attention_encoding.py

import numpy as np
import pandas as pd
from keras.layers import Dense, Input, Dropout, Conv1D, Reshape, MaxPooling1D, ZeroPadding1D, AveragePooling1D
from keras.layers import  Add, Activation, ZeroPadding2D, BatchNormalization, Flatten, Conv2D, AveragePooling2D, MaxPooling2D, GlobalMaxPooling2D
# from keras.layers import Conv1D
from keras.layers.merge import concatenate
from keras.optimizers import SGD
from keras.models import Model
from keras.regularizers import l2
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import RobustScaler
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import roc_curve
from keras.utils import np_utils
import matplotlib.pyplot as plt
from sklearn.metrics import roc_auc_score
import sklearn.metrics as metrics
from sklearn.metrics import confusion_matrix
from sklearn.metrics import matthews_corrcoef,accuracy_score, precision_score,recall_score
from sklearn.manifold import TSNE
from attention import *

from keras.initializers import glorot_uniform

from xgboost import XGBClassifier
import time

start = time.time()

def define_model():
    
    ########################################################"Channel-1" ########################################################
    
    input_1 = Input(shape=(573, ), name='Protein_a')

    p1 = Reshape((3, 191))(input_1)


    ##attention
    d1 = p1.get_shape().as_list()
    d1 = d1[2]
    X = Self_Attention(d1)(p1)

    # 均值池化
    X = AveragePooling1D(pool_size=2, padding='same')(X)

    ########################################################"Channel-2" ########################################################

    input_2 = Input(shape=(573, ), name='Protein_b')

    p2 = Reshape((3, 191))(input_2)

    ##attention
    d10 = p2.get_shape().as_list()
    d10 = d10[2]
    p2 = Self_Attention(d10)(p2)

    # 均值池化
    p2 = AveragePooling1D(pool_size=2, padding='same')(p2)

    ##################################### Merge Abstraction features ##################################################
    
    merged = concatenate([X,p2], name='merged_protein1_2')
    
    ##################################### Prediction Module ##########################################################
    
   
    merged = Flatten()(merged)

    pre_output = Dense(64, activation='relu', kernel_initializer='glorot_normal', name='Merged_feature_1')(merged)
    pre_output = Dense(32, activation='relu', kernel_initializer='glorot_normal', name='Merged_feature_2')(pre_output)
    pre_output = Dense(16, activation='relu', kernel_initializer='he_uniform', name='Merged_feature_3')(pre_output)


    
    pre_output=Dropout(0.2)(pre_output)

    output = Dense(1, activation='sigmoid', name='output')(pre_output)

    model = Model(input=[input_1, input_2], output=output)
   
    sgd = SGD(lr=0.01, momentum=0.9, decay=0.001)

    model.compile(loss='binary_crossentropy', optimizer=sgd, metrics=['accuracy'])

    return model


##################################### Load Positive and Negative Dataset ##########################################################
    
df_pos = pd.read_csv('/P.csv', header=None)
df_neg = pd.read_csv('/N.csv', header=None)

df_neg['Status'] = 0
df_pos['Status'] = 1
df_neg=df_neg.sample(n=len(df_pos))

df = pd.concat([df_pos,df_neg])
df = df.reset_index()
df=df.sample(frac=1)
df = df.iloc[:,1:]

X = df.iloc[:,0:1146].values
y = df.iloc[:,1146:].values
Trainlabels=y
scaler = StandardScaler().fit(X)
#scaler = RobustScaler().fit(X)
X = scaler.transform(X)

X1_train = X[:, :573]
X2_train = X[:, 573:]

##################################### Five-fold Cross-Validation ##########################################################
    
kf=StratifiedKFold(n_splits=5)

accuracy1 = []
specificity1 = []
sensitivity1 = []
precision1=[]
recall1=[]

m_coef=[]
dnn_fpr_list=[]
dnn_tpr_list=[]
dnn_auc_list = []
o=0
max_accuracy=float("-inf")
dnn_fpr=None
dnn_tpr=None

for train, test in kf.split(X,y):
    global model
    model=define_model()
    o=o+1

    model.fit([X1_train[train],X2_train[train]],y[train],epochs=50,batch_size=64,verbose=1)

    y_test=y[test]

    y_score = model.predict([X1_train[test],X2_train[test]])

    fpr, tpr, _= roc_curve(y_test,  y_score)
    auc = metrics.roc_auc_score(y_test, y_score)
    
    dnn_auc_list.append(auc)
    
    y_score=y_score[:,0]
    
    for i in range(0,len(y_score)):
        if(y_score[i]>0.5):
            y_score[i]=1
        else:
            y_score[i]=0
            
    cm1=confusion_matrix(y[test][:,0],y_score)
    acc1 = accuracy_score(y[test][:,0], y_score, sample_weight=None)
    spec1= (cm1[0,0])/(cm1[0,0]+cm1[0,1])
    sens1 = recall_score(y[test][:,0], y_score, sample_weight=None)
    prec1=precision_score(y[test][:,0], y_score, sample_weight=None)
    

    sensitivity1.append(sens1)
    specificity1.append(spec1)
    accuracy1.append(acc1)
    precision1.append(prec1)
    
    coef=matthews_corrcoef(y[test], y_score, sample_weight=None)
    m_coef.append(coef)
    # dnn_fpr_list.append(fpr)
    # dnn_tpr_list.append(tpr)

    if acc1>max_accuracy:
        max_accuracy=acc1
        dnn_fpr=fpr[:]
        dnn_tpr=tpr[:]

dnn_fpr=pd.DataFrame(dnn_fpr)
dnn_tpr=pd.DataFrame(dnn_tpr)
dnn_fpr.to_csv('fprDNN.csv',header=False, index=False)
dnn_tpr.to_csv('tprDNN.csv',header=False, index=False)

mean_acc1=np.mean(accuracy1)
std_acc1=np.std(accuracy1)
var_acc1=np.var(accuracy1)
print("Accuracy1:"+str(mean_acc1)+" ± "+str(std_acc1))
print("Accuracy_Var:"+str(mean_acc1)+" ± "+str(var_acc1))
mean_spec1=np.mean(specificity1)
std_spec1=np.std(specificity1)
print("Specificity1:"+str(mean_spec1)+" ± "+str(std_spec1))
mean_sens1=np.mean(sensitivity1)
std_sens1=np.std(sensitivity1)
print("Sensitivity1:"+str(mean_sens1)+" ± "+str(std_sens1))
mean_prec1=np.mean(precision1)
std_prec1=np.std(precision1)
print("Precison1:"+str(mean_prec1)+" ± "+str(std_prec1))

mean_coef=np.mean(m_coef)
std_coef=np.std(m_coef)
print("MCC1:"+str(mean_coef)+" ± "+str(std_coef))

print("AUC1:"+str(np.mean(dnn_auc_list)))


end1 = time.time()
end11=end1 - start
print(f"Runtime of the program is {end1 - start}")