_train,
          batch_size=100,
          epochs=100,
          callbacks=[my_callbacks],
          validation_data=(X_test,y_test),
          verbose=0)


    

    




acc: 0.687 , loss: 0.590 , val_acc: 0.733 , val_loss: 0.529 , val_auc: 0.816
Epoch 00000: val_acc improved from -inf to 0.73298, saving model to CYP1A2_conv1.hdf5
acc: 0.727 , loss: 0.536 , val_acc: 0.757 , val_loss: 0.512 , val_auc: 0.831
Epoch 00001: val_acc improved from 0.73298 to 0.75746, saving model to CYP1A2_conv1.hdf5
acc: 0.749 , loss: 0.514 , val_acc: 0.757 , val_loss: 0.497 , val_auc: 0.843
Epoch 00002: val_acc did not improve
acc: 0.762 , loss: 0.498 , val_acc: 0.761 , val_loss: 0.493 , val_auc: 0.847
Epoch 00003: val_acc improved from 0.75746 to 0.76115, saving model to CYP1A2_conv1.hdf5
acc: 0.764 , loss: 0.492 , val_acc: 0.741 , val_loss: 0.523 , val_auc: 0.856
Epoch 00004: val_acc did not improve
acc: 0.762 , loss: 0.494 , val_acc: 0.745 , val_loss: 0.518 , val_auc: 0.855
Epoch 00005: val_acc did not improve
acc: 0.773 , loss: 0.479 , val_acc: 0.783 , val_loss: 0.467 , val_auc: 0.860
Epoch 00006: val_acc improved from 0.76115 to 0.78296, saving model to CYP1A2_conv1.hdf5
acc: 0.772 , loss: 0.479 , val_acc: 0.783 , val_loss: 0.466 , val_auc: 0.861
Epoch 00007: val_acc improved from 0.78296 to 0.78296, saving model to CYP1A2_conv1.hdf5
acc: 0.784 , loss: 0.462 , val_acc: 0.777 , val_loss: 0.472 , val_auc: 0.865
Epoch 00008: val_acc did not improve
acc: 0.787 , loss: 0.458 , val_acc: 0.759 , val_loss: 0.505 , val_auc: 0.866
Epoch 00009: val_acc did not improve
acc: 0.792 , loss: 0.456 , val_acc: 0.793 , val_loss: 0.453 , val_auc: 0.871
Epoch 00010: val_acc improved from 0.78296 to 0.79336, saving model to CYP1A2_conv1.hdf5
acc: 0.793 , loss: 0.450 , val_acc: 0.762 , val_loss: 0.496 , val_auc: 0.871
Epoch 00011: val_acc did not improve
acc: 0.802 , loss: 0.440 , val_acc: 0.784 , val_loss: 0.482 , val_auc: 0.874
Epoch 00012: val_acc did not improve
acc: 0.795 , loss: 0.443 , val_acc: 0.795 , val_loss: 0.455 , val_auc: 0.875
Epoch 00013: val_acc improved from 0.79336 to 0.79537, saving model to CYP1A2_conv1.hdf5
acc: 0.804 , loss: 0.431 , val_acc: 0.787 , val_loss: 0.463 , val_auc: 0.877
Epoch 00014: val_acc did not improve
acc: 0.801 , loss: 0.428 , val_acc: 0.799 , val_loss: 0.449 , val_auc: 0.876
Epoch 00015: val_acc improved from 0.79537 to 0.79940, saving model to CYP1A2_conv1.hdf5
acc: 0.808 , loss: 0.425 , val_acc: 0.794 , val_loss: 0.455 , val_auc: 0.879
Epoch 00016: val_acc did not improve
acc: 0.808 , loss: 0.424 , val_acc: 0.802 , val_loss: 0.447 , val_auc: 0.880
Epoch 00017: val_acc improved from 0.79940 to 0.80174, saving model to CYP1A2_conv1.hdf5
acc: 0.812 , loss: 0.418 , val_acc: 0.792 , val_loss: 0.458 , val_auc: 0.880
Epoch 00018: val_acc did not improve
acc: 0.815 , loss: 0.411 , val_acc: 0.804 , val_loss: 0.445 , val_auc: 0.880
Epoch 00019: val_acc improved from 0.80174 to 0.80409, saving model to CYP1A2_conv1.hdf5
acc: 0.819 , loss: 0.406 , val_acc: 0.801 , val_loss: 0.447 , val_auc: 0.881
Epoch 00020: val_acc did not improve
acc: 0.811 , loss: 0.411 , val_acc: 0.782 , val_loss: 0.484 , val_auc: 0.882
Epoch 00021: val_acc did not improve
acc: 0.818 , loss: 0.412 , val_acc: 0.791 , val_loss: 0.457 , val_auc: 0.882
Epoch 00022: val_acc did not improve
acc: 0.819 , loss: 0.403 , val_acc: 0.802 , val_loss: 0.445 , val_auc: 0.882
Epoch 00023: val_acc did not improve
acc: 0.824 , loss: 0.401 , val_acc: 0.806 , val_loss: 0.447 , val_auc: 0.882
Epoch 00024: val_acc improved from 0.80409 to 0.80577, saving model to CYP1A2_conv1.hdf5
acc: 0.822 , loss: 0.397 , val_acc: 0.802 , val_loss: 0.449 , val_auc: 0.882
Epoch 00025: val_acc did not improve
acc: 0.820 , loss: 0.403 , val_acc: 0.808 , val_loss: 0.438 , val_auc: 0.883
Epoch 00026: val_acc improved from 0.80577 to 0.80778, saving model to CYP1A2_conv1.hdf5
acc: 0.819 , loss: 0.401 , val_acc: 0.795 , val_loss: 0.463 , val_auc: 0.882
Epoch 00027: val_acc did not improve
acc: 0.831 , loss: 0.382 , val_acc: 0.797 , val_loss: 0.460 , val_auc: 0.883
Epoch 00028: val_acc did not improve
acc: 0.826 , loss: 0.393 , val_acc: 0.807 , val_loss: 0.444 , val_auc: 0.883
Epoch 00029: val_acc did not improve
acc: 0.827 , loss: 0.394 , val_acc: 0.786 , val_loss: 0.480 , val_auc: 0.883
Epoch 00030: val_acc did not improve
acc: 0.829 , loss: 0.383 , val_acc: 0.808 , val_loss: 0.447 , val_auc: 0.884
Epoch 00031: val_acc did not improve
acc: 0.827 , loss: 0.386 , val_acc: 0.790 , val_loss: 0.479 , val_auc: 0.884
Epoch 00032: val_acc did not improve
acc: 0.832 , loss: 0.384 , val_acc: 0.792 , val_loss: 0.493 , val_auc: 0.886
Epoch 00033: val_acc did not improve
acc: 0.829 , loss: 0.380 , val_acc: 0.795 , val_loss: 0.474 , val_auc: 0.885
Epoch 00034: val_acc did not improve
acc: 0.835 , loss: 0.375 , val_acc: 0.808 , val_loss: 0.451 , val_auc: 0.884
Epoch 00035: val_acc improved from 0.80778 to 0.80812, saving model to CYP1A2_conv1.hdf5
acc: 0.822 , loss: 0.398 , val_acc: 0.801 , val_loss: 0.457 , val_auc: 0.884
Epoch 00036: val_acc did not improve
acc: 0.837 , loss: 0.373 , val_acc: 0.787 , val_loss: 0.496 , val_auc: 0.884
Epoch 00037: val_acc did not improve
acc: 0.833 , loss: 0.378 , val_acc: 0.804 , val_loss: 0.452 , val_auc: 0.886
Epoch 00038: val_acc did not improve
acc: 0.834 , loss: 0.371 , val_acc: 0.813 , val_loss: 0.457 , val_auc: 0.883
Epoch 00039: val_acc improved from 0.80812 to 0.81349, saving model to CYP1A2_conv1.hdf5
acc: 0.834 , loss: 0.368 , val_acc: 0.808 , val_loss: 0.451 , val_auc: 0.882
Epoch 00040: val_acc did not improve
acc: 0.836 , loss: 0.367 , val_acc: 0.804 , val_loss: 0.457 , val_auc: 0.883
Epoch 00041: val_acc did not improve
acc: 0.836 , loss: 0.373 , val_acc: 0.797 , val_loss: 0.467 , val_auc: 0.885
Epoch 00042: val_acc did not improve
acc: 0.835 , loss: 0.369 , val_acc: 0.804 , val_loss: 0.460 , val_auc: 0.885
Epoch 00043: val_acc did not improve
acc: 0.842 , loss: 0.365 , val_acc: 0.813 , val_loss: 0.463 , val_auc: 0.884
Epoch 00044: val_acc did not improve
acc: 0.840 , loss: 0.360 , val_acc: 0.804 , val_loss: 0.470 , val_auc: 0.884
Epoch 00045: val_acc did not improve
acc: 0.844 , loss: 0.356 , val_acc: 0.812 , val_loss: 0.460 , val_auc: 0.883
Epoch 00046: val_acc did not improve
acc: 0.837 , loss: 0.368 , val_acc: 0.813 , val_loss: 0.450 , val_auc: 0.882
Epoch 00047: val_acc did not improve
acc: 0.842 , loss: 0.363 , val_acc: 0.811 , val_loss: 0.452 , val_auc: 0.883
Epoch 00048: val_acc did not improve
acc: 0.839 , loss: 0.364 , val_acc: 0.794 , val_loss: 0.507 , val_auc: 0.884
Epoch 00049: val_acc did not improve
acc: 0.844 , loss: 0.357 , val_acc: 0.808 , val_loss: 0.461 , val_auc: 0.883
Epoch 00050: val_acc did not improve
acc: 0.843 , loss: 0.359 , val_acc: 0.808 , val_loss: 0.457 , val_auc: 0.884
Epoch 00051: val_acc did not improve
acc: 0.848 , loss: 0.350 , val_acc: 0.800 , val_loss: 0.497 , val_auc: 0.883
Epoch 00052: val_acc did not improve
acc: 0.845 , loss: 0.348 , val_acc: 0.807 , val_loss: 0.466 , val_auc: 0.883
Epoch 00053: val_acc did not improve
acc: 0.849 , loss: 0.342 , val_acc: 0.804 , val_loss: 0.497 , val_auc: 0.883
Epoch 00054: val_acc did not improve
acc: 0.846 , loss: 0.347 , val_acc: 0.806 , val_loss: 0.468 , val_auc: 0.883
Epoch 00055: val_acc did not improve
acc: 0.854 , loss: 0.340 , val_acc: 0.803 , val_loss: 0.483 , val_auc: 0.883
Epoch 00056: val_acc did not improve
acc: 0.849 , loss: 0.347 , val_acc: 0.806 , val_loss: 0.482 , val_auc: 0.885
Epoch 00057: val_acc did not improve
acc: 0.846 , loss: 0.346 , val_acc: 0.813 , val_loss: 0.463 , val_auc: 0.885
Epoch 00058: val_acc did not improve
acc: 0.846 , loss: 0.353 , val_acc: 0.804 , val_loss: 0.484 , val_auc: 0.885
Epoch 00059: val_acc did not improve
acc: 0.849 , loss: 0.350 , val_acc: 0.814 , val_loss: 0.460 , val_auc: 0.886
Epoch 00060: val_acc improved from 0.81349 to 0.81449, saving model to CYP1A2_conv1.hdf5
acc: 0.850 , loss: 0.341 , val_acc: 0.813 , val_loss: 0.463 , val_auc: 0.888
Epoch 00061: val_acc did not improve
acc: 0.852 , loss: 0.342 , val_acc: 0.807 , val_loss: 0.467 , val_auc: 0.887
Epoch 00062: val_acc did not improve
acc: 0.856 , loss: 0.337 , val_acc: 0.809 , val_loss: 0.483 , val_auc: 0.886
Epoch 00063: val_acc did not improve
acc: 0.851 , loss: 0.338 , val_acc: 0.801 , val_loss: 0.486 , val_auc: 0.884
Epoch 00064: val_acc did not improve
acc: 0.855 , loss: 0.338 , val_acc: 0.813 , val_loss: 0.460 , val_auc: 0.884
Epoch 00065: val_acc did not improve
acc: 0.849 , loss: 0.345 , val_acc: 0.777 , val_loss: 0.540 , val_auc: 0.885
Epoch 00066: val_acc did not improve
acc: 0.852 , loss: 0.339 , val_acc: 0.795 , val_loss: 0.474 , val_auc: 0.885
Epoch 00067: val_acc did not improve
acc: 0.848 , loss: 0.338 , val_acc: 0.817 , val_loss: 0.465 , val_auc: 0.886
Epoch 00068: val_acc improved from 0.81449 to 0.81718, saving model to CYP1A2_conv1.hdf5
acc: 0.852 , loss: 0.334 , val_acc: 0.800 , val_loss: 0.495 , val_auc: 0.887
Epoch 00069: val_acc did not improve
acc: 0.851 , loss: 0.337 , val_acc: 0.808 , val_loss: 0.467 , val_auc: 0.885
Epoch 00070: val_acc did not improve
acc: 0.855 , loss: 0.337 , val_acc: 0.801 , val_loss: 0.483 , val_auc: 0.886
Epoch 00071: val_acc did not improve
acc: 0.855 , loss: 0.334 , val_acc: 0.819 , val_loss: 0.464 , val_auc: 0.885
Epoch 00072: val_acc improved from 0.81718 to 0.81885, saving model to CYP1A2_conv1.hdf5
acc: 0.853 , loss: 0.334 , val_acc: 0.811 , val_loss: 0.472 , val_auc: 0.886
Epoch 00073: val_acc did not improve
acc: 0.858 , loss: 0.326 , val_acc: 0.810 , val_loss: 0.486 , val_auc: 0.885
Epoch 00074: val_acc did not improve
acc: 0.860 , loss: 0.325 , val_acc: 0.812 , val_loss: 0.479 , val_auc: 0.885
Epoch 00075: val_acc did not improve
acc: 0.852 , loss: 0.342 , val_acc: 0.798 , val_loss: 0.512 , val_auc: 0.885
Epoch 00076: val_acc did not improve
acc: 0.853 , loss: 0.329 , val_acc: 0.813 , val_loss: 0.478 , val_auc: 0.885
Epoch 00077: val_acc did not improve
acc: 0.859 , loss: 0.330 , val_acc: 0.809 , val_loss: 0.463 , val_auc: 0.887
Epoch 00078: val_acc did not improve
acc: 0.863 , loss: 0.321 , val_acc: 0.806 , val_loss: 0.495 , val_auc: 0.885
Epoch 00079: val_acc did not improve
acc: 0.851 , loss: 0.334 , val_acc: 0.813 , val_loss: 0.470 , val_auc: 0.888
Epoch 00080: val_acc did not improve
acc: 0.856 , loss: 0.332 , val_acc: 0.794 , val_loss: 0.521 , val_auc: 0.888
Epoch 00081: val_acc did not improve
acc: 0.859 , loss: 0.323 , val_acc: 0.814 , val_loss: 0.476 , val_auc: 0.888
Epoch 00082: val_acc did not improve
acc: 0.858 , loss: 0.324 , val_acc: 0.817 , val_loss: 0.471 , val_auc: 0.888
Epoch 00083: val_acc did not improve
acc: 0.858 , loss: 0.327 , val_acc: 0.781 , val_loss: 0.596 , val_auc: 0.889
Epoch 00084: val_acc did not improve
acc: 0.853 , loss: 0.337 , val_acc: 0.797 , val_loss: 0.495 , val_auc: 0.887
Epoch 00085: val_acc did not improve
acc: 0.862 , loss: 0.319 , val_acc: 0.808 , val_loss: 0.493 , val_auc: 0.885
Epoch 00086: val_acc did not improve
acc: 0.867 , loss: 0.310 , val_acc: 0.812 , val_loss: 0.485 , val_auc: 0.887
Epoch 00087: val_acc did not improve
acc: 0.867 , loss: 0.309 , val_acc: 0.812 , val_loss: 0.484 , val_auc: 0.886
Epoch 00088: val_acc did not improve
acc: 0.860 , loss: 0.316 , val_acc: 0.804 , val_loss: 0.508 , val_auc: 0.887
Epoch 00089: val_acc did not improve
acc: 0.863 , loss: 0.314 , val_acc: 0.812 , val_loss: 0.496 , val_auc: 0.888
Epoch 00090: val_acc did not improve
acc: 0.864 , loss: 0.315 , val_acc: 0.807 , val_loss: 0.483 , val_auc: 0.887
Epoch 00091: val_acc did not improve
acc: 0.867 , loss: 0.316 , val_acc: 0.793 , val_loss: 0.527 , val_auc: 0.888
Epoch 00092: val_acc did not improve
acc: 0.860 , loss: 0.326 , val_acc: 0.816 , val_loss: 0.463 , val_auc: 0.888
Epoch 00093: val_acc did not improve
acc: 0.861 , loss: 0.322 , val_acc: 0.815 , val_loss: 0.467 , val_auc: 0.887
Epoch 00094: val_acc did not improve
acc: 0.864 , loss: 0.316 , val_acc: 0.805 , val_loss: 0.505 , val_auc: 0.888
Epoch 00095: val_acc did not improve
acc: 0.867 , loss: 0.304 , val_acc: 0.817 , val_loss: 0.502 , val_auc: 0.888
Epoch 00096: val_acc did not improve
acc: 0.862 , loss: 0.314 , val_acc: 0.805 , val_loss: 0.494 , val_auc: 0.884
Epoch 00097: val_acc did not improve
acc: 0.863 , loss: 0.311 , val_acc: 0.805 , val_loss: 0.478 , val_auc: 0.886
Epoch 00098: val_acc did not improve
acc: 0.868 , loss: 0.309 , val_acc: 0.805 , val_loss: 0.507 , val_auc: 0.887
Epoch 00099: val_acc did not improve






    
Out[196]:





<keras.callbacks.History at 0x7f90d73ae4a8>

In [173]:

    

from keras import models
best_model=models.load_model("CYP1A2_conv1.hdf5")
#print(best_model.summary())


    

In [197]:

    

%matplotlib inline
import matplotlib.pyplot as plt

y_score=best_model.predict(X_test)
y_score_get_value=[]
for item in y_score:
    y_score_get_value.append(item[0])
    
fpr, tpr, _ = roc_curve(y_test, y_score_get_value)

plt.figure()
lw = 2
plt.plot(fpr, tpr, color='darkorange',
         lw=lw, label='ROC curve (area = %0.3f)' % auc(fpr,tpr))
plt.plot([0, 1], [0, 1], color='navy', lw=lw, linestyle='--')
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('Receiver operating characteristic ')
plt.legend(loc="lower right")
plt.savefig("roc.svg", format="svg")
plt.show()