tested steroids

DeepDrug.py

@@ -11,7 +11,7 @@
 import numpy as np
 import tensorflow as tf
 from sklearn.preprocessing import LabelEncoder
-from keras.models import Sequential
+from keras.models import Sequential, load_model
 from keras import optimizers, callbacks
 from keras.layers import Dense, Flatten, TimeDistributed, Dropout
 from keras import Input, Model
@@ -23,16 +23,6 @@ from keras.layers import Convolution3D, MaxPooling3D
 
 # ### used to store model prediction in order to plot roc curve
 
-# In[ ]:
-
-
-class prediction_history(callbacks.Callback):
-    def __init__(self):
-        self.predhis = []
-    def on_epoch_end(self, epoch, logs={}):
-        self.predhis.append(model.predict(predictor_train))
-
-
 # ### Creating input and ouputs
 
 # In[ ]:
@@ -130,7 +120,7 @@ def model_heavy(): # créer un objet modèle
 # In[8]:
 
 
-def model_new(): # créer un objet modèle
+def model_fast_k32(): # créer un objet modèle
     """
     Return a simple sequentiel model
     
@@ -138,10 +128,10 @@ def model_new(): # créer un objet modèle
         - model : keras.Model
     """
     inputs = Input(shape=(14,32,32,32))
-    conv_1 = Convolution3D(filters=64, kernel_size=5, padding="valid", data_format='channels_first')(inputs)
+    conv_1 = Convolution3D(filters=64, kernel_size=32, padding="valid", data_format='channels_first')(inputs)
     activation_1 = LeakyReLU(alpha = 0.1)(conv_1)
     drop_1 = Dropout(0.2)(activation_1)
-    conv_2 = Convolution3D(filters=64, kernel_size=3, padding="valid", data_format='channels_first')(drop_1)
+    conv_2 = Convolution3D(filters=128, kernel_size=32, padding="valid", data_format='channels_first')(drop_1)
     activation_2 = LeakyReLU(alpha = 0.1)(conv_2)
     maxpool = MaxPooling3D(pool_size=(2,2,2),
                             strides=None,
@@ -149,7 +139,37 @@ def model_new(): # créer un objet modèle
                             data_format='channels_first')(activation_2)
     drop_2 = Dropout(0.4)(maxpool)
     flatters = Flatten()(drop_2)
-    dense = Dense(128)(flatters)
+    dense = Dense(256)(flatters)
+    activation_3 = LeakyReLU(alpha = 0.1)(dense)
+    drop_3 = Dropout(0.4)(activation_3)
+    output = Dense(3, activation='softmax')(drop_3)
+    model = Model(inputs=inputs, outputs=output)
+    my_opt = optimizers.Adam(learning_rate=0.000001, beta_1=0.9, beta_2=0.999, amsgrad=False)
+    print(model.summary)
+    model.compile(optimizer=my_opt, loss="categorical_crossentropy",
+                  metrics=["accuracy"])
+    return model
+
+def model_fast_k16(): # créer un objet modèle
+    """
+    Return a simple sequentiel model
+    
+    Returns :
+        - model : keras.Model
+    """
+    inputs = Input(shape=(14,32,32,32))
+    conv_1 = Convolution3D(filters=64, kernel_size=16, padding="valid", data_format='channels_first')(inputs)
+    activation_1 = LeakyReLU(alpha = 0.1)(conv_1)
+    drop_1 = Dropout(0.2)(activation_1)
+    conv_2 = Convolution3D(filters=128, kernel_size=16, padding="valid", data_format='channels_first')(drop_1)
+    activation_2 = LeakyReLU(alpha = 0.1)(conv_2)
+    maxpool = MaxPooling3D(pool_size=(2,2,2),
+                            strides=None,
+                            padding='valid',
+                            data_format='channels_first')(activation_2)
+    drop_2 = Dropout(0.4)(maxpool)
+    flatters = Flatten()(drop_2)
+    dense = Dense(256)(flatters)
     activation_3 = LeakyReLU(alpha = 0.1)(dense)
     drop_3 = Dropout(0.4)(activation_3)
     output = Dense(3, activation='softmax')(drop_3)
@@ -200,51 +220,72 @@ def model_light(): # créer un objet modèle
 
 # In[ ]:
 
-
+sample = 1000
-data = in_out_lists(1400)
+data = in_out_lists(sample)
 pockets = np.cumsum(data[1], axis=0)[-1]
 
 
 # In[ ]:
 
 
-print("with random seed=9001 and a 1400 pockets dataset the rates are:\n      {} heme, {} nucleotide, {} control\n      Total avaible dataset are composed of the following proportions:\n      {} heme, {} nucleotide, {} control".format(pockets[0]/1400, pockets[1]/1400,pockets[2]/1400,
+print("with random seed=9001 and a {} pockets dataset the rates are:\n      {} heme, {} nucleotide, {} control\n      Total avaible dataset are composed of the following proportions:\n      {} heme, {} nucleotide, {} control".format(sample, pockets[0]/sample,
+                                                                                       pockets[1]/sample,pockets[2]/sample,
                                                                                        0.145, 0.380, 0.475))
 
 
 # In[ ]:
 
+train = int(sample*0.6)
 
 data_onehot = data[0]
 output = data[1]
-X_train = data_onehot[0:1000,]
+
-Y_train = output[0:1000,]
+X_train = data_onehot[0:train,]
-X_test = data_onehot[1000:,]
+Y_train = output[0:train,]
-Y_test = output[1000:,]
+X_test = data_onehot[train:,]
+Y_test = output[train:,]
 
 
 # In[ ]:
 
 
-my_model = model_new()
+my_model = model_fast_k16()
 
 
 # In[ ]:
 
 
 tf.test.is_gpu_available()
-#my_model.fit(X_train, Y_train, epochs=50, batch_size=30)
 
 
 # In[ ]:
 
 
-history_mild_2mp = my_model.fit(X_train, Y_train, validation_data=(X_test, Y_test), epochs=30, batch_size=32)
+my_model.fit(X_train, Y_train, validation_data=(X_test, Y_test), epochs=50, batch_size=32)
-my_model.save('new_model_e30_b32_t1000.h5')
+#my_model.save('new_model_e50_b32_t1600.h5')
+#my_model = load_model('new_model_e50_b32_t1600.h5')
 
+# ## Testing steroids
 
-# In[ ]:
+with open("steroid.list", "r") as filin: 
+    steroid = filin.read() 
+    steroid = steroid.split("\n") 
+    steroid.pop()
 
+X_steroid = np.ndarray(shape=(69, 14, 32, 32, 32))
 
-#predictions=prediction_history()
+i = -1
+for pocket in steroid:
+    i += 1
+    X_steroid[i,] = np.load("deepdrug3d_voxel_data/"+pocket+".npy")
+
+Y_pred_steroid = my_model.predict(X_steroid)
+Y_pred_steroid = Y_pred_steroid.round()
+
+steroid_predict = Y_pred_steroid.cumsum(axis=0)
+print("On 69 steroid-binded pockets, prediction are the following:\n\
+      predicted as heme:\t{}\npredicted as nucleotide:\t{}\n\
+      predicted as control:\t{}\n".format(steroid_predict[0],
+                                          steroid_predict[1],
+                                          steroid_predict[2]))