first commit

Feature construction/Feature_code.m

@@ -0,0 +1,11 @@
+function code=Feature_code(SR,SD,seq)
+[m,n]=size(seq);
+code=[];
+for i=1:m
+    a=all_negative(i,1);
+    b=all_negative(i,2);
+    code_miRNA_disease=[SR(a,:),SD(b,:)];
+   code=[code;code_miRNA_disease];
+
+end
+end

Feature construction/Feature_extraction.m

@@ -0,0 +1,24 @@
+function feature=Feature_extraction(Positive,Negative)%samples with labels
+[m1,n1]=size(Positive);
+[m2,n2]=size(Negative);
+csp=zeros(n1,1);
+csn=zeros(n2,1);
+sps=zeros(n1,1);
+for i=1:n1
+    csp(i)=sum(Positive(:,i));
+    csn(i)=sum(Negative(:,i));
+    sps(i)=(csp(i)+csn(i))*log(m1/csp(i)+m2/csn(i));
+end
+[spssort, index] = sort(sps,'descend');
+[m,n]=size(index);
+micfeature=[];
+disfeature=[];
+for j=1:m
+    if  1<=index(j)&&index(j)<=2204
+        micfeature=[micfeature;index(j)];
+    else
+        disfeature=[disfeature;index(j)];
+    end
+end
+feature=[micfeature(1:100);disfeature(1:100)];
+end

Predictors/IMPMD/linear_regression.m

@@ -0,0 +1,43 @@
+[m1,n1]=size(Positive_200);
+[m2,n2]=size(Negative_200);
+Y=[ones(m1,1);zeros(m2,1)]; Y=Y(rank);
+feature=[prdY_svm,prdY_knn,prdY_reg];
+[m3,n3]=size(feature);
+b=regress(Y,feature);
+prdY=zeros(m3,1);
+for i=1:size(feature,1)
+    prdY(i)=b'*feature(i,:)';
+end
+for i=1:length(prdY)
+    if prdY(i)>=0.5    %¿Éµ÷
+       predictedY(i,:)=1;
+    else
+       predictedY(i,:)=-1;
+    end
+end
+%%%%%%%%%%%%%%%%%%roc curve
+targets=Y;
+targets1=[];
+targets1=[targets1;Y];
+zz=find(targets1==0);
+for i=1:length(zz)
+    targets1(zz(i))=-1;
+end
+[tpr,fpr,thresholds] = roc(targets',prdY');
+AUCbb=AUCwang(tpr,fpr);
+hold on, plot(fpr,tpr)
+
+
+P=size(find(targets1==1),1);
+TP=size(find(predictedY==targets1 & predictedY==1),1);
+FN=size(find(predictedY~=targets1&targets1==1),1);
+N=size(find(targets1==-1),1);
+TN=size(find(predictedY==targets1&predictedY==-1),1);
+FP=size(find(predictedY~=targets1&targets1==-1),1);
+
+acc=length(find(predictedY==targets1))/length(targets1);
+sen=TP/(P);
+spe=TN/(N);
+mcc=(TP*TN-FP*FN)/sqrt((TP+FN)*(TP+FP)*(TN+FP)*(TN+FN));
+auc=AUCbb;
+pre=TP/(TP+FP);

Predictors/KNN/KNN.m

@@ -0,0 +1,27 @@
+function PredictY=KNN(K,trainset,testset)
+[trainm,trainn]=size(trainset);
+[testm,~]=size(testset);
+PredictY=zeros(testm,1);
+distancev=zeros(trainm,1);
+for i=1:testm
+    for j=1:trainm
+        distancev(j)=0;
+        for k=1:trainn-1
+            distancev(j)=distancev(j)+(testset(i,k)-trainset(j,k))^2;
+        end
+        distancev(j)=sqrt(distancev(j));
+    end 
+    [~,val]=sort(distancev);
+    val=val(1:K);
+    class1=0;
+    class2=0;
+    for k=1:size(val,1)   
+        if trainset(val(k),end)==1
+            class1=class1+1;
+        else
+            class2=class2+1;
+        end
+    end
+    PredictY(i)=class1/K;
+end
+end

Predictors/KNN/Kfold_KNN.m

@@ -0,0 +1,70 @@
+function [acc,mcc,sen,spe,auc,pre,prdY,fpr,tpr,predictedY]=Kfold_KNN(rank,Xpos,Xneg,pth,cv,K)
+%  [acc,mcc,sen,spe,auc,pre,prdY,fpr,tpr,predictedY]=crossvalidation(randperm(60)',Positive_200(1:30,:),Negative_200(1:30,:),0.5,10,3);
+[m1,n1]=size(Xpos);
+ [m2,n2]=size(Xneg);
+ X=[Xpos;Xneg];    
+ Y=[ones(m1,1);-ones(m2,1)]; 
+ [m3,n3]=size(X);  
+%  p=randperm(m3)';
+  X=X(rank,:);
+  Y=Y(rank);
+ K_neighbor=K; 
+ prdY=[]; predictedY=[]; 
+ targets=[];
+ predictedY=[];
+ Ytest=[];
+ kfold=cv;
+t=fix(m3/kfold);
+ for k=1:kfold
+   if k==kfold
+        train_data=X(1:t*(k-1),:);
+        test_data=X(t*(k-1)+1:m3,:);
+        train_target=Y(1:t*(k-1),:);
+        test_target=Y(t*(k-1)+1:m3,:);
+
+   else  
+       train_data=[X(1:t*(k-1),:);X(t*k+1:m3,:)];
+       test_data=X(t*(k-1)+1:t*k,:);
+       train_target=[Y(1:t*(k-1));Y(t*k+1:m3)];
+       test_target=Y(t*(k-1)+1:t*k,:);  
+   end       
+     prdY_k=KNN( K_neighbor,[ train_data,train_target],[test_data,test_target]);
+     prdY=[prdY;prdY_k];
+%     test_target=[test_target;test_target];
+ end
+
+ for i=1:length(prdY)
+    if prdY(i)>=pth    
+       predictedY(i,:)=1;
+    else
+       predictedY(i,:)=-1;
+    end
+end
+%%%%%%%%%%%%%%%%%%roc curve
+targets=Y;
+targets1=[];
+targets1=[targets1;Y];
+zz=find(targets==-1);
+for i=1:length(zz)
+    targets(zz(i))=0;
+end
+[tpr,fpr,thresholds] = roc(targets',prdY');
+AUCbb=AUCwang(tpr,fpr);
+hold on, plot(fpr,tpr)
+
+
+P=size(find(targets1==1),1);
+TP=size(find(predictedY==targets1 & predictedY==1),1);
+FN=size(find(predictedY~=targets1&targets1==1),1);
+N=size(find(targets1==-1),1);
+TN=size(find(predictedY==targets1&predictedY==-1),1);
+FP=size(find(predictedY~=targets1&targets1==-1),1);
+
+acc=length(find(predictedY==targets1))/length(targets1);
+sen=TP/(P);
+spe=TN/(N);
+mcc=(TP*TN-FP*FN)/sqrt((TP+FN)*(TP+FP)*(TN+FP)*(TN+FN));
+auc=AUCbb;
+pre=TP/(TP+FP);
+end
+

Predictors/LR/Kfold_regression.m

@@ -0,0 +1,90 @@
+function [acc,mcc,sen,spe,auc,pre,prdY,fpr,tpr,predictedY]=Kfold_regression(Xpos,Xneg,pth,cv, alpha, threshold, maxTimes)
+
+[m1,n1]=size(Xpos);
+[m2,n2]=size(Xneg);
+X=[Xpos;Xneg];
+Y=[ones(m1,1);-ones(m2,1)];
+[m3,n3]=size(X);  
+%  p=randperm(m3)';
+%  X=X(p,:);
+%  Y=Y(p);
+prdY=[];
+targets=[];
+predictedY=[];
+indices=crossvalind('Kfold',X(1:m3,n3),cv);
+for k=1:cv 
+    test = (indices == k);
+    train = ~test;
+    train_data=X(train,:);
+    train_target=Y(train,:);
+    test_data=X(test,:);
+    test_target=Y(test,:);
+    %     maxIndex=n1;
+    %     [theta,prdY_k]=logisticRegression([train_data,train_target], maxIndex, alpha, threshold, maxTimes);
+    %     prdY=[prdY;prdY_k];
+    trainset=[train_data,train_target];
+    testset=[test_data,test_target];
+    dataSize = size(trainset);
+    dataLen = dataSize(1);
+    paramLen = dataSize(2);
+    maxIndex=paramLen-1;
+    theta = zeros(paramLen, 1);
+    times = 0;
+    cost0 = 0;
+    cost1 = 1;
+    while abs(cost1-cost0) > threshold && times < maxTimes
+        times = times + 1;
+        theta0 = theta;
+        cost0 = LogisticRegressionCostFun(theta, trainset);
+        for i = 1 : dataLen
+            tmp = ((1 / (1 + exp(-theta0' * [1, trainset(i,1:maxIndex)]'))) - trainset(i, paramLen)) / dataLen;
+            theta(1) = theta(1) - alpha * tmp;
+            for j = 2 : paramLen
+                theta(j) = theta(j) - alpha * tmp * trainset(i, j - 1);
+            end
+        end
+        cost1 = LogisticRegressionCostFun(theta, trainset);
+    end
+    [m,n]=size(testset);
+    predY_k=zeros(m,1);
+    for kk=1:m
+        x=testset(kk,1:maxIndex);
+        predY_k(kk)=1/(1+exp(-theta'*[1,x]'));
+    end
+    prdY=[prdY;prdY_k];
+end
+for i=1:length(prdY)
+    if prdY(i)>=pth    
+        predictedY(i,:)=1;
+    else
+        predictedY(i,:)=-1;
+    end
+end
+%%%%%%%%%%%%%%%%%%roc curve
+targets=[targets;Ytest];
+targets1=[];
+targets1=[targets1;Ytest];
+zz=find(targets==-1);
+for i=1:length(zz)
+    targets(zz(i))=0;
+end
+[tpr,fpr,thresholds] = roc(targets',prdY');
+AUCbb=AUCwang(tpr,fpr);
+hold on, plot(fpr,tpr)
+
+
+P=size(find(targets1==1),1);
+TP=size(find(predictedY==targets1 & predictedY==1),1);
+FN=size(find(predictedY~=targets1&targets1==1),1);
+N=size(find(targets1==-1),1);
+TN=size(find(predictedY==targets1&predictedY==-1),1);
+FP=size(find(predictedY~=targets1&targets1==-1),1);
+
+acc=length(find(predictedY==targets1))/length(targets1);
+sen=TP/(P);
+spe=TN/(N);
+mcc=(TP*TN-FP*FN)/sqrt((TP+FN)*(TP+FP)*(TN+FP)*(TN+FN));
+auc=AUCbb;
+pre=TP/(TP+FP);
+end
+

Predictors/LR/LogisticRegressionCostFun.m

@@ -0,0 +1,18 @@
+function cost=LogisticRegressionCostFun(theta, data)
+
+
+
+paramLen = length(theta);
+X = zeros(paramLen, 1);
+dataSize = size(data);
+dataLen = dataSize(1);
+cost = 0;
+for i = 1 : dataLen 
+    X(1) = 1; 
+    for k = 1 : paramLen - 1   
+        X(k + 1) = data(i, k);   
+    end
+    cost = cost +(-data(i, dataSize(2)) * log(1/(1 + exp(-(theta' * X))))  - (1 - data(i, dataSize(2))) * log(1 - 1/(1 + exp(-(theta' * X)))));
+end
+cost = cost / dataLen;
+end

Predictors/LR/logisticRegression.m

@@ -0,0 +1,48 @@
+% function theta=logisticRegression(x,y)
+
+% % x = [1,2,3;2,3,4;9,8,1;4,5,6;5,6,7;2,6,10;1,5,10];
+% % y = [0, 0, 1, 0 ,0 , 1, 1]'; 
+% [m,n]=size(x);
+% theta = glmfit(x, [y ones(m,1)], 'binomial', 'link', 'logit')
+% end
+function [theta,predictY]=logisticRegression(trainset, testset,maxIndex, alpha, threshold, maxTimes)
+
+
+%     data = [0 0 0; 0 1 0; 0 1.5 0; 0.5 0.5 0; 0.5 1 0; 1 0.95 0; 0.5 1.4 0; 1.5 0.51 0; 2 0 0;
+% 
+%         1.9 0 0; 0 3 1; 0 2.1 1; 0.5 1.8 1; 0.8 1.5 1; 1 1.2 1; 1.5 2 1; 3 0 1; 3 1 1; 2 2 1;
+% 
+%         3 4 1; 1.8 0.5 1];
+% 
+%     maxIndex = 2;
+%     alpha = 0.1;
+%     threshold = 0.00001;
+%     maxTimes = 1000;
+
+dataSize = size(trainset);
+dataLen = dataSize(1);
+param = maxIndex + 1;
+theta = zeros(param, 1);
+times = 0;
+cost0 = 0;
+cost1 = 1;
+while abs(cost1-cost0) > threshold && times < maxTimes
+    times = times + 1;  
+    theta0 = theta;
+    cost0 = LogisticRegressionCostFun(theta, trainset);
+    for i = 1 : dataLen
+        tmp = ((1 / (1 + exp(-theta0' * [1, trainset(i,1:maxIndex)]'))) - trainset(i, param)) / dataLen;
+        theta(1) = theta(1) - alpha * tmp;
+        for j = 2 : param
+            theta(j) = theta(j) - alpha * tmp * trainset(i, j - 1); 
+        end  
+    end
+    cost1 = LogisticRegressionCostFun(theta, trainset);
+end
+[m,n]=size(testset);
+predictY=zeros(m,1);
+for k=1:m
+    x=testset(k,1:maxIndex);
+    predictY(k)=1/(1+exp(-theta'*[1,x]'));
+end
+end

Predictors/SVM/AUCwang.m

@@ -0,0 +1,7 @@
+function  AUCbb=AUCwang(tpr,fpr)
+
+for i=1:length(tpr)-1
+r(i)=(fpr(i+1)-fpr(i))*tpr(i);
+s(i)=0.5*(fpr(i+1)-fpr(i))*(tpr(i+1)-tpr(i));
+end
+AUCbb=sum(r)+sum(s);

Predictors/SVM/Change_format.m

@@ -0,0 +1,18 @@
+function Change_format(X,file)
+%X is the priminal data; file is the output filename(Note it must be a char!)  
+%The first colum is the label of the samples, so the dim of the sample is
+%n-1 dimensions
+
+[m,n]=size(X);
+filename=file;
+fid=fopen(filename,'w');
+for i=1:m
+    fprintf(fid,'%5d ',X(i,1));
+    for j=2:n
+        fprintf(fid,'%5d: %12.4f ',j-1,X(i,j));
+    end
+    fprintf(fid,'\n');    
+end
+
+fclose(fid);
+clear fid i j;